xref: /linux/fs/btrfs/ioctl.c (revision 4246b92cf9fb32da8d8b060c92d8302797c6fbea)
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/compat.h>
37 #include <linux/bit_spinlock.h>
38 #include <linux/security.h>
39 #include <linux/xattr.h>
40 #include <linux/mm.h>
41 #include <linux/slab.h>
42 #include <linux/blkdev.h>
43 #include <linux/uuid.h>
44 #include <linux/btrfs.h>
45 #include <linux/uaccess.h>
46 #include "ctree.h"
47 #include "disk-io.h"
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "volumes.h"
52 #include "locking.h"
53 #include "inode-map.h"
54 #include "backref.h"
55 #include "rcu-string.h"
56 #include "send.h"
57 #include "dev-replace.h"
58 #include "props.h"
59 #include "sysfs.h"
60 #include "qgroup.h"
61 #include "tree-log.h"
62 #include "compression.h"
63 
64 #ifdef CONFIG_64BIT
65 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
66  * structures are incorrect, as the timespec structure from userspace
67  * is 4 bytes too small. We define these alternatives here to teach
68  * the kernel about the 32-bit struct packing.
69  */
70 struct btrfs_ioctl_timespec_32 {
71 	__u64 sec;
72 	__u32 nsec;
73 } __attribute__ ((__packed__));
74 
75 struct btrfs_ioctl_received_subvol_args_32 {
76 	char	uuid[BTRFS_UUID_SIZE];	/* in */
77 	__u64	stransid;		/* in */
78 	__u64	rtransid;		/* out */
79 	struct btrfs_ioctl_timespec_32 stime; /* in */
80 	struct btrfs_ioctl_timespec_32 rtime; /* out */
81 	__u64	flags;			/* in */
82 	__u64	reserved[16];		/* in */
83 } __attribute__ ((__packed__));
84 
85 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
86 				struct btrfs_ioctl_received_subvol_args_32)
87 #endif
88 
89 
90 static int btrfs_clone(struct inode *src, struct inode *inode,
91 		       u64 off, u64 olen, u64 olen_aligned, u64 destoff,
92 		       int no_time_update);
93 
94 /* Mask out flags that are inappropriate for the given type of inode. */
95 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
96 {
97 	if (S_ISDIR(mode))
98 		return flags;
99 	else if (S_ISREG(mode))
100 		return flags & ~FS_DIRSYNC_FL;
101 	else
102 		return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
103 }
104 
105 /*
106  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
107  */
108 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
109 {
110 	unsigned int iflags = 0;
111 
112 	if (flags & BTRFS_INODE_SYNC)
113 		iflags |= FS_SYNC_FL;
114 	if (flags & BTRFS_INODE_IMMUTABLE)
115 		iflags |= FS_IMMUTABLE_FL;
116 	if (flags & BTRFS_INODE_APPEND)
117 		iflags |= FS_APPEND_FL;
118 	if (flags & BTRFS_INODE_NODUMP)
119 		iflags |= FS_NODUMP_FL;
120 	if (flags & BTRFS_INODE_NOATIME)
121 		iflags |= FS_NOATIME_FL;
122 	if (flags & BTRFS_INODE_DIRSYNC)
123 		iflags |= FS_DIRSYNC_FL;
124 	if (flags & BTRFS_INODE_NODATACOW)
125 		iflags |= FS_NOCOW_FL;
126 
127 	if (flags & BTRFS_INODE_NOCOMPRESS)
128 		iflags |= FS_NOCOMP_FL;
129 	else if (flags & BTRFS_INODE_COMPRESS)
130 		iflags |= FS_COMPR_FL;
131 
132 	return iflags;
133 }
134 
135 /*
136  * Update inode->i_flags based on the btrfs internal flags.
137  */
138 void btrfs_update_iflags(struct inode *inode)
139 {
140 	struct btrfs_inode *ip = BTRFS_I(inode);
141 	unsigned int new_fl = 0;
142 
143 	if (ip->flags & BTRFS_INODE_SYNC)
144 		new_fl |= S_SYNC;
145 	if (ip->flags & BTRFS_INODE_IMMUTABLE)
146 		new_fl |= S_IMMUTABLE;
147 	if (ip->flags & BTRFS_INODE_APPEND)
148 		new_fl |= S_APPEND;
149 	if (ip->flags & BTRFS_INODE_NOATIME)
150 		new_fl |= S_NOATIME;
151 	if (ip->flags & BTRFS_INODE_DIRSYNC)
152 		new_fl |= S_DIRSYNC;
153 
154 	set_mask_bits(&inode->i_flags,
155 		      S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
156 		      new_fl);
157 }
158 
159 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
160 {
161 	struct btrfs_inode *ip = BTRFS_I(file_inode(file));
162 	unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
163 
164 	if (copy_to_user(arg, &flags, sizeof(flags)))
165 		return -EFAULT;
166 	return 0;
167 }
168 
169 static int check_flags(unsigned int flags)
170 {
171 	if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
172 		      FS_NOATIME_FL | FS_NODUMP_FL | \
173 		      FS_SYNC_FL | FS_DIRSYNC_FL | \
174 		      FS_NOCOMP_FL | FS_COMPR_FL |
175 		      FS_NOCOW_FL))
176 		return -EOPNOTSUPP;
177 
178 	if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
179 		return -EINVAL;
180 
181 	return 0;
182 }
183 
184 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
185 {
186 	struct inode *inode = file_inode(file);
187 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
188 	struct btrfs_inode *ip = BTRFS_I(inode);
189 	struct btrfs_root *root = ip->root;
190 	struct btrfs_trans_handle *trans;
191 	unsigned int flags, oldflags;
192 	int ret;
193 	u64 ip_oldflags;
194 	unsigned int i_oldflags;
195 	umode_t mode;
196 
197 	if (!inode_owner_or_capable(inode))
198 		return -EPERM;
199 
200 	if (btrfs_root_readonly(root))
201 		return -EROFS;
202 
203 	if (copy_from_user(&flags, arg, sizeof(flags)))
204 		return -EFAULT;
205 
206 	ret = check_flags(flags);
207 	if (ret)
208 		return ret;
209 
210 	ret = mnt_want_write_file(file);
211 	if (ret)
212 		return ret;
213 
214 	inode_lock(inode);
215 
216 	ip_oldflags = ip->flags;
217 	i_oldflags = inode->i_flags;
218 	mode = inode->i_mode;
219 
220 	flags = btrfs_mask_flags(inode->i_mode, flags);
221 	oldflags = btrfs_flags_to_ioctl(ip->flags);
222 	if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
223 		if (!capable(CAP_LINUX_IMMUTABLE)) {
224 			ret = -EPERM;
225 			goto out_unlock;
226 		}
227 	}
228 
229 	if (flags & FS_SYNC_FL)
230 		ip->flags |= BTRFS_INODE_SYNC;
231 	else
232 		ip->flags &= ~BTRFS_INODE_SYNC;
233 	if (flags & FS_IMMUTABLE_FL)
234 		ip->flags |= BTRFS_INODE_IMMUTABLE;
235 	else
236 		ip->flags &= ~BTRFS_INODE_IMMUTABLE;
237 	if (flags & FS_APPEND_FL)
238 		ip->flags |= BTRFS_INODE_APPEND;
239 	else
240 		ip->flags &= ~BTRFS_INODE_APPEND;
241 	if (flags & FS_NODUMP_FL)
242 		ip->flags |= BTRFS_INODE_NODUMP;
243 	else
244 		ip->flags &= ~BTRFS_INODE_NODUMP;
245 	if (flags & FS_NOATIME_FL)
246 		ip->flags |= BTRFS_INODE_NOATIME;
247 	else
248 		ip->flags &= ~BTRFS_INODE_NOATIME;
249 	if (flags & FS_DIRSYNC_FL)
250 		ip->flags |= BTRFS_INODE_DIRSYNC;
251 	else
252 		ip->flags &= ~BTRFS_INODE_DIRSYNC;
253 	if (flags & FS_NOCOW_FL) {
254 		if (S_ISREG(mode)) {
255 			/*
256 			 * It's safe to turn csums off here, no extents exist.
257 			 * Otherwise we want the flag to reflect the real COW
258 			 * status of the file and will not set it.
259 			 */
260 			if (inode->i_size == 0)
261 				ip->flags |= BTRFS_INODE_NODATACOW
262 					   | BTRFS_INODE_NODATASUM;
263 		} else {
264 			ip->flags |= BTRFS_INODE_NODATACOW;
265 		}
266 	} else {
267 		/*
268 		 * Revert back under same assumptions as above
269 		 */
270 		if (S_ISREG(mode)) {
271 			if (inode->i_size == 0)
272 				ip->flags &= ~(BTRFS_INODE_NODATACOW
273 				             | BTRFS_INODE_NODATASUM);
274 		} else {
275 			ip->flags &= ~BTRFS_INODE_NODATACOW;
276 		}
277 	}
278 
279 	/*
280 	 * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
281 	 * flag may be changed automatically if compression code won't make
282 	 * things smaller.
283 	 */
284 	if (flags & FS_NOCOMP_FL) {
285 		ip->flags &= ~BTRFS_INODE_COMPRESS;
286 		ip->flags |= BTRFS_INODE_NOCOMPRESS;
287 
288 		ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
289 		if (ret && ret != -ENODATA)
290 			goto out_drop;
291 	} else if (flags & FS_COMPR_FL) {
292 		const char *comp;
293 
294 		ip->flags |= BTRFS_INODE_COMPRESS;
295 		ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
296 
297 		if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
298 			comp = "lzo";
299 		else if (fs_info->compress_type == BTRFS_COMPRESS_ZLIB)
300 			comp = "zlib";
301 		else
302 			comp = "zstd";
303 		ret = btrfs_set_prop(inode, "btrfs.compression",
304 				     comp, strlen(comp), 0);
305 		if (ret)
306 			goto out_drop;
307 
308 	} else {
309 		ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
310 		if (ret && ret != -ENODATA)
311 			goto out_drop;
312 		ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
313 	}
314 
315 	trans = btrfs_start_transaction(root, 1);
316 	if (IS_ERR(trans)) {
317 		ret = PTR_ERR(trans);
318 		goto out_drop;
319 	}
320 
321 	btrfs_update_iflags(inode);
322 	inode_inc_iversion(inode);
323 	inode->i_ctime = current_time(inode);
324 	ret = btrfs_update_inode(trans, root, inode);
325 
326 	btrfs_end_transaction(trans);
327  out_drop:
328 	if (ret) {
329 		ip->flags = ip_oldflags;
330 		inode->i_flags = i_oldflags;
331 	}
332 
333  out_unlock:
334 	inode_unlock(inode);
335 	mnt_drop_write_file(file);
336 	return ret;
337 }
338 
339 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
340 {
341 	struct inode *inode = file_inode(file);
342 
343 	return put_user(inode->i_generation, arg);
344 }
345 
346 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
347 {
348 	struct inode *inode = file_inode(file);
349 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
350 	struct btrfs_device *device;
351 	struct request_queue *q;
352 	struct fstrim_range range;
353 	u64 minlen = ULLONG_MAX;
354 	u64 num_devices = 0;
355 	u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
356 	int ret;
357 
358 	if (!capable(CAP_SYS_ADMIN))
359 		return -EPERM;
360 
361 	rcu_read_lock();
362 	list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
363 				dev_list) {
364 		if (!device->bdev)
365 			continue;
366 		q = bdev_get_queue(device->bdev);
367 		if (blk_queue_discard(q)) {
368 			num_devices++;
369 			minlen = min_t(u64, q->limits.discard_granularity,
370 				     minlen);
371 		}
372 	}
373 	rcu_read_unlock();
374 
375 	if (!num_devices)
376 		return -EOPNOTSUPP;
377 	if (copy_from_user(&range, arg, sizeof(range)))
378 		return -EFAULT;
379 	if (range.start > total_bytes ||
380 	    range.len < fs_info->sb->s_blocksize)
381 		return -EINVAL;
382 
383 	range.len = min(range.len, total_bytes - range.start);
384 	range.minlen = max(range.minlen, minlen);
385 	ret = btrfs_trim_fs(fs_info, &range);
386 	if (ret < 0)
387 		return ret;
388 
389 	if (copy_to_user(arg, &range, sizeof(range)))
390 		return -EFAULT;
391 
392 	return 0;
393 }
394 
395 int btrfs_is_empty_uuid(u8 *uuid)
396 {
397 	int i;
398 
399 	for (i = 0; i < BTRFS_UUID_SIZE; i++) {
400 		if (uuid[i])
401 			return 0;
402 	}
403 	return 1;
404 }
405 
406 static noinline int create_subvol(struct inode *dir,
407 				  struct dentry *dentry,
408 				  const char *name, int namelen,
409 				  u64 *async_transid,
410 				  struct btrfs_qgroup_inherit *inherit)
411 {
412 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
413 	struct btrfs_trans_handle *trans;
414 	struct btrfs_key key;
415 	struct btrfs_root_item *root_item;
416 	struct btrfs_inode_item *inode_item;
417 	struct extent_buffer *leaf;
418 	struct btrfs_root *root = BTRFS_I(dir)->root;
419 	struct btrfs_root *new_root;
420 	struct btrfs_block_rsv block_rsv;
421 	struct timespec cur_time = current_time(dir);
422 	struct inode *inode;
423 	int ret;
424 	int err;
425 	u64 objectid;
426 	u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
427 	u64 index = 0;
428 	u64 qgroup_reserved;
429 	uuid_le new_uuid;
430 
431 	root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
432 	if (!root_item)
433 		return -ENOMEM;
434 
435 	ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid);
436 	if (ret)
437 		goto fail_free;
438 
439 	/*
440 	 * Don't create subvolume whose level is not zero. Or qgroup will be
441 	 * screwed up since it assumes subvolume qgroup's level to be 0.
442 	 */
443 	if (btrfs_qgroup_level(objectid)) {
444 		ret = -ENOSPC;
445 		goto fail_free;
446 	}
447 
448 	btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
449 	/*
450 	 * The same as the snapshot creation, please see the comment
451 	 * of create_snapshot().
452 	 */
453 	ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
454 					       8, &qgroup_reserved, false);
455 	if (ret)
456 		goto fail_free;
457 
458 	trans = btrfs_start_transaction(root, 0);
459 	if (IS_ERR(trans)) {
460 		ret = PTR_ERR(trans);
461 		btrfs_subvolume_release_metadata(fs_info, &block_rsv);
462 		goto fail_free;
463 	}
464 	trans->block_rsv = &block_rsv;
465 	trans->bytes_reserved = block_rsv.size;
466 
467 	ret = btrfs_qgroup_inherit(trans, fs_info, 0, objectid, inherit);
468 	if (ret)
469 		goto fail;
470 
471 	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
472 	if (IS_ERR(leaf)) {
473 		ret = PTR_ERR(leaf);
474 		goto fail;
475 	}
476 
477 	memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
478 	btrfs_set_header_bytenr(leaf, leaf->start);
479 	btrfs_set_header_generation(leaf, trans->transid);
480 	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
481 	btrfs_set_header_owner(leaf, objectid);
482 
483 	write_extent_buffer_fsid(leaf, fs_info->fsid);
484 	write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
485 	btrfs_mark_buffer_dirty(leaf);
486 
487 	inode_item = &root_item->inode;
488 	btrfs_set_stack_inode_generation(inode_item, 1);
489 	btrfs_set_stack_inode_size(inode_item, 3);
490 	btrfs_set_stack_inode_nlink(inode_item, 1);
491 	btrfs_set_stack_inode_nbytes(inode_item,
492 				     fs_info->nodesize);
493 	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
494 
495 	btrfs_set_root_flags(root_item, 0);
496 	btrfs_set_root_limit(root_item, 0);
497 	btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
498 
499 	btrfs_set_root_bytenr(root_item, leaf->start);
500 	btrfs_set_root_generation(root_item, trans->transid);
501 	btrfs_set_root_level(root_item, 0);
502 	btrfs_set_root_refs(root_item, 1);
503 	btrfs_set_root_used(root_item, leaf->len);
504 	btrfs_set_root_last_snapshot(root_item, 0);
505 
506 	btrfs_set_root_generation_v2(root_item,
507 			btrfs_root_generation(root_item));
508 	uuid_le_gen(&new_uuid);
509 	memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
510 	btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
511 	btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
512 	root_item->ctime = root_item->otime;
513 	btrfs_set_root_ctransid(root_item, trans->transid);
514 	btrfs_set_root_otransid(root_item, trans->transid);
515 
516 	btrfs_tree_unlock(leaf);
517 	free_extent_buffer(leaf);
518 	leaf = NULL;
519 
520 	btrfs_set_root_dirid(root_item, new_dirid);
521 
522 	key.objectid = objectid;
523 	key.offset = 0;
524 	key.type = BTRFS_ROOT_ITEM_KEY;
525 	ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
526 				root_item);
527 	if (ret)
528 		goto fail;
529 
530 	key.offset = (u64)-1;
531 	new_root = btrfs_read_fs_root_no_name(fs_info, &key);
532 	if (IS_ERR(new_root)) {
533 		ret = PTR_ERR(new_root);
534 		btrfs_abort_transaction(trans, ret);
535 		goto fail;
536 	}
537 
538 	btrfs_record_root_in_trans(trans, new_root);
539 
540 	ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
541 	if (ret) {
542 		/* We potentially lose an unused inode item here */
543 		btrfs_abort_transaction(trans, ret);
544 		goto fail;
545 	}
546 
547 	mutex_lock(&new_root->objectid_mutex);
548 	new_root->highest_objectid = new_dirid;
549 	mutex_unlock(&new_root->objectid_mutex);
550 
551 	/*
552 	 * insert the directory item
553 	 */
554 	ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
555 	if (ret) {
556 		btrfs_abort_transaction(trans, ret);
557 		goto fail;
558 	}
559 
560 	ret = btrfs_insert_dir_item(trans, root,
561 				    name, namelen, BTRFS_I(dir), &key,
562 				    BTRFS_FT_DIR, index);
563 	if (ret) {
564 		btrfs_abort_transaction(trans, ret);
565 		goto fail;
566 	}
567 
568 	btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
569 	ret = btrfs_update_inode(trans, root, dir);
570 	BUG_ON(ret);
571 
572 	ret = btrfs_add_root_ref(trans, fs_info,
573 				 objectid, root->root_key.objectid,
574 				 btrfs_ino(BTRFS_I(dir)), index, name, namelen);
575 	BUG_ON(ret);
576 
577 	ret = btrfs_uuid_tree_add(trans, fs_info, root_item->uuid,
578 				  BTRFS_UUID_KEY_SUBVOL, objectid);
579 	if (ret)
580 		btrfs_abort_transaction(trans, ret);
581 
582 fail:
583 	kfree(root_item);
584 	trans->block_rsv = NULL;
585 	trans->bytes_reserved = 0;
586 	btrfs_subvolume_release_metadata(fs_info, &block_rsv);
587 
588 	if (async_transid) {
589 		*async_transid = trans->transid;
590 		err = btrfs_commit_transaction_async(trans, 1);
591 		if (err)
592 			err = btrfs_commit_transaction(trans);
593 	} else {
594 		err = btrfs_commit_transaction(trans);
595 	}
596 	if (err && !ret)
597 		ret = err;
598 
599 	if (!ret) {
600 		inode = btrfs_lookup_dentry(dir, dentry);
601 		if (IS_ERR(inode))
602 			return PTR_ERR(inode);
603 		d_instantiate(dentry, inode);
604 	}
605 	return ret;
606 
607 fail_free:
608 	kfree(root_item);
609 	return ret;
610 }
611 
612 static void btrfs_wait_for_no_snapshotting_writes(struct btrfs_root *root)
613 {
614 	s64 writers;
615 	DEFINE_WAIT(wait);
616 
617 	do {
618 		prepare_to_wait(&root->subv_writers->wait, &wait,
619 				TASK_UNINTERRUPTIBLE);
620 
621 		writers = percpu_counter_sum(&root->subv_writers->counter);
622 		if (writers)
623 			schedule();
624 
625 		finish_wait(&root->subv_writers->wait, &wait);
626 	} while (writers);
627 }
628 
629 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
630 			   struct dentry *dentry,
631 			   u64 *async_transid, bool readonly,
632 			   struct btrfs_qgroup_inherit *inherit)
633 {
634 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
635 	struct inode *inode;
636 	struct btrfs_pending_snapshot *pending_snapshot;
637 	struct btrfs_trans_handle *trans;
638 	int ret;
639 
640 	if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
641 		return -EINVAL;
642 
643 	pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
644 	if (!pending_snapshot)
645 		return -ENOMEM;
646 
647 	pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
648 			GFP_KERNEL);
649 	pending_snapshot->path = btrfs_alloc_path();
650 	if (!pending_snapshot->root_item || !pending_snapshot->path) {
651 		ret = -ENOMEM;
652 		goto free_pending;
653 	}
654 
655 	atomic_inc(&root->will_be_snapshotted);
656 	smp_mb__after_atomic();
657 	btrfs_wait_for_no_snapshotting_writes(root);
658 
659 	ret = btrfs_start_delalloc_inodes(root, 0);
660 	if (ret)
661 		goto dec_and_free;
662 
663 	btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
664 
665 	btrfs_init_block_rsv(&pending_snapshot->block_rsv,
666 			     BTRFS_BLOCK_RSV_TEMP);
667 	/*
668 	 * 1 - parent dir inode
669 	 * 2 - dir entries
670 	 * 1 - root item
671 	 * 2 - root ref/backref
672 	 * 1 - root of snapshot
673 	 * 1 - UUID item
674 	 */
675 	ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
676 					&pending_snapshot->block_rsv, 8,
677 					&pending_snapshot->qgroup_reserved,
678 					false);
679 	if (ret)
680 		goto dec_and_free;
681 
682 	pending_snapshot->dentry = dentry;
683 	pending_snapshot->root = root;
684 	pending_snapshot->readonly = readonly;
685 	pending_snapshot->dir = dir;
686 	pending_snapshot->inherit = inherit;
687 
688 	trans = btrfs_start_transaction(root, 0);
689 	if (IS_ERR(trans)) {
690 		ret = PTR_ERR(trans);
691 		goto fail;
692 	}
693 
694 	spin_lock(&fs_info->trans_lock);
695 	list_add(&pending_snapshot->list,
696 		 &trans->transaction->pending_snapshots);
697 	spin_unlock(&fs_info->trans_lock);
698 	if (async_transid) {
699 		*async_transid = trans->transid;
700 		ret = btrfs_commit_transaction_async(trans, 1);
701 		if (ret)
702 			ret = btrfs_commit_transaction(trans);
703 	} else {
704 		ret = btrfs_commit_transaction(trans);
705 	}
706 	if (ret)
707 		goto fail;
708 
709 	ret = pending_snapshot->error;
710 	if (ret)
711 		goto fail;
712 
713 	ret = btrfs_orphan_cleanup(pending_snapshot->snap);
714 	if (ret)
715 		goto fail;
716 
717 	inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
718 	if (IS_ERR(inode)) {
719 		ret = PTR_ERR(inode);
720 		goto fail;
721 	}
722 
723 	d_instantiate(dentry, inode);
724 	ret = 0;
725 fail:
726 	btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
727 dec_and_free:
728 	if (atomic_dec_and_test(&root->will_be_snapshotted))
729 		wake_up_atomic_t(&root->will_be_snapshotted);
730 free_pending:
731 	kfree(pending_snapshot->root_item);
732 	btrfs_free_path(pending_snapshot->path);
733 	kfree(pending_snapshot);
734 
735 	return ret;
736 }
737 
738 /*  copy of may_delete in fs/namei.c()
739  *	Check whether we can remove a link victim from directory dir, check
740  *  whether the type of victim is right.
741  *  1. We can't do it if dir is read-only (done in permission())
742  *  2. We should have write and exec permissions on dir
743  *  3. We can't remove anything from append-only dir
744  *  4. We can't do anything with immutable dir (done in permission())
745  *  5. If the sticky bit on dir is set we should either
746  *	a. be owner of dir, or
747  *	b. be owner of victim, or
748  *	c. have CAP_FOWNER capability
749  *  6. If the victim is append-only or immutable we can't do anything with
750  *     links pointing to it.
751  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
752  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
753  *  9. We can't remove a root or mountpoint.
754  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
755  *     nfs_async_unlink().
756  */
757 
758 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
759 {
760 	int error;
761 
762 	if (d_really_is_negative(victim))
763 		return -ENOENT;
764 
765 	BUG_ON(d_inode(victim->d_parent) != dir);
766 	audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
767 
768 	error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
769 	if (error)
770 		return error;
771 	if (IS_APPEND(dir))
772 		return -EPERM;
773 	if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
774 	    IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
775 		return -EPERM;
776 	if (isdir) {
777 		if (!d_is_dir(victim))
778 			return -ENOTDIR;
779 		if (IS_ROOT(victim))
780 			return -EBUSY;
781 	} else if (d_is_dir(victim))
782 		return -EISDIR;
783 	if (IS_DEADDIR(dir))
784 		return -ENOENT;
785 	if (victim->d_flags & DCACHE_NFSFS_RENAMED)
786 		return -EBUSY;
787 	return 0;
788 }
789 
790 /* copy of may_create in fs/namei.c() */
791 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
792 {
793 	if (d_really_is_positive(child))
794 		return -EEXIST;
795 	if (IS_DEADDIR(dir))
796 		return -ENOENT;
797 	return inode_permission(dir, MAY_WRITE | MAY_EXEC);
798 }
799 
800 /*
801  * Create a new subvolume below @parent.  This is largely modeled after
802  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
803  * inside this filesystem so it's quite a bit simpler.
804  */
805 static noinline int btrfs_mksubvol(const struct path *parent,
806 				   const char *name, int namelen,
807 				   struct btrfs_root *snap_src,
808 				   u64 *async_transid, bool readonly,
809 				   struct btrfs_qgroup_inherit *inherit)
810 {
811 	struct inode *dir = d_inode(parent->dentry);
812 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
813 	struct dentry *dentry;
814 	int error;
815 
816 	error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
817 	if (error == -EINTR)
818 		return error;
819 
820 	dentry = lookup_one_len(name, parent->dentry, namelen);
821 	error = PTR_ERR(dentry);
822 	if (IS_ERR(dentry))
823 		goto out_unlock;
824 
825 	error = btrfs_may_create(dir, dentry);
826 	if (error)
827 		goto out_dput;
828 
829 	/*
830 	 * even if this name doesn't exist, we may get hash collisions.
831 	 * check for them now when we can safely fail
832 	 */
833 	error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
834 					       dir->i_ino, name,
835 					       namelen);
836 	if (error)
837 		goto out_dput;
838 
839 	down_read(&fs_info->subvol_sem);
840 
841 	if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
842 		goto out_up_read;
843 
844 	if (snap_src) {
845 		error = create_snapshot(snap_src, dir, dentry,
846 					async_transid, readonly, inherit);
847 	} else {
848 		error = create_subvol(dir, dentry, name, namelen,
849 				      async_transid, inherit);
850 	}
851 	if (!error)
852 		fsnotify_mkdir(dir, dentry);
853 out_up_read:
854 	up_read(&fs_info->subvol_sem);
855 out_dput:
856 	dput(dentry);
857 out_unlock:
858 	inode_unlock(dir);
859 	return error;
860 }
861 
862 /*
863  * When we're defragging a range, we don't want to kick it off again
864  * if it is really just waiting for delalloc to send it down.
865  * If we find a nice big extent or delalloc range for the bytes in the
866  * file you want to defrag, we return 0 to let you know to skip this
867  * part of the file
868  */
869 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
870 {
871 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
872 	struct extent_map *em = NULL;
873 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
874 	u64 end;
875 
876 	read_lock(&em_tree->lock);
877 	em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
878 	read_unlock(&em_tree->lock);
879 
880 	if (em) {
881 		end = extent_map_end(em);
882 		free_extent_map(em);
883 		if (end - offset > thresh)
884 			return 0;
885 	}
886 	/* if we already have a nice delalloc here, just stop */
887 	thresh /= 2;
888 	end = count_range_bits(io_tree, &offset, offset + thresh,
889 			       thresh, EXTENT_DELALLOC, 1);
890 	if (end >= thresh)
891 		return 0;
892 	return 1;
893 }
894 
895 /*
896  * helper function to walk through a file and find extents
897  * newer than a specific transid, and smaller than thresh.
898  *
899  * This is used by the defragging code to find new and small
900  * extents
901  */
902 static int find_new_extents(struct btrfs_root *root,
903 			    struct inode *inode, u64 newer_than,
904 			    u64 *off, u32 thresh)
905 {
906 	struct btrfs_path *path;
907 	struct btrfs_key min_key;
908 	struct extent_buffer *leaf;
909 	struct btrfs_file_extent_item *extent;
910 	int type;
911 	int ret;
912 	u64 ino = btrfs_ino(BTRFS_I(inode));
913 
914 	path = btrfs_alloc_path();
915 	if (!path)
916 		return -ENOMEM;
917 
918 	min_key.objectid = ino;
919 	min_key.type = BTRFS_EXTENT_DATA_KEY;
920 	min_key.offset = *off;
921 
922 	while (1) {
923 		ret = btrfs_search_forward(root, &min_key, path, newer_than);
924 		if (ret != 0)
925 			goto none;
926 process_slot:
927 		if (min_key.objectid != ino)
928 			goto none;
929 		if (min_key.type != BTRFS_EXTENT_DATA_KEY)
930 			goto none;
931 
932 		leaf = path->nodes[0];
933 		extent = btrfs_item_ptr(leaf, path->slots[0],
934 					struct btrfs_file_extent_item);
935 
936 		type = btrfs_file_extent_type(leaf, extent);
937 		if (type == BTRFS_FILE_EXTENT_REG &&
938 		    btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
939 		    check_defrag_in_cache(inode, min_key.offset, thresh)) {
940 			*off = min_key.offset;
941 			btrfs_free_path(path);
942 			return 0;
943 		}
944 
945 		path->slots[0]++;
946 		if (path->slots[0] < btrfs_header_nritems(leaf)) {
947 			btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
948 			goto process_slot;
949 		}
950 
951 		if (min_key.offset == (u64)-1)
952 			goto none;
953 
954 		min_key.offset++;
955 		btrfs_release_path(path);
956 	}
957 none:
958 	btrfs_free_path(path);
959 	return -ENOENT;
960 }
961 
962 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
963 {
964 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
965 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
966 	struct extent_map *em;
967 	u64 len = PAGE_SIZE;
968 
969 	/*
970 	 * hopefully we have this extent in the tree already, try without
971 	 * the full extent lock
972 	 */
973 	read_lock(&em_tree->lock);
974 	em = lookup_extent_mapping(em_tree, start, len);
975 	read_unlock(&em_tree->lock);
976 
977 	if (!em) {
978 		struct extent_state *cached = NULL;
979 		u64 end = start + len - 1;
980 
981 		/* get the big lock and read metadata off disk */
982 		lock_extent_bits(io_tree, start, end, &cached);
983 		em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
984 		unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
985 
986 		if (IS_ERR(em))
987 			return NULL;
988 	}
989 
990 	return em;
991 }
992 
993 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
994 {
995 	struct extent_map *next;
996 	bool ret = true;
997 
998 	/* this is the last extent */
999 	if (em->start + em->len >= i_size_read(inode))
1000 		return false;
1001 
1002 	next = defrag_lookup_extent(inode, em->start + em->len);
1003 	if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1004 		ret = false;
1005 	else if ((em->block_start + em->block_len == next->block_start) &&
1006 		 (em->block_len > SZ_128K && next->block_len > SZ_128K))
1007 		ret = false;
1008 
1009 	free_extent_map(next);
1010 	return ret;
1011 }
1012 
1013 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1014 			       u64 *last_len, u64 *skip, u64 *defrag_end,
1015 			       int compress)
1016 {
1017 	struct extent_map *em;
1018 	int ret = 1;
1019 	bool next_mergeable = true;
1020 	bool prev_mergeable = true;
1021 
1022 	/*
1023 	 * make sure that once we start defragging an extent, we keep on
1024 	 * defragging it
1025 	 */
1026 	if (start < *defrag_end)
1027 		return 1;
1028 
1029 	*skip = 0;
1030 
1031 	em = defrag_lookup_extent(inode, start);
1032 	if (!em)
1033 		return 0;
1034 
1035 	/* this will cover holes, and inline extents */
1036 	if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1037 		ret = 0;
1038 		goto out;
1039 	}
1040 
1041 	if (!*defrag_end)
1042 		prev_mergeable = false;
1043 
1044 	next_mergeable = defrag_check_next_extent(inode, em);
1045 	/*
1046 	 * we hit a real extent, if it is big or the next extent is not a
1047 	 * real extent, don't bother defragging it
1048 	 */
1049 	if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1050 	    (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1051 		ret = 0;
1052 out:
1053 	/*
1054 	 * last_len ends up being a counter of how many bytes we've defragged.
1055 	 * every time we choose not to defrag an extent, we reset *last_len
1056 	 * so that the next tiny extent will force a defrag.
1057 	 *
1058 	 * The end result of this is that tiny extents before a single big
1059 	 * extent will force at least part of that big extent to be defragged.
1060 	 */
1061 	if (ret) {
1062 		*defrag_end = extent_map_end(em);
1063 	} else {
1064 		*last_len = 0;
1065 		*skip = extent_map_end(em);
1066 		*defrag_end = 0;
1067 	}
1068 
1069 	free_extent_map(em);
1070 	return ret;
1071 }
1072 
1073 /*
1074  * it doesn't do much good to defrag one or two pages
1075  * at a time.  This pulls in a nice chunk of pages
1076  * to COW and defrag.
1077  *
1078  * It also makes sure the delalloc code has enough
1079  * dirty data to avoid making new small extents as part
1080  * of the defrag
1081  *
1082  * It's a good idea to start RA on this range
1083  * before calling this.
1084  */
1085 static int cluster_pages_for_defrag(struct inode *inode,
1086 				    struct page **pages,
1087 				    unsigned long start_index,
1088 				    unsigned long num_pages)
1089 {
1090 	unsigned long file_end;
1091 	u64 isize = i_size_read(inode);
1092 	u64 page_start;
1093 	u64 page_end;
1094 	u64 page_cnt;
1095 	int ret;
1096 	int i;
1097 	int i_done;
1098 	struct btrfs_ordered_extent *ordered;
1099 	struct extent_state *cached_state = NULL;
1100 	struct extent_io_tree *tree;
1101 	struct extent_changeset *data_reserved = NULL;
1102 	gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1103 
1104 	file_end = (isize - 1) >> PAGE_SHIFT;
1105 	if (!isize || start_index > file_end)
1106 		return 0;
1107 
1108 	page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1109 
1110 	ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
1111 			start_index << PAGE_SHIFT,
1112 			page_cnt << PAGE_SHIFT);
1113 	if (ret)
1114 		return ret;
1115 	i_done = 0;
1116 	tree = &BTRFS_I(inode)->io_tree;
1117 
1118 	/* step one, lock all the pages */
1119 	for (i = 0; i < page_cnt; i++) {
1120 		struct page *page;
1121 again:
1122 		page = find_or_create_page(inode->i_mapping,
1123 					   start_index + i, mask);
1124 		if (!page)
1125 			break;
1126 
1127 		page_start = page_offset(page);
1128 		page_end = page_start + PAGE_SIZE - 1;
1129 		while (1) {
1130 			lock_extent_bits(tree, page_start, page_end,
1131 					 &cached_state);
1132 			ordered = btrfs_lookup_ordered_extent(inode,
1133 							      page_start);
1134 			unlock_extent_cached(tree, page_start, page_end,
1135 					     &cached_state, GFP_NOFS);
1136 			if (!ordered)
1137 				break;
1138 
1139 			unlock_page(page);
1140 			btrfs_start_ordered_extent(inode, ordered, 1);
1141 			btrfs_put_ordered_extent(ordered);
1142 			lock_page(page);
1143 			/*
1144 			 * we unlocked the page above, so we need check if
1145 			 * it was released or not.
1146 			 */
1147 			if (page->mapping != inode->i_mapping) {
1148 				unlock_page(page);
1149 				put_page(page);
1150 				goto again;
1151 			}
1152 		}
1153 
1154 		if (!PageUptodate(page)) {
1155 			btrfs_readpage(NULL, page);
1156 			lock_page(page);
1157 			if (!PageUptodate(page)) {
1158 				unlock_page(page);
1159 				put_page(page);
1160 				ret = -EIO;
1161 				break;
1162 			}
1163 		}
1164 
1165 		if (page->mapping != inode->i_mapping) {
1166 			unlock_page(page);
1167 			put_page(page);
1168 			goto again;
1169 		}
1170 
1171 		pages[i] = page;
1172 		i_done++;
1173 	}
1174 	if (!i_done || ret)
1175 		goto out;
1176 
1177 	if (!(inode->i_sb->s_flags & MS_ACTIVE))
1178 		goto out;
1179 
1180 	/*
1181 	 * so now we have a nice long stream of locked
1182 	 * and up to date pages, lets wait on them
1183 	 */
1184 	for (i = 0; i < i_done; i++)
1185 		wait_on_page_writeback(pages[i]);
1186 
1187 	page_start = page_offset(pages[0]);
1188 	page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1189 
1190 	lock_extent_bits(&BTRFS_I(inode)->io_tree,
1191 			 page_start, page_end - 1, &cached_state);
1192 	clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1193 			  page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1194 			  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1195 			  &cached_state, GFP_NOFS);
1196 
1197 	if (i_done != page_cnt) {
1198 		spin_lock(&BTRFS_I(inode)->lock);
1199 		BTRFS_I(inode)->outstanding_extents++;
1200 		spin_unlock(&BTRFS_I(inode)->lock);
1201 		btrfs_delalloc_release_space(inode, data_reserved,
1202 				start_index << PAGE_SHIFT,
1203 				(page_cnt - i_done) << PAGE_SHIFT);
1204 	}
1205 
1206 
1207 	set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1208 			  &cached_state);
1209 
1210 	unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1211 			     page_start, page_end - 1, &cached_state,
1212 			     GFP_NOFS);
1213 
1214 	for (i = 0; i < i_done; i++) {
1215 		clear_page_dirty_for_io(pages[i]);
1216 		ClearPageChecked(pages[i]);
1217 		set_page_extent_mapped(pages[i]);
1218 		set_page_dirty(pages[i]);
1219 		unlock_page(pages[i]);
1220 		put_page(pages[i]);
1221 	}
1222 	extent_changeset_free(data_reserved);
1223 	return i_done;
1224 out:
1225 	for (i = 0; i < i_done; i++) {
1226 		unlock_page(pages[i]);
1227 		put_page(pages[i]);
1228 	}
1229 	btrfs_delalloc_release_space(inode, data_reserved,
1230 			start_index << PAGE_SHIFT,
1231 			page_cnt << PAGE_SHIFT);
1232 	extent_changeset_free(data_reserved);
1233 	return ret;
1234 
1235 }
1236 
1237 int btrfs_defrag_file(struct inode *inode, struct file *file,
1238 		      struct btrfs_ioctl_defrag_range_args *range,
1239 		      u64 newer_than, unsigned long max_to_defrag)
1240 {
1241 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1242 	struct btrfs_root *root = BTRFS_I(inode)->root;
1243 	struct file_ra_state *ra = NULL;
1244 	unsigned long last_index;
1245 	u64 isize = i_size_read(inode);
1246 	u64 last_len = 0;
1247 	u64 skip = 0;
1248 	u64 defrag_end = 0;
1249 	u64 newer_off = range->start;
1250 	unsigned long i;
1251 	unsigned long ra_index = 0;
1252 	int ret;
1253 	int defrag_count = 0;
1254 	int compress_type = BTRFS_COMPRESS_ZLIB;
1255 	u32 extent_thresh = range->extent_thresh;
1256 	unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1257 	unsigned long cluster = max_cluster;
1258 	u64 new_align = ~((u64)SZ_128K - 1);
1259 	struct page **pages = NULL;
1260 	bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1261 
1262 	if (isize == 0)
1263 		return 0;
1264 
1265 	if (range->start >= isize)
1266 		return -EINVAL;
1267 
1268 	if (do_compress) {
1269 		if (range->compress_type > BTRFS_COMPRESS_TYPES)
1270 			return -EINVAL;
1271 		if (range->compress_type)
1272 			compress_type = range->compress_type;
1273 	}
1274 
1275 	if (extent_thresh == 0)
1276 		extent_thresh = SZ_256K;
1277 
1278 	/*
1279 	 * If we were not given a file, allocate a readahead context. As
1280 	 * readahead is just an optimization, defrag will work without it so
1281 	 * we don't error out.
1282 	 */
1283 	if (!file) {
1284 		ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1285 		if (ra)
1286 			file_ra_state_init(ra, inode->i_mapping);
1287 	} else {
1288 		ra = &file->f_ra;
1289 	}
1290 
1291 	pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1292 	if (!pages) {
1293 		ret = -ENOMEM;
1294 		goto out_ra;
1295 	}
1296 
1297 	/* find the last page to defrag */
1298 	if (range->start + range->len > range->start) {
1299 		last_index = min_t(u64, isize - 1,
1300 			 range->start + range->len - 1) >> PAGE_SHIFT;
1301 	} else {
1302 		last_index = (isize - 1) >> PAGE_SHIFT;
1303 	}
1304 
1305 	if (newer_than) {
1306 		ret = find_new_extents(root, inode, newer_than,
1307 				       &newer_off, SZ_64K);
1308 		if (!ret) {
1309 			range->start = newer_off;
1310 			/*
1311 			 * we always align our defrag to help keep
1312 			 * the extents in the file evenly spaced
1313 			 */
1314 			i = (newer_off & new_align) >> PAGE_SHIFT;
1315 		} else
1316 			goto out_ra;
1317 	} else {
1318 		i = range->start >> PAGE_SHIFT;
1319 	}
1320 	if (!max_to_defrag)
1321 		max_to_defrag = last_index - i + 1;
1322 
1323 	/*
1324 	 * make writeback starts from i, so the defrag range can be
1325 	 * written sequentially.
1326 	 */
1327 	if (i < inode->i_mapping->writeback_index)
1328 		inode->i_mapping->writeback_index = i;
1329 
1330 	while (i <= last_index && defrag_count < max_to_defrag &&
1331 	       (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1332 		/*
1333 		 * make sure we stop running if someone unmounts
1334 		 * the FS
1335 		 */
1336 		if (!(inode->i_sb->s_flags & MS_ACTIVE))
1337 			break;
1338 
1339 		if (btrfs_defrag_cancelled(fs_info)) {
1340 			btrfs_debug(fs_info, "defrag_file cancelled");
1341 			ret = -EAGAIN;
1342 			break;
1343 		}
1344 
1345 		if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1346 					 extent_thresh, &last_len, &skip,
1347 					 &defrag_end, do_compress)){
1348 			unsigned long next;
1349 			/*
1350 			 * the should_defrag function tells us how much to skip
1351 			 * bump our counter by the suggested amount
1352 			 */
1353 			next = DIV_ROUND_UP(skip, PAGE_SIZE);
1354 			i = max(i + 1, next);
1355 			continue;
1356 		}
1357 
1358 		if (!newer_than) {
1359 			cluster = (PAGE_ALIGN(defrag_end) >>
1360 				   PAGE_SHIFT) - i;
1361 			cluster = min(cluster, max_cluster);
1362 		} else {
1363 			cluster = max_cluster;
1364 		}
1365 
1366 		if (i + cluster > ra_index) {
1367 			ra_index = max(i, ra_index);
1368 			if (ra)
1369 				page_cache_sync_readahead(inode->i_mapping, ra,
1370 						file, ra_index, cluster);
1371 			ra_index += cluster;
1372 		}
1373 
1374 		inode_lock(inode);
1375 		if (do_compress)
1376 			BTRFS_I(inode)->defrag_compress = compress_type;
1377 		ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1378 		if (ret < 0) {
1379 			inode_unlock(inode);
1380 			goto out_ra;
1381 		}
1382 
1383 		defrag_count += ret;
1384 		balance_dirty_pages_ratelimited(inode->i_mapping);
1385 		inode_unlock(inode);
1386 
1387 		if (newer_than) {
1388 			if (newer_off == (u64)-1)
1389 				break;
1390 
1391 			if (ret > 0)
1392 				i += ret;
1393 
1394 			newer_off = max(newer_off + 1,
1395 					(u64)i << PAGE_SHIFT);
1396 
1397 			ret = find_new_extents(root, inode, newer_than,
1398 					       &newer_off, SZ_64K);
1399 			if (!ret) {
1400 				range->start = newer_off;
1401 				i = (newer_off & new_align) >> PAGE_SHIFT;
1402 			} else {
1403 				break;
1404 			}
1405 		} else {
1406 			if (ret > 0) {
1407 				i += ret;
1408 				last_len += ret << PAGE_SHIFT;
1409 			} else {
1410 				i++;
1411 				last_len = 0;
1412 			}
1413 		}
1414 	}
1415 
1416 	if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1417 		filemap_flush(inode->i_mapping);
1418 		if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1419 			     &BTRFS_I(inode)->runtime_flags))
1420 			filemap_flush(inode->i_mapping);
1421 	}
1422 
1423 	if (do_compress) {
1424 		/* the filemap_flush will queue IO into the worker threads, but
1425 		 * we have to make sure the IO is actually started and that
1426 		 * ordered extents get created before we return
1427 		 */
1428 		atomic_inc(&fs_info->async_submit_draining);
1429 		while (atomic_read(&fs_info->nr_async_submits) ||
1430 		       atomic_read(&fs_info->async_delalloc_pages)) {
1431 			wait_event(fs_info->async_submit_wait,
1432 				   (atomic_read(&fs_info->nr_async_submits) == 0 &&
1433 				    atomic_read(&fs_info->async_delalloc_pages) == 0));
1434 		}
1435 		atomic_dec(&fs_info->async_submit_draining);
1436 	}
1437 
1438 	if (range->compress_type == BTRFS_COMPRESS_LZO) {
1439 		btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1440 	} else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1441 		btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1442 	}
1443 
1444 	ret = defrag_count;
1445 
1446 out_ra:
1447 	if (do_compress) {
1448 		inode_lock(inode);
1449 		BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1450 		inode_unlock(inode);
1451 	}
1452 	if (!file)
1453 		kfree(ra);
1454 	kfree(pages);
1455 	return ret;
1456 }
1457 
1458 static noinline int btrfs_ioctl_resize(struct file *file,
1459 					void __user *arg)
1460 {
1461 	struct inode *inode = file_inode(file);
1462 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1463 	u64 new_size;
1464 	u64 old_size;
1465 	u64 devid = 1;
1466 	struct btrfs_root *root = BTRFS_I(inode)->root;
1467 	struct btrfs_ioctl_vol_args *vol_args;
1468 	struct btrfs_trans_handle *trans;
1469 	struct btrfs_device *device = NULL;
1470 	char *sizestr;
1471 	char *retptr;
1472 	char *devstr = NULL;
1473 	int ret = 0;
1474 	int mod = 0;
1475 
1476 	if (!capable(CAP_SYS_ADMIN))
1477 		return -EPERM;
1478 
1479 	ret = mnt_want_write_file(file);
1480 	if (ret)
1481 		return ret;
1482 
1483 	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
1484 		mnt_drop_write_file(file);
1485 		return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1486 	}
1487 
1488 	mutex_lock(&fs_info->volume_mutex);
1489 	vol_args = memdup_user(arg, sizeof(*vol_args));
1490 	if (IS_ERR(vol_args)) {
1491 		ret = PTR_ERR(vol_args);
1492 		goto out;
1493 	}
1494 
1495 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1496 
1497 	sizestr = vol_args->name;
1498 	devstr = strchr(sizestr, ':');
1499 	if (devstr) {
1500 		sizestr = devstr + 1;
1501 		*devstr = '\0';
1502 		devstr = vol_args->name;
1503 		ret = kstrtoull(devstr, 10, &devid);
1504 		if (ret)
1505 			goto out_free;
1506 		if (!devid) {
1507 			ret = -EINVAL;
1508 			goto out_free;
1509 		}
1510 		btrfs_info(fs_info, "resizing devid %llu", devid);
1511 	}
1512 
1513 	device = btrfs_find_device(fs_info, devid, NULL, NULL);
1514 	if (!device) {
1515 		btrfs_info(fs_info, "resizer unable to find device %llu",
1516 			   devid);
1517 		ret = -ENODEV;
1518 		goto out_free;
1519 	}
1520 
1521 	if (!device->writeable) {
1522 		btrfs_info(fs_info,
1523 			   "resizer unable to apply on readonly device %llu",
1524 		       devid);
1525 		ret = -EPERM;
1526 		goto out_free;
1527 	}
1528 
1529 	if (!strcmp(sizestr, "max"))
1530 		new_size = device->bdev->bd_inode->i_size;
1531 	else {
1532 		if (sizestr[0] == '-') {
1533 			mod = -1;
1534 			sizestr++;
1535 		} else if (sizestr[0] == '+') {
1536 			mod = 1;
1537 			sizestr++;
1538 		}
1539 		new_size = memparse(sizestr, &retptr);
1540 		if (*retptr != '\0' || new_size == 0) {
1541 			ret = -EINVAL;
1542 			goto out_free;
1543 		}
1544 	}
1545 
1546 	if (device->is_tgtdev_for_dev_replace) {
1547 		ret = -EPERM;
1548 		goto out_free;
1549 	}
1550 
1551 	old_size = btrfs_device_get_total_bytes(device);
1552 
1553 	if (mod < 0) {
1554 		if (new_size > old_size) {
1555 			ret = -EINVAL;
1556 			goto out_free;
1557 		}
1558 		new_size = old_size - new_size;
1559 	} else if (mod > 0) {
1560 		if (new_size > ULLONG_MAX - old_size) {
1561 			ret = -ERANGE;
1562 			goto out_free;
1563 		}
1564 		new_size = old_size + new_size;
1565 	}
1566 
1567 	if (new_size < SZ_256M) {
1568 		ret = -EINVAL;
1569 		goto out_free;
1570 	}
1571 	if (new_size > device->bdev->bd_inode->i_size) {
1572 		ret = -EFBIG;
1573 		goto out_free;
1574 	}
1575 
1576 	new_size = round_down(new_size, fs_info->sectorsize);
1577 
1578 	btrfs_info_in_rcu(fs_info, "new size for %s is %llu",
1579 			  rcu_str_deref(device->name), new_size);
1580 
1581 	if (new_size > old_size) {
1582 		trans = btrfs_start_transaction(root, 0);
1583 		if (IS_ERR(trans)) {
1584 			ret = PTR_ERR(trans);
1585 			goto out_free;
1586 		}
1587 		ret = btrfs_grow_device(trans, device, new_size);
1588 		btrfs_commit_transaction(trans);
1589 	} else if (new_size < old_size) {
1590 		ret = btrfs_shrink_device(device, new_size);
1591 	} /* equal, nothing need to do */
1592 
1593 out_free:
1594 	kfree(vol_args);
1595 out:
1596 	mutex_unlock(&fs_info->volume_mutex);
1597 	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
1598 	mnt_drop_write_file(file);
1599 	return ret;
1600 }
1601 
1602 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1603 				const char *name, unsigned long fd, int subvol,
1604 				u64 *transid, bool readonly,
1605 				struct btrfs_qgroup_inherit *inherit)
1606 {
1607 	int namelen;
1608 	int ret = 0;
1609 
1610 	if (!S_ISDIR(file_inode(file)->i_mode))
1611 		return -ENOTDIR;
1612 
1613 	ret = mnt_want_write_file(file);
1614 	if (ret)
1615 		goto out;
1616 
1617 	namelen = strlen(name);
1618 	if (strchr(name, '/')) {
1619 		ret = -EINVAL;
1620 		goto out_drop_write;
1621 	}
1622 
1623 	if (name[0] == '.' &&
1624 	   (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1625 		ret = -EEXIST;
1626 		goto out_drop_write;
1627 	}
1628 
1629 	if (subvol) {
1630 		ret = btrfs_mksubvol(&file->f_path, name, namelen,
1631 				     NULL, transid, readonly, inherit);
1632 	} else {
1633 		struct fd src = fdget(fd);
1634 		struct inode *src_inode;
1635 		if (!src.file) {
1636 			ret = -EINVAL;
1637 			goto out_drop_write;
1638 		}
1639 
1640 		src_inode = file_inode(src.file);
1641 		if (src_inode->i_sb != file_inode(file)->i_sb) {
1642 			btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1643 				   "Snapshot src from another FS");
1644 			ret = -EXDEV;
1645 		} else if (!inode_owner_or_capable(src_inode)) {
1646 			/*
1647 			 * Subvolume creation is not restricted, but snapshots
1648 			 * are limited to own subvolumes only
1649 			 */
1650 			ret = -EPERM;
1651 		} else {
1652 			ret = btrfs_mksubvol(&file->f_path, name, namelen,
1653 					     BTRFS_I(src_inode)->root,
1654 					     transid, readonly, inherit);
1655 		}
1656 		fdput(src);
1657 	}
1658 out_drop_write:
1659 	mnt_drop_write_file(file);
1660 out:
1661 	return ret;
1662 }
1663 
1664 static noinline int btrfs_ioctl_snap_create(struct file *file,
1665 					    void __user *arg, int subvol)
1666 {
1667 	struct btrfs_ioctl_vol_args *vol_args;
1668 	int ret;
1669 
1670 	if (!S_ISDIR(file_inode(file)->i_mode))
1671 		return -ENOTDIR;
1672 
1673 	vol_args = memdup_user(arg, sizeof(*vol_args));
1674 	if (IS_ERR(vol_args))
1675 		return PTR_ERR(vol_args);
1676 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1677 
1678 	ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1679 					      vol_args->fd, subvol,
1680 					      NULL, false, NULL);
1681 
1682 	kfree(vol_args);
1683 	return ret;
1684 }
1685 
1686 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1687 					       void __user *arg, int subvol)
1688 {
1689 	struct btrfs_ioctl_vol_args_v2 *vol_args;
1690 	int ret;
1691 	u64 transid = 0;
1692 	u64 *ptr = NULL;
1693 	bool readonly = false;
1694 	struct btrfs_qgroup_inherit *inherit = NULL;
1695 
1696 	if (!S_ISDIR(file_inode(file)->i_mode))
1697 		return -ENOTDIR;
1698 
1699 	vol_args = memdup_user(arg, sizeof(*vol_args));
1700 	if (IS_ERR(vol_args))
1701 		return PTR_ERR(vol_args);
1702 	vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1703 
1704 	if (vol_args->flags &
1705 	    ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1706 	      BTRFS_SUBVOL_QGROUP_INHERIT)) {
1707 		ret = -EOPNOTSUPP;
1708 		goto free_args;
1709 	}
1710 
1711 	if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1712 		ptr = &transid;
1713 	if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1714 		readonly = true;
1715 	if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1716 		if (vol_args->size > PAGE_SIZE) {
1717 			ret = -EINVAL;
1718 			goto free_args;
1719 		}
1720 		inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1721 		if (IS_ERR(inherit)) {
1722 			ret = PTR_ERR(inherit);
1723 			goto free_args;
1724 		}
1725 	}
1726 
1727 	ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1728 					      vol_args->fd, subvol, ptr,
1729 					      readonly, inherit);
1730 	if (ret)
1731 		goto free_inherit;
1732 
1733 	if (ptr && copy_to_user(arg +
1734 				offsetof(struct btrfs_ioctl_vol_args_v2,
1735 					transid),
1736 				ptr, sizeof(*ptr)))
1737 		ret = -EFAULT;
1738 
1739 free_inherit:
1740 	kfree(inherit);
1741 free_args:
1742 	kfree(vol_args);
1743 	return ret;
1744 }
1745 
1746 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1747 						void __user *arg)
1748 {
1749 	struct inode *inode = file_inode(file);
1750 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1751 	struct btrfs_root *root = BTRFS_I(inode)->root;
1752 	int ret = 0;
1753 	u64 flags = 0;
1754 
1755 	if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1756 		return -EINVAL;
1757 
1758 	down_read(&fs_info->subvol_sem);
1759 	if (btrfs_root_readonly(root))
1760 		flags |= BTRFS_SUBVOL_RDONLY;
1761 	up_read(&fs_info->subvol_sem);
1762 
1763 	if (copy_to_user(arg, &flags, sizeof(flags)))
1764 		ret = -EFAULT;
1765 
1766 	return ret;
1767 }
1768 
1769 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1770 					      void __user *arg)
1771 {
1772 	struct inode *inode = file_inode(file);
1773 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1774 	struct btrfs_root *root = BTRFS_I(inode)->root;
1775 	struct btrfs_trans_handle *trans;
1776 	u64 root_flags;
1777 	u64 flags;
1778 	int ret = 0;
1779 
1780 	if (!inode_owner_or_capable(inode))
1781 		return -EPERM;
1782 
1783 	ret = mnt_want_write_file(file);
1784 	if (ret)
1785 		goto out;
1786 
1787 	if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1788 		ret = -EINVAL;
1789 		goto out_drop_write;
1790 	}
1791 
1792 	if (copy_from_user(&flags, arg, sizeof(flags))) {
1793 		ret = -EFAULT;
1794 		goto out_drop_write;
1795 	}
1796 
1797 	if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1798 		ret = -EINVAL;
1799 		goto out_drop_write;
1800 	}
1801 
1802 	if (flags & ~BTRFS_SUBVOL_RDONLY) {
1803 		ret = -EOPNOTSUPP;
1804 		goto out_drop_write;
1805 	}
1806 
1807 	down_write(&fs_info->subvol_sem);
1808 
1809 	/* nothing to do */
1810 	if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1811 		goto out_drop_sem;
1812 
1813 	root_flags = btrfs_root_flags(&root->root_item);
1814 	if (flags & BTRFS_SUBVOL_RDONLY) {
1815 		btrfs_set_root_flags(&root->root_item,
1816 				     root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1817 	} else {
1818 		/*
1819 		 * Block RO -> RW transition if this subvolume is involved in
1820 		 * send
1821 		 */
1822 		spin_lock(&root->root_item_lock);
1823 		if (root->send_in_progress == 0) {
1824 			btrfs_set_root_flags(&root->root_item,
1825 				     root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1826 			spin_unlock(&root->root_item_lock);
1827 		} else {
1828 			spin_unlock(&root->root_item_lock);
1829 			btrfs_warn(fs_info,
1830 				   "Attempt to set subvolume %llu read-write during send",
1831 				   root->root_key.objectid);
1832 			ret = -EPERM;
1833 			goto out_drop_sem;
1834 		}
1835 	}
1836 
1837 	trans = btrfs_start_transaction(root, 1);
1838 	if (IS_ERR(trans)) {
1839 		ret = PTR_ERR(trans);
1840 		goto out_reset;
1841 	}
1842 
1843 	ret = btrfs_update_root(trans, fs_info->tree_root,
1844 				&root->root_key, &root->root_item);
1845 
1846 	btrfs_commit_transaction(trans);
1847 out_reset:
1848 	if (ret)
1849 		btrfs_set_root_flags(&root->root_item, root_flags);
1850 out_drop_sem:
1851 	up_write(&fs_info->subvol_sem);
1852 out_drop_write:
1853 	mnt_drop_write_file(file);
1854 out:
1855 	return ret;
1856 }
1857 
1858 /*
1859  * helper to check if the subvolume references other subvolumes
1860  */
1861 static noinline int may_destroy_subvol(struct btrfs_root *root)
1862 {
1863 	struct btrfs_fs_info *fs_info = root->fs_info;
1864 	struct btrfs_path *path;
1865 	struct btrfs_dir_item *di;
1866 	struct btrfs_key key;
1867 	u64 dir_id;
1868 	int ret;
1869 
1870 	path = btrfs_alloc_path();
1871 	if (!path)
1872 		return -ENOMEM;
1873 
1874 	/* Make sure this root isn't set as the default subvol */
1875 	dir_id = btrfs_super_root_dir(fs_info->super_copy);
1876 	di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
1877 				   dir_id, "default", 7, 0);
1878 	if (di && !IS_ERR(di)) {
1879 		btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1880 		if (key.objectid == root->root_key.objectid) {
1881 			ret = -EPERM;
1882 			btrfs_err(fs_info,
1883 				  "deleting default subvolume %llu is not allowed",
1884 				  key.objectid);
1885 			goto out;
1886 		}
1887 		btrfs_release_path(path);
1888 	}
1889 
1890 	key.objectid = root->root_key.objectid;
1891 	key.type = BTRFS_ROOT_REF_KEY;
1892 	key.offset = (u64)-1;
1893 
1894 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1895 	if (ret < 0)
1896 		goto out;
1897 	BUG_ON(ret == 0);
1898 
1899 	ret = 0;
1900 	if (path->slots[0] > 0) {
1901 		path->slots[0]--;
1902 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1903 		if (key.objectid == root->root_key.objectid &&
1904 		    key.type == BTRFS_ROOT_REF_KEY)
1905 			ret = -ENOTEMPTY;
1906 	}
1907 out:
1908 	btrfs_free_path(path);
1909 	return ret;
1910 }
1911 
1912 static noinline int key_in_sk(struct btrfs_key *key,
1913 			      struct btrfs_ioctl_search_key *sk)
1914 {
1915 	struct btrfs_key test;
1916 	int ret;
1917 
1918 	test.objectid = sk->min_objectid;
1919 	test.type = sk->min_type;
1920 	test.offset = sk->min_offset;
1921 
1922 	ret = btrfs_comp_cpu_keys(key, &test);
1923 	if (ret < 0)
1924 		return 0;
1925 
1926 	test.objectid = sk->max_objectid;
1927 	test.type = sk->max_type;
1928 	test.offset = sk->max_offset;
1929 
1930 	ret = btrfs_comp_cpu_keys(key, &test);
1931 	if (ret > 0)
1932 		return 0;
1933 	return 1;
1934 }
1935 
1936 static noinline int copy_to_sk(struct btrfs_path *path,
1937 			       struct btrfs_key *key,
1938 			       struct btrfs_ioctl_search_key *sk,
1939 			       size_t *buf_size,
1940 			       char __user *ubuf,
1941 			       unsigned long *sk_offset,
1942 			       int *num_found)
1943 {
1944 	u64 found_transid;
1945 	struct extent_buffer *leaf;
1946 	struct btrfs_ioctl_search_header sh;
1947 	struct btrfs_key test;
1948 	unsigned long item_off;
1949 	unsigned long item_len;
1950 	int nritems;
1951 	int i;
1952 	int slot;
1953 	int ret = 0;
1954 
1955 	leaf = path->nodes[0];
1956 	slot = path->slots[0];
1957 	nritems = btrfs_header_nritems(leaf);
1958 
1959 	if (btrfs_header_generation(leaf) > sk->max_transid) {
1960 		i = nritems;
1961 		goto advance_key;
1962 	}
1963 	found_transid = btrfs_header_generation(leaf);
1964 
1965 	for (i = slot; i < nritems; i++) {
1966 		item_off = btrfs_item_ptr_offset(leaf, i);
1967 		item_len = btrfs_item_size_nr(leaf, i);
1968 
1969 		btrfs_item_key_to_cpu(leaf, key, i);
1970 		if (!key_in_sk(key, sk))
1971 			continue;
1972 
1973 		if (sizeof(sh) + item_len > *buf_size) {
1974 			if (*num_found) {
1975 				ret = 1;
1976 				goto out;
1977 			}
1978 
1979 			/*
1980 			 * return one empty item back for v1, which does not
1981 			 * handle -EOVERFLOW
1982 			 */
1983 
1984 			*buf_size = sizeof(sh) + item_len;
1985 			item_len = 0;
1986 			ret = -EOVERFLOW;
1987 		}
1988 
1989 		if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
1990 			ret = 1;
1991 			goto out;
1992 		}
1993 
1994 		sh.objectid = key->objectid;
1995 		sh.offset = key->offset;
1996 		sh.type = key->type;
1997 		sh.len = item_len;
1998 		sh.transid = found_transid;
1999 
2000 		/* copy search result header */
2001 		if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
2002 			ret = -EFAULT;
2003 			goto out;
2004 		}
2005 
2006 		*sk_offset += sizeof(sh);
2007 
2008 		if (item_len) {
2009 			char __user *up = ubuf + *sk_offset;
2010 			/* copy the item */
2011 			if (read_extent_buffer_to_user(leaf, up,
2012 						       item_off, item_len)) {
2013 				ret = -EFAULT;
2014 				goto out;
2015 			}
2016 
2017 			*sk_offset += item_len;
2018 		}
2019 		(*num_found)++;
2020 
2021 		if (ret) /* -EOVERFLOW from above */
2022 			goto out;
2023 
2024 		if (*num_found >= sk->nr_items) {
2025 			ret = 1;
2026 			goto out;
2027 		}
2028 	}
2029 advance_key:
2030 	ret = 0;
2031 	test.objectid = sk->max_objectid;
2032 	test.type = sk->max_type;
2033 	test.offset = sk->max_offset;
2034 	if (btrfs_comp_cpu_keys(key, &test) >= 0)
2035 		ret = 1;
2036 	else if (key->offset < (u64)-1)
2037 		key->offset++;
2038 	else if (key->type < (u8)-1) {
2039 		key->offset = 0;
2040 		key->type++;
2041 	} else if (key->objectid < (u64)-1) {
2042 		key->offset = 0;
2043 		key->type = 0;
2044 		key->objectid++;
2045 	} else
2046 		ret = 1;
2047 out:
2048 	/*
2049 	 *  0: all items from this leaf copied, continue with next
2050 	 *  1: * more items can be copied, but unused buffer is too small
2051 	 *     * all items were found
2052 	 *     Either way, it will stops the loop which iterates to the next
2053 	 *     leaf
2054 	 *  -EOVERFLOW: item was to large for buffer
2055 	 *  -EFAULT: could not copy extent buffer back to userspace
2056 	 */
2057 	return ret;
2058 }
2059 
2060 static noinline int search_ioctl(struct inode *inode,
2061 				 struct btrfs_ioctl_search_key *sk,
2062 				 size_t *buf_size,
2063 				 char __user *ubuf)
2064 {
2065 	struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2066 	struct btrfs_root *root;
2067 	struct btrfs_key key;
2068 	struct btrfs_path *path;
2069 	int ret;
2070 	int num_found = 0;
2071 	unsigned long sk_offset = 0;
2072 
2073 	if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2074 		*buf_size = sizeof(struct btrfs_ioctl_search_header);
2075 		return -EOVERFLOW;
2076 	}
2077 
2078 	path = btrfs_alloc_path();
2079 	if (!path)
2080 		return -ENOMEM;
2081 
2082 	if (sk->tree_id == 0) {
2083 		/* search the root of the inode that was passed */
2084 		root = BTRFS_I(inode)->root;
2085 	} else {
2086 		key.objectid = sk->tree_id;
2087 		key.type = BTRFS_ROOT_ITEM_KEY;
2088 		key.offset = (u64)-1;
2089 		root = btrfs_read_fs_root_no_name(info, &key);
2090 		if (IS_ERR(root)) {
2091 			btrfs_free_path(path);
2092 			return -ENOENT;
2093 		}
2094 	}
2095 
2096 	key.objectid = sk->min_objectid;
2097 	key.type = sk->min_type;
2098 	key.offset = sk->min_offset;
2099 
2100 	while (1) {
2101 		ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2102 		if (ret != 0) {
2103 			if (ret > 0)
2104 				ret = 0;
2105 			goto err;
2106 		}
2107 		ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2108 				 &sk_offset, &num_found);
2109 		btrfs_release_path(path);
2110 		if (ret)
2111 			break;
2112 
2113 	}
2114 	if (ret > 0)
2115 		ret = 0;
2116 err:
2117 	sk->nr_items = num_found;
2118 	btrfs_free_path(path);
2119 	return ret;
2120 }
2121 
2122 static noinline int btrfs_ioctl_tree_search(struct file *file,
2123 					   void __user *argp)
2124 {
2125 	struct btrfs_ioctl_search_args __user *uargs;
2126 	struct btrfs_ioctl_search_key sk;
2127 	struct inode *inode;
2128 	int ret;
2129 	size_t buf_size;
2130 
2131 	if (!capable(CAP_SYS_ADMIN))
2132 		return -EPERM;
2133 
2134 	uargs = (struct btrfs_ioctl_search_args __user *)argp;
2135 
2136 	if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2137 		return -EFAULT;
2138 
2139 	buf_size = sizeof(uargs->buf);
2140 
2141 	inode = file_inode(file);
2142 	ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2143 
2144 	/*
2145 	 * In the origin implementation an overflow is handled by returning a
2146 	 * search header with a len of zero, so reset ret.
2147 	 */
2148 	if (ret == -EOVERFLOW)
2149 		ret = 0;
2150 
2151 	if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2152 		ret = -EFAULT;
2153 	return ret;
2154 }
2155 
2156 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2157 					       void __user *argp)
2158 {
2159 	struct btrfs_ioctl_search_args_v2 __user *uarg;
2160 	struct btrfs_ioctl_search_args_v2 args;
2161 	struct inode *inode;
2162 	int ret;
2163 	size_t buf_size;
2164 	const size_t buf_limit = SZ_16M;
2165 
2166 	if (!capable(CAP_SYS_ADMIN))
2167 		return -EPERM;
2168 
2169 	/* copy search header and buffer size */
2170 	uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2171 	if (copy_from_user(&args, uarg, sizeof(args)))
2172 		return -EFAULT;
2173 
2174 	buf_size = args.buf_size;
2175 
2176 	/* limit result size to 16MB */
2177 	if (buf_size > buf_limit)
2178 		buf_size = buf_limit;
2179 
2180 	inode = file_inode(file);
2181 	ret = search_ioctl(inode, &args.key, &buf_size,
2182 			   (char *)(&uarg->buf[0]));
2183 	if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2184 		ret = -EFAULT;
2185 	else if (ret == -EOVERFLOW &&
2186 		copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2187 		ret = -EFAULT;
2188 
2189 	return ret;
2190 }
2191 
2192 /*
2193  * Search INODE_REFs to identify path name of 'dirid' directory
2194  * in a 'tree_id' tree. and sets path name to 'name'.
2195  */
2196 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2197 				u64 tree_id, u64 dirid, char *name)
2198 {
2199 	struct btrfs_root *root;
2200 	struct btrfs_key key;
2201 	char *ptr;
2202 	int ret = -1;
2203 	int slot;
2204 	int len;
2205 	int total_len = 0;
2206 	struct btrfs_inode_ref *iref;
2207 	struct extent_buffer *l;
2208 	struct btrfs_path *path;
2209 
2210 	if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2211 		name[0]='\0';
2212 		return 0;
2213 	}
2214 
2215 	path = btrfs_alloc_path();
2216 	if (!path)
2217 		return -ENOMEM;
2218 
2219 	ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2220 
2221 	key.objectid = tree_id;
2222 	key.type = BTRFS_ROOT_ITEM_KEY;
2223 	key.offset = (u64)-1;
2224 	root = btrfs_read_fs_root_no_name(info, &key);
2225 	if (IS_ERR(root)) {
2226 		btrfs_err(info, "could not find root %llu", tree_id);
2227 		ret = -ENOENT;
2228 		goto out;
2229 	}
2230 
2231 	key.objectid = dirid;
2232 	key.type = BTRFS_INODE_REF_KEY;
2233 	key.offset = (u64)-1;
2234 
2235 	while (1) {
2236 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2237 		if (ret < 0)
2238 			goto out;
2239 		else if (ret > 0) {
2240 			ret = btrfs_previous_item(root, path, dirid,
2241 						  BTRFS_INODE_REF_KEY);
2242 			if (ret < 0)
2243 				goto out;
2244 			else if (ret > 0) {
2245 				ret = -ENOENT;
2246 				goto out;
2247 			}
2248 		}
2249 
2250 		l = path->nodes[0];
2251 		slot = path->slots[0];
2252 		btrfs_item_key_to_cpu(l, &key, slot);
2253 
2254 		iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2255 		len = btrfs_inode_ref_name_len(l, iref);
2256 		ptr -= len + 1;
2257 		total_len += len + 1;
2258 		if (ptr < name) {
2259 			ret = -ENAMETOOLONG;
2260 			goto out;
2261 		}
2262 
2263 		*(ptr + len) = '/';
2264 		read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2265 
2266 		if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2267 			break;
2268 
2269 		btrfs_release_path(path);
2270 		key.objectid = key.offset;
2271 		key.offset = (u64)-1;
2272 		dirid = key.objectid;
2273 	}
2274 	memmove(name, ptr, total_len);
2275 	name[total_len] = '\0';
2276 	ret = 0;
2277 out:
2278 	btrfs_free_path(path);
2279 	return ret;
2280 }
2281 
2282 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2283 					   void __user *argp)
2284 {
2285 	 struct btrfs_ioctl_ino_lookup_args *args;
2286 	 struct inode *inode;
2287 	int ret = 0;
2288 
2289 	args = memdup_user(argp, sizeof(*args));
2290 	if (IS_ERR(args))
2291 		return PTR_ERR(args);
2292 
2293 	inode = file_inode(file);
2294 
2295 	/*
2296 	 * Unprivileged query to obtain the containing subvolume root id. The
2297 	 * path is reset so it's consistent with btrfs_search_path_in_tree.
2298 	 */
2299 	if (args->treeid == 0)
2300 		args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2301 
2302 	if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2303 		args->name[0] = 0;
2304 		goto out;
2305 	}
2306 
2307 	if (!capable(CAP_SYS_ADMIN)) {
2308 		ret = -EPERM;
2309 		goto out;
2310 	}
2311 
2312 	ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2313 					args->treeid, args->objectid,
2314 					args->name);
2315 
2316 out:
2317 	if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2318 		ret = -EFAULT;
2319 
2320 	kfree(args);
2321 	return ret;
2322 }
2323 
2324 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2325 					     void __user *arg)
2326 {
2327 	struct dentry *parent = file->f_path.dentry;
2328 	struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2329 	struct dentry *dentry;
2330 	struct inode *dir = d_inode(parent);
2331 	struct inode *inode;
2332 	struct btrfs_root *root = BTRFS_I(dir)->root;
2333 	struct btrfs_root *dest = NULL;
2334 	struct btrfs_ioctl_vol_args *vol_args;
2335 	struct btrfs_trans_handle *trans;
2336 	struct btrfs_block_rsv block_rsv;
2337 	u64 root_flags;
2338 	u64 qgroup_reserved;
2339 	int namelen;
2340 	int ret;
2341 	int err = 0;
2342 
2343 	if (!S_ISDIR(dir->i_mode))
2344 		return -ENOTDIR;
2345 
2346 	vol_args = memdup_user(arg, sizeof(*vol_args));
2347 	if (IS_ERR(vol_args))
2348 		return PTR_ERR(vol_args);
2349 
2350 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2351 	namelen = strlen(vol_args->name);
2352 	if (strchr(vol_args->name, '/') ||
2353 	    strncmp(vol_args->name, "..", namelen) == 0) {
2354 		err = -EINVAL;
2355 		goto out;
2356 	}
2357 
2358 	err = mnt_want_write_file(file);
2359 	if (err)
2360 		goto out;
2361 
2362 
2363 	err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2364 	if (err == -EINTR)
2365 		goto out_drop_write;
2366 	dentry = lookup_one_len(vol_args->name, parent, namelen);
2367 	if (IS_ERR(dentry)) {
2368 		err = PTR_ERR(dentry);
2369 		goto out_unlock_dir;
2370 	}
2371 
2372 	if (d_really_is_negative(dentry)) {
2373 		err = -ENOENT;
2374 		goto out_dput;
2375 	}
2376 
2377 	inode = d_inode(dentry);
2378 	dest = BTRFS_I(inode)->root;
2379 	if (!capable(CAP_SYS_ADMIN)) {
2380 		/*
2381 		 * Regular user.  Only allow this with a special mount
2382 		 * option, when the user has write+exec access to the
2383 		 * subvol root, and when rmdir(2) would have been
2384 		 * allowed.
2385 		 *
2386 		 * Note that this is _not_ check that the subvol is
2387 		 * empty or doesn't contain data that we wouldn't
2388 		 * otherwise be able to delete.
2389 		 *
2390 		 * Users who want to delete empty subvols should try
2391 		 * rmdir(2).
2392 		 */
2393 		err = -EPERM;
2394 		if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2395 			goto out_dput;
2396 
2397 		/*
2398 		 * Do not allow deletion if the parent dir is the same
2399 		 * as the dir to be deleted.  That means the ioctl
2400 		 * must be called on the dentry referencing the root
2401 		 * of the subvol, not a random directory contained
2402 		 * within it.
2403 		 */
2404 		err = -EINVAL;
2405 		if (root == dest)
2406 			goto out_dput;
2407 
2408 		err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2409 		if (err)
2410 			goto out_dput;
2411 	}
2412 
2413 	/* check if subvolume may be deleted by a user */
2414 	err = btrfs_may_delete(dir, dentry, 1);
2415 	if (err)
2416 		goto out_dput;
2417 
2418 	if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2419 		err = -EINVAL;
2420 		goto out_dput;
2421 	}
2422 
2423 	inode_lock(inode);
2424 
2425 	/*
2426 	 * Don't allow to delete a subvolume with send in progress. This is
2427 	 * inside the i_mutex so the error handling that has to drop the bit
2428 	 * again is not run concurrently.
2429 	 */
2430 	spin_lock(&dest->root_item_lock);
2431 	root_flags = btrfs_root_flags(&dest->root_item);
2432 	if (dest->send_in_progress == 0) {
2433 		btrfs_set_root_flags(&dest->root_item,
2434 				root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2435 		spin_unlock(&dest->root_item_lock);
2436 	} else {
2437 		spin_unlock(&dest->root_item_lock);
2438 		btrfs_warn(fs_info,
2439 			   "Attempt to delete subvolume %llu during send",
2440 			   dest->root_key.objectid);
2441 		err = -EPERM;
2442 		goto out_unlock_inode;
2443 	}
2444 
2445 	down_write(&fs_info->subvol_sem);
2446 
2447 	err = may_destroy_subvol(dest);
2448 	if (err)
2449 		goto out_up_write;
2450 
2451 	btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2452 	/*
2453 	 * One for dir inode, two for dir entries, two for root
2454 	 * ref/backref.
2455 	 */
2456 	err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2457 					       5, &qgroup_reserved, true);
2458 	if (err)
2459 		goto out_up_write;
2460 
2461 	trans = btrfs_start_transaction(root, 0);
2462 	if (IS_ERR(trans)) {
2463 		err = PTR_ERR(trans);
2464 		goto out_release;
2465 	}
2466 	trans->block_rsv = &block_rsv;
2467 	trans->bytes_reserved = block_rsv.size;
2468 
2469 	btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
2470 
2471 	ret = btrfs_unlink_subvol(trans, root, dir,
2472 				dest->root_key.objectid,
2473 				dentry->d_name.name,
2474 				dentry->d_name.len);
2475 	if (ret) {
2476 		err = ret;
2477 		btrfs_abort_transaction(trans, ret);
2478 		goto out_end_trans;
2479 	}
2480 
2481 	btrfs_record_root_in_trans(trans, dest);
2482 
2483 	memset(&dest->root_item.drop_progress, 0,
2484 		sizeof(dest->root_item.drop_progress));
2485 	dest->root_item.drop_level = 0;
2486 	btrfs_set_root_refs(&dest->root_item, 0);
2487 
2488 	if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2489 		ret = btrfs_insert_orphan_item(trans,
2490 					fs_info->tree_root,
2491 					dest->root_key.objectid);
2492 		if (ret) {
2493 			btrfs_abort_transaction(trans, ret);
2494 			err = ret;
2495 			goto out_end_trans;
2496 		}
2497 	}
2498 
2499 	ret = btrfs_uuid_tree_rem(trans, fs_info, dest->root_item.uuid,
2500 				  BTRFS_UUID_KEY_SUBVOL,
2501 				  dest->root_key.objectid);
2502 	if (ret && ret != -ENOENT) {
2503 		btrfs_abort_transaction(trans, ret);
2504 		err = ret;
2505 		goto out_end_trans;
2506 	}
2507 	if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2508 		ret = btrfs_uuid_tree_rem(trans, fs_info,
2509 					  dest->root_item.received_uuid,
2510 					  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2511 					  dest->root_key.objectid);
2512 		if (ret && ret != -ENOENT) {
2513 			btrfs_abort_transaction(trans, ret);
2514 			err = ret;
2515 			goto out_end_trans;
2516 		}
2517 	}
2518 
2519 out_end_trans:
2520 	trans->block_rsv = NULL;
2521 	trans->bytes_reserved = 0;
2522 	ret = btrfs_end_transaction(trans);
2523 	if (ret && !err)
2524 		err = ret;
2525 	inode->i_flags |= S_DEAD;
2526 out_release:
2527 	btrfs_subvolume_release_metadata(fs_info, &block_rsv);
2528 out_up_write:
2529 	up_write(&fs_info->subvol_sem);
2530 	if (err) {
2531 		spin_lock(&dest->root_item_lock);
2532 		root_flags = btrfs_root_flags(&dest->root_item);
2533 		btrfs_set_root_flags(&dest->root_item,
2534 				root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2535 		spin_unlock(&dest->root_item_lock);
2536 	}
2537 out_unlock_inode:
2538 	inode_unlock(inode);
2539 	if (!err) {
2540 		d_invalidate(dentry);
2541 		btrfs_invalidate_inodes(dest);
2542 		d_delete(dentry);
2543 		ASSERT(dest->send_in_progress == 0);
2544 
2545 		/* the last ref */
2546 		if (dest->ino_cache_inode) {
2547 			iput(dest->ino_cache_inode);
2548 			dest->ino_cache_inode = NULL;
2549 		}
2550 	}
2551 out_dput:
2552 	dput(dentry);
2553 out_unlock_dir:
2554 	inode_unlock(dir);
2555 out_drop_write:
2556 	mnt_drop_write_file(file);
2557 out:
2558 	kfree(vol_args);
2559 	return err;
2560 }
2561 
2562 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2563 {
2564 	struct inode *inode = file_inode(file);
2565 	struct btrfs_root *root = BTRFS_I(inode)->root;
2566 	struct btrfs_ioctl_defrag_range_args *range;
2567 	int ret;
2568 
2569 	ret = mnt_want_write_file(file);
2570 	if (ret)
2571 		return ret;
2572 
2573 	if (btrfs_root_readonly(root)) {
2574 		ret = -EROFS;
2575 		goto out;
2576 	}
2577 
2578 	switch (inode->i_mode & S_IFMT) {
2579 	case S_IFDIR:
2580 		if (!capable(CAP_SYS_ADMIN)) {
2581 			ret = -EPERM;
2582 			goto out;
2583 		}
2584 		ret = btrfs_defrag_root(root);
2585 		break;
2586 	case S_IFREG:
2587 		if (!(file->f_mode & FMODE_WRITE)) {
2588 			ret = -EINVAL;
2589 			goto out;
2590 		}
2591 
2592 		range = kzalloc(sizeof(*range), GFP_KERNEL);
2593 		if (!range) {
2594 			ret = -ENOMEM;
2595 			goto out;
2596 		}
2597 
2598 		if (argp) {
2599 			if (copy_from_user(range, argp,
2600 					   sizeof(*range))) {
2601 				ret = -EFAULT;
2602 				kfree(range);
2603 				goto out;
2604 			}
2605 			/* compression requires us to start the IO */
2606 			if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2607 				range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2608 				range->extent_thresh = (u32)-1;
2609 			}
2610 		} else {
2611 			/* the rest are all set to zero by kzalloc */
2612 			range->len = (u64)-1;
2613 		}
2614 		ret = btrfs_defrag_file(file_inode(file), file,
2615 					range, 0, 0);
2616 		if (ret > 0)
2617 			ret = 0;
2618 		kfree(range);
2619 		break;
2620 	default:
2621 		ret = -EINVAL;
2622 	}
2623 out:
2624 	mnt_drop_write_file(file);
2625 	return ret;
2626 }
2627 
2628 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
2629 {
2630 	struct btrfs_ioctl_vol_args *vol_args;
2631 	int ret;
2632 
2633 	if (!capable(CAP_SYS_ADMIN))
2634 		return -EPERM;
2635 
2636 	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
2637 		return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2638 
2639 	mutex_lock(&fs_info->volume_mutex);
2640 	vol_args = memdup_user(arg, sizeof(*vol_args));
2641 	if (IS_ERR(vol_args)) {
2642 		ret = PTR_ERR(vol_args);
2643 		goto out;
2644 	}
2645 
2646 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2647 	ret = btrfs_init_new_device(fs_info, vol_args->name);
2648 
2649 	if (!ret)
2650 		btrfs_info(fs_info, "disk added %s", vol_args->name);
2651 
2652 	kfree(vol_args);
2653 out:
2654 	mutex_unlock(&fs_info->volume_mutex);
2655 	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2656 	return ret;
2657 }
2658 
2659 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
2660 {
2661 	struct inode *inode = file_inode(file);
2662 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2663 	struct btrfs_ioctl_vol_args_v2 *vol_args;
2664 	int ret;
2665 
2666 	if (!capable(CAP_SYS_ADMIN))
2667 		return -EPERM;
2668 
2669 	ret = mnt_want_write_file(file);
2670 	if (ret)
2671 		return ret;
2672 
2673 	vol_args = memdup_user(arg, sizeof(*vol_args));
2674 	if (IS_ERR(vol_args)) {
2675 		ret = PTR_ERR(vol_args);
2676 		goto err_drop;
2677 	}
2678 
2679 	/* Check for compatibility reject unknown flags */
2680 	if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED)
2681 		return -EOPNOTSUPP;
2682 
2683 	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
2684 		ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2685 		goto out;
2686 	}
2687 
2688 	mutex_lock(&fs_info->volume_mutex);
2689 	if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
2690 		ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
2691 	} else {
2692 		vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2693 		ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2694 	}
2695 	mutex_unlock(&fs_info->volume_mutex);
2696 	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2697 
2698 	if (!ret) {
2699 		if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
2700 			btrfs_info(fs_info, "device deleted: id %llu",
2701 					vol_args->devid);
2702 		else
2703 			btrfs_info(fs_info, "device deleted: %s",
2704 					vol_args->name);
2705 	}
2706 out:
2707 	kfree(vol_args);
2708 err_drop:
2709 	mnt_drop_write_file(file);
2710 	return ret;
2711 }
2712 
2713 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2714 {
2715 	struct inode *inode = file_inode(file);
2716 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2717 	struct btrfs_ioctl_vol_args *vol_args;
2718 	int ret;
2719 
2720 	if (!capable(CAP_SYS_ADMIN))
2721 		return -EPERM;
2722 
2723 	ret = mnt_want_write_file(file);
2724 	if (ret)
2725 		return ret;
2726 
2727 	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
2728 		ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2729 		goto out_drop_write;
2730 	}
2731 
2732 	vol_args = memdup_user(arg, sizeof(*vol_args));
2733 	if (IS_ERR(vol_args)) {
2734 		ret = PTR_ERR(vol_args);
2735 		goto out;
2736 	}
2737 
2738 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2739 	mutex_lock(&fs_info->volume_mutex);
2740 	ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2741 	mutex_unlock(&fs_info->volume_mutex);
2742 
2743 	if (!ret)
2744 		btrfs_info(fs_info, "disk deleted %s", vol_args->name);
2745 	kfree(vol_args);
2746 out:
2747 	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2748 out_drop_write:
2749 	mnt_drop_write_file(file);
2750 
2751 	return ret;
2752 }
2753 
2754 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
2755 				void __user *arg)
2756 {
2757 	struct btrfs_ioctl_fs_info_args *fi_args;
2758 	struct btrfs_device *device;
2759 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2760 	int ret = 0;
2761 
2762 	fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2763 	if (!fi_args)
2764 		return -ENOMEM;
2765 
2766 	mutex_lock(&fs_devices->device_list_mutex);
2767 	fi_args->num_devices = fs_devices->num_devices;
2768 	memcpy(&fi_args->fsid, fs_info->fsid, sizeof(fi_args->fsid));
2769 
2770 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
2771 		if (device->devid > fi_args->max_id)
2772 			fi_args->max_id = device->devid;
2773 	}
2774 	mutex_unlock(&fs_devices->device_list_mutex);
2775 
2776 	fi_args->nodesize = fs_info->nodesize;
2777 	fi_args->sectorsize = fs_info->sectorsize;
2778 	fi_args->clone_alignment = fs_info->sectorsize;
2779 
2780 	if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2781 		ret = -EFAULT;
2782 
2783 	kfree(fi_args);
2784 	return ret;
2785 }
2786 
2787 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
2788 				 void __user *arg)
2789 {
2790 	struct btrfs_ioctl_dev_info_args *di_args;
2791 	struct btrfs_device *dev;
2792 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2793 	int ret = 0;
2794 	char *s_uuid = NULL;
2795 
2796 	di_args = memdup_user(arg, sizeof(*di_args));
2797 	if (IS_ERR(di_args))
2798 		return PTR_ERR(di_args);
2799 
2800 	if (!btrfs_is_empty_uuid(di_args->uuid))
2801 		s_uuid = di_args->uuid;
2802 
2803 	mutex_lock(&fs_devices->device_list_mutex);
2804 	dev = btrfs_find_device(fs_info, di_args->devid, s_uuid, NULL);
2805 
2806 	if (!dev) {
2807 		ret = -ENODEV;
2808 		goto out;
2809 	}
2810 
2811 	di_args->devid = dev->devid;
2812 	di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2813 	di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2814 	memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2815 	if (dev->name) {
2816 		struct rcu_string *name;
2817 
2818 		rcu_read_lock();
2819 		name = rcu_dereference(dev->name);
2820 		strncpy(di_args->path, name->str, sizeof(di_args->path));
2821 		rcu_read_unlock();
2822 		di_args->path[sizeof(di_args->path) - 1] = 0;
2823 	} else {
2824 		di_args->path[0] = '\0';
2825 	}
2826 
2827 out:
2828 	mutex_unlock(&fs_devices->device_list_mutex);
2829 	if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2830 		ret = -EFAULT;
2831 
2832 	kfree(di_args);
2833 	return ret;
2834 }
2835 
2836 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
2837 {
2838 	struct page *page;
2839 
2840 	page = grab_cache_page(inode->i_mapping, index);
2841 	if (!page)
2842 		return ERR_PTR(-ENOMEM);
2843 
2844 	if (!PageUptodate(page)) {
2845 		int ret;
2846 
2847 		ret = btrfs_readpage(NULL, page);
2848 		if (ret)
2849 			return ERR_PTR(ret);
2850 		lock_page(page);
2851 		if (!PageUptodate(page)) {
2852 			unlock_page(page);
2853 			put_page(page);
2854 			return ERR_PTR(-EIO);
2855 		}
2856 		if (page->mapping != inode->i_mapping) {
2857 			unlock_page(page);
2858 			put_page(page);
2859 			return ERR_PTR(-EAGAIN);
2860 		}
2861 	}
2862 
2863 	return page;
2864 }
2865 
2866 static int gather_extent_pages(struct inode *inode, struct page **pages,
2867 			       int num_pages, u64 off)
2868 {
2869 	int i;
2870 	pgoff_t index = off >> PAGE_SHIFT;
2871 
2872 	for (i = 0; i < num_pages; i++) {
2873 again:
2874 		pages[i] = extent_same_get_page(inode, index + i);
2875 		if (IS_ERR(pages[i])) {
2876 			int err = PTR_ERR(pages[i]);
2877 
2878 			if (err == -EAGAIN)
2879 				goto again;
2880 			pages[i] = NULL;
2881 			return err;
2882 		}
2883 	}
2884 	return 0;
2885 }
2886 
2887 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
2888 			     bool retry_range_locking)
2889 {
2890 	/*
2891 	 * Do any pending delalloc/csum calculations on inode, one way or
2892 	 * another, and lock file content.
2893 	 * The locking order is:
2894 	 *
2895 	 *   1) pages
2896 	 *   2) range in the inode's io tree
2897 	 */
2898 	while (1) {
2899 		struct btrfs_ordered_extent *ordered;
2900 		lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2901 		ordered = btrfs_lookup_first_ordered_extent(inode,
2902 							    off + len - 1);
2903 		if ((!ordered ||
2904 		     ordered->file_offset + ordered->len <= off ||
2905 		     ordered->file_offset >= off + len) &&
2906 		    !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2907 				    off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2908 			if (ordered)
2909 				btrfs_put_ordered_extent(ordered);
2910 			break;
2911 		}
2912 		unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2913 		if (ordered)
2914 			btrfs_put_ordered_extent(ordered);
2915 		if (!retry_range_locking)
2916 			return -EAGAIN;
2917 		btrfs_wait_ordered_range(inode, off, len);
2918 	}
2919 	return 0;
2920 }
2921 
2922 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
2923 {
2924 	inode_unlock(inode1);
2925 	inode_unlock(inode2);
2926 }
2927 
2928 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
2929 {
2930 	if (inode1 < inode2)
2931 		swap(inode1, inode2);
2932 
2933 	inode_lock_nested(inode1, I_MUTEX_PARENT);
2934 	inode_lock_nested(inode2, I_MUTEX_CHILD);
2935 }
2936 
2937 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
2938 				      struct inode *inode2, u64 loff2, u64 len)
2939 {
2940 	unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2941 	unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2942 }
2943 
2944 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
2945 				    struct inode *inode2, u64 loff2, u64 len,
2946 				    bool retry_range_locking)
2947 {
2948 	int ret;
2949 
2950 	if (inode1 < inode2) {
2951 		swap(inode1, inode2);
2952 		swap(loff1, loff2);
2953 	}
2954 	ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
2955 	if (ret)
2956 		return ret;
2957 	ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
2958 	if (ret)
2959 		unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
2960 			      loff1 + len - 1);
2961 	return ret;
2962 }
2963 
2964 struct cmp_pages {
2965 	int		num_pages;
2966 	struct page	**src_pages;
2967 	struct page	**dst_pages;
2968 };
2969 
2970 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
2971 {
2972 	int i;
2973 	struct page *pg;
2974 
2975 	for (i = 0; i < cmp->num_pages; i++) {
2976 		pg = cmp->src_pages[i];
2977 		if (pg) {
2978 			unlock_page(pg);
2979 			put_page(pg);
2980 		}
2981 		pg = cmp->dst_pages[i];
2982 		if (pg) {
2983 			unlock_page(pg);
2984 			put_page(pg);
2985 		}
2986 	}
2987 	kfree(cmp->src_pages);
2988 	kfree(cmp->dst_pages);
2989 }
2990 
2991 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
2992 				  struct inode *dst, u64 dst_loff,
2993 				  u64 len, struct cmp_pages *cmp)
2994 {
2995 	int ret;
2996 	int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT;
2997 	struct page **src_pgarr, **dst_pgarr;
2998 
2999 	/*
3000 	 * We must gather up all the pages before we initiate our
3001 	 * extent locking. We use an array for the page pointers. Size
3002 	 * of the array is bounded by len, which is in turn bounded by
3003 	 * BTRFS_MAX_DEDUPE_LEN.
3004 	 */
3005 	src_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
3006 	dst_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
3007 	if (!src_pgarr || !dst_pgarr) {
3008 		kfree(src_pgarr);
3009 		kfree(dst_pgarr);
3010 		return -ENOMEM;
3011 	}
3012 	cmp->num_pages = num_pages;
3013 	cmp->src_pages = src_pgarr;
3014 	cmp->dst_pages = dst_pgarr;
3015 
3016 	/*
3017 	 * If deduping ranges in the same inode, locking rules make it mandatory
3018 	 * to always lock pages in ascending order to avoid deadlocks with
3019 	 * concurrent tasks (such as starting writeback/delalloc).
3020 	 */
3021 	if (src == dst && dst_loff < loff) {
3022 		swap(src_pgarr, dst_pgarr);
3023 		swap(loff, dst_loff);
3024 	}
3025 
3026 	ret = gather_extent_pages(src, src_pgarr, cmp->num_pages, loff);
3027 	if (ret)
3028 		goto out;
3029 
3030 	ret = gather_extent_pages(dst, dst_pgarr, cmp->num_pages, dst_loff);
3031 
3032 out:
3033 	if (ret)
3034 		btrfs_cmp_data_free(cmp);
3035 	return ret;
3036 }
3037 
3038 static int btrfs_cmp_data(u64 len, struct cmp_pages *cmp)
3039 {
3040 	int ret = 0;
3041 	int i;
3042 	struct page *src_page, *dst_page;
3043 	unsigned int cmp_len = PAGE_SIZE;
3044 	void *addr, *dst_addr;
3045 
3046 	i = 0;
3047 	while (len) {
3048 		if (len < PAGE_SIZE)
3049 			cmp_len = len;
3050 
3051 		BUG_ON(i >= cmp->num_pages);
3052 
3053 		src_page = cmp->src_pages[i];
3054 		dst_page = cmp->dst_pages[i];
3055 		ASSERT(PageLocked(src_page));
3056 		ASSERT(PageLocked(dst_page));
3057 
3058 		addr = kmap_atomic(src_page);
3059 		dst_addr = kmap_atomic(dst_page);
3060 
3061 		flush_dcache_page(src_page);
3062 		flush_dcache_page(dst_page);
3063 
3064 		if (memcmp(addr, dst_addr, cmp_len))
3065 			ret = -EBADE;
3066 
3067 		kunmap_atomic(addr);
3068 		kunmap_atomic(dst_addr);
3069 
3070 		if (ret)
3071 			break;
3072 
3073 		len -= cmp_len;
3074 		i++;
3075 	}
3076 
3077 	return ret;
3078 }
3079 
3080 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3081 				     u64 olen)
3082 {
3083 	u64 len = *plen;
3084 	u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3085 
3086 	if (off + olen > inode->i_size || off + olen < off)
3087 		return -EINVAL;
3088 
3089 	/* if we extend to eof, continue to block boundary */
3090 	if (off + len == inode->i_size)
3091 		*plen = len = ALIGN(inode->i_size, bs) - off;
3092 
3093 	/* Check that we are block aligned - btrfs_clone() requires this */
3094 	if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3095 		return -EINVAL;
3096 
3097 	return 0;
3098 }
3099 
3100 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3101 			     struct inode *dst, u64 dst_loff)
3102 {
3103 	int ret;
3104 	u64 len = olen;
3105 	struct cmp_pages cmp;
3106 	bool same_inode = (src == dst);
3107 	u64 same_lock_start = 0;
3108 	u64 same_lock_len = 0;
3109 
3110 	if (len == 0)
3111 		return 0;
3112 
3113 	if (same_inode)
3114 		inode_lock(src);
3115 	else
3116 		btrfs_double_inode_lock(src, dst);
3117 
3118 	ret = extent_same_check_offsets(src, loff, &len, olen);
3119 	if (ret)
3120 		goto out_unlock;
3121 
3122 	ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3123 	if (ret)
3124 		goto out_unlock;
3125 
3126 	if (same_inode) {
3127 		/*
3128 		 * Single inode case wants the same checks, except we
3129 		 * don't want our length pushed out past i_size as
3130 		 * comparing that data range makes no sense.
3131 		 *
3132 		 * extent_same_check_offsets() will do this for an
3133 		 * unaligned length at i_size, so catch it here and
3134 		 * reject the request.
3135 		 *
3136 		 * This effectively means we require aligned extents
3137 		 * for the single-inode case, whereas the other cases
3138 		 * allow an unaligned length so long as it ends at
3139 		 * i_size.
3140 		 */
3141 		if (len != olen) {
3142 			ret = -EINVAL;
3143 			goto out_unlock;
3144 		}
3145 
3146 		/* Check for overlapping ranges */
3147 		if (dst_loff + len > loff && dst_loff < loff + len) {
3148 			ret = -EINVAL;
3149 			goto out_unlock;
3150 		}
3151 
3152 		same_lock_start = min_t(u64, loff, dst_loff);
3153 		same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3154 	}
3155 
3156 	/* don't make the dst file partly checksummed */
3157 	if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3158 	    (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3159 		ret = -EINVAL;
3160 		goto out_unlock;
3161 	}
3162 
3163 again:
3164 	ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp);
3165 	if (ret)
3166 		goto out_unlock;
3167 
3168 	if (same_inode)
3169 		ret = lock_extent_range(src, same_lock_start, same_lock_len,
3170 					false);
3171 	else
3172 		ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3173 					       false);
3174 	/*
3175 	 * If one of the inodes has dirty pages in the respective range or
3176 	 * ordered extents, we need to flush dellaloc and wait for all ordered
3177 	 * extents in the range. We must unlock the pages and the ranges in the
3178 	 * io trees to avoid deadlocks when flushing delalloc (requires locking
3179 	 * pages) and when waiting for ordered extents to complete (they require
3180 	 * range locking).
3181 	 */
3182 	if (ret == -EAGAIN) {
3183 		/*
3184 		 * Ranges in the io trees already unlocked. Now unlock all
3185 		 * pages before waiting for all IO to complete.
3186 		 */
3187 		btrfs_cmp_data_free(&cmp);
3188 		if (same_inode) {
3189 			btrfs_wait_ordered_range(src, same_lock_start,
3190 						 same_lock_len);
3191 		} else {
3192 			btrfs_wait_ordered_range(src, loff, len);
3193 			btrfs_wait_ordered_range(dst, dst_loff, len);
3194 		}
3195 		goto again;
3196 	}
3197 	ASSERT(ret == 0);
3198 	if (WARN_ON(ret)) {
3199 		/* ranges in the io trees already unlocked */
3200 		btrfs_cmp_data_free(&cmp);
3201 		return ret;
3202 	}
3203 
3204 	/* pass original length for comparison so we stay within i_size */
3205 	ret = btrfs_cmp_data(olen, &cmp);
3206 	if (ret == 0)
3207 		ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3208 
3209 	if (same_inode)
3210 		unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3211 			      same_lock_start + same_lock_len - 1);
3212 	else
3213 		btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3214 
3215 	btrfs_cmp_data_free(&cmp);
3216 out_unlock:
3217 	if (same_inode)
3218 		inode_unlock(src);
3219 	else
3220 		btrfs_double_inode_unlock(src, dst);
3221 
3222 	return ret;
3223 }
3224 
3225 #define BTRFS_MAX_DEDUPE_LEN	SZ_16M
3226 
3227 ssize_t btrfs_dedupe_file_range(struct file *src_file, u64 loff, u64 olen,
3228 				struct file *dst_file, u64 dst_loff)
3229 {
3230 	struct inode *src = file_inode(src_file);
3231 	struct inode *dst = file_inode(dst_file);
3232 	u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3233 	ssize_t res;
3234 
3235 	if (olen > BTRFS_MAX_DEDUPE_LEN)
3236 		olen = BTRFS_MAX_DEDUPE_LEN;
3237 
3238 	if (WARN_ON_ONCE(bs < PAGE_SIZE)) {
3239 		/*
3240 		 * Btrfs does not support blocksize < page_size. As a
3241 		 * result, btrfs_cmp_data() won't correctly handle
3242 		 * this situation without an update.
3243 		 */
3244 		return -EINVAL;
3245 	}
3246 
3247 	res = btrfs_extent_same(src, loff, olen, dst, dst_loff);
3248 	if (res)
3249 		return res;
3250 	return olen;
3251 }
3252 
3253 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3254 				     struct inode *inode,
3255 				     u64 endoff,
3256 				     const u64 destoff,
3257 				     const u64 olen,
3258 				     int no_time_update)
3259 {
3260 	struct btrfs_root *root = BTRFS_I(inode)->root;
3261 	int ret;
3262 
3263 	inode_inc_iversion(inode);
3264 	if (!no_time_update)
3265 		inode->i_mtime = inode->i_ctime = current_time(inode);
3266 	/*
3267 	 * We round up to the block size at eof when determining which
3268 	 * extents to clone above, but shouldn't round up the file size.
3269 	 */
3270 	if (endoff > destoff + olen)
3271 		endoff = destoff + olen;
3272 	if (endoff > inode->i_size)
3273 		btrfs_i_size_write(BTRFS_I(inode), endoff);
3274 
3275 	ret = btrfs_update_inode(trans, root, inode);
3276 	if (ret) {
3277 		btrfs_abort_transaction(trans, ret);
3278 		btrfs_end_transaction(trans);
3279 		goto out;
3280 	}
3281 	ret = btrfs_end_transaction(trans);
3282 out:
3283 	return ret;
3284 }
3285 
3286 static void clone_update_extent_map(struct btrfs_inode *inode,
3287 				    const struct btrfs_trans_handle *trans,
3288 				    const struct btrfs_path *path,
3289 				    const u64 hole_offset,
3290 				    const u64 hole_len)
3291 {
3292 	struct extent_map_tree *em_tree = &inode->extent_tree;
3293 	struct extent_map *em;
3294 	int ret;
3295 
3296 	em = alloc_extent_map();
3297 	if (!em) {
3298 		set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3299 		return;
3300 	}
3301 
3302 	if (path) {
3303 		struct btrfs_file_extent_item *fi;
3304 
3305 		fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3306 				    struct btrfs_file_extent_item);
3307 		btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3308 		em->generation = -1;
3309 		if (btrfs_file_extent_type(path->nodes[0], fi) ==
3310 		    BTRFS_FILE_EXTENT_INLINE)
3311 			set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3312 					&inode->runtime_flags);
3313 	} else {
3314 		em->start = hole_offset;
3315 		em->len = hole_len;
3316 		em->ram_bytes = em->len;
3317 		em->orig_start = hole_offset;
3318 		em->block_start = EXTENT_MAP_HOLE;
3319 		em->block_len = 0;
3320 		em->orig_block_len = 0;
3321 		em->compress_type = BTRFS_COMPRESS_NONE;
3322 		em->generation = trans->transid;
3323 	}
3324 
3325 	while (1) {
3326 		write_lock(&em_tree->lock);
3327 		ret = add_extent_mapping(em_tree, em, 1);
3328 		write_unlock(&em_tree->lock);
3329 		if (ret != -EEXIST) {
3330 			free_extent_map(em);
3331 			break;
3332 		}
3333 		btrfs_drop_extent_cache(inode, em->start,
3334 					em->start + em->len - 1, 0);
3335 	}
3336 
3337 	if (ret)
3338 		set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3339 }
3340 
3341 /*
3342  * Make sure we do not end up inserting an inline extent into a file that has
3343  * already other (non-inline) extents. If a file has an inline extent it can
3344  * not have any other extents and the (single) inline extent must start at the
3345  * file offset 0. Failing to respect these rules will lead to file corruption,
3346  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3347  *
3348  * We can have extents that have been already written to disk or we can have
3349  * dirty ranges still in delalloc, in which case the extent maps and items are
3350  * created only when we run delalloc, and the delalloc ranges might fall outside
3351  * the range we are currently locking in the inode's io tree. So we check the
3352  * inode's i_size because of that (i_size updates are done while holding the
3353  * i_mutex, which we are holding here).
3354  * We also check to see if the inode has a size not greater than "datal" but has
3355  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3356  * protected against such concurrent fallocate calls by the i_mutex).
3357  *
3358  * If the file has no extents but a size greater than datal, do not allow the
3359  * copy because we would need turn the inline extent into a non-inline one (even
3360  * with NO_HOLES enabled). If we find our destination inode only has one inline
3361  * extent, just overwrite it with the source inline extent if its size is less
3362  * than the source extent's size, or we could copy the source inline extent's
3363  * data into the destination inode's inline extent if the later is greater then
3364  * the former.
3365  */
3366 static int clone_copy_inline_extent(struct inode *dst,
3367 				    struct btrfs_trans_handle *trans,
3368 				    struct btrfs_path *path,
3369 				    struct btrfs_key *new_key,
3370 				    const u64 drop_start,
3371 				    const u64 datal,
3372 				    const u64 skip,
3373 				    const u64 size,
3374 				    char *inline_data)
3375 {
3376 	struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb);
3377 	struct btrfs_root *root = BTRFS_I(dst)->root;
3378 	const u64 aligned_end = ALIGN(new_key->offset + datal,
3379 				      fs_info->sectorsize);
3380 	int ret;
3381 	struct btrfs_key key;
3382 
3383 	if (new_key->offset > 0)
3384 		return -EOPNOTSUPP;
3385 
3386 	key.objectid = btrfs_ino(BTRFS_I(dst));
3387 	key.type = BTRFS_EXTENT_DATA_KEY;
3388 	key.offset = 0;
3389 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3390 	if (ret < 0) {
3391 		return ret;
3392 	} else if (ret > 0) {
3393 		if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3394 			ret = btrfs_next_leaf(root, path);
3395 			if (ret < 0)
3396 				return ret;
3397 			else if (ret > 0)
3398 				goto copy_inline_extent;
3399 		}
3400 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3401 		if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3402 		    key.type == BTRFS_EXTENT_DATA_KEY) {
3403 			ASSERT(key.offset > 0);
3404 			return -EOPNOTSUPP;
3405 		}
3406 	} else if (i_size_read(dst) <= datal) {
3407 		struct btrfs_file_extent_item *ei;
3408 		u64 ext_len;
3409 
3410 		/*
3411 		 * If the file size is <= datal, make sure there are no other
3412 		 * extents following (can happen do to an fallocate call with
3413 		 * the flag FALLOC_FL_KEEP_SIZE).
3414 		 */
3415 		ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3416 				    struct btrfs_file_extent_item);
3417 		/*
3418 		 * If it's an inline extent, it can not have other extents
3419 		 * following it.
3420 		 */
3421 		if (btrfs_file_extent_type(path->nodes[0], ei) ==
3422 		    BTRFS_FILE_EXTENT_INLINE)
3423 			goto copy_inline_extent;
3424 
3425 		ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3426 		if (ext_len > aligned_end)
3427 			return -EOPNOTSUPP;
3428 
3429 		ret = btrfs_next_item(root, path);
3430 		if (ret < 0) {
3431 			return ret;
3432 		} else if (ret == 0) {
3433 			btrfs_item_key_to_cpu(path->nodes[0], &key,
3434 					      path->slots[0]);
3435 			if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3436 			    key.type == BTRFS_EXTENT_DATA_KEY)
3437 				return -EOPNOTSUPP;
3438 		}
3439 	}
3440 
3441 copy_inline_extent:
3442 	/*
3443 	 * We have no extent items, or we have an extent at offset 0 which may
3444 	 * or may not be inlined. All these cases are dealt the same way.
3445 	 */
3446 	if (i_size_read(dst) > datal) {
3447 		/*
3448 		 * If the destination inode has an inline extent...
3449 		 * This would require copying the data from the source inline
3450 		 * extent into the beginning of the destination's inline extent.
3451 		 * But this is really complex, both extents can be compressed
3452 		 * or just one of them, which would require decompressing and
3453 		 * re-compressing data (which could increase the new compressed
3454 		 * size, not allowing the compressed data to fit anymore in an
3455 		 * inline extent).
3456 		 * So just don't support this case for now (it should be rare,
3457 		 * we are not really saving space when cloning inline extents).
3458 		 */
3459 		return -EOPNOTSUPP;
3460 	}
3461 
3462 	btrfs_release_path(path);
3463 	ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3464 	if (ret)
3465 		return ret;
3466 	ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3467 	if (ret)
3468 		return ret;
3469 
3470 	if (skip) {
3471 		const u32 start = btrfs_file_extent_calc_inline_size(0);
3472 
3473 		memmove(inline_data + start, inline_data + start + skip, datal);
3474 	}
3475 
3476 	write_extent_buffer(path->nodes[0], inline_data,
3477 			    btrfs_item_ptr_offset(path->nodes[0],
3478 						  path->slots[0]),
3479 			    size);
3480 	inode_add_bytes(dst, datal);
3481 
3482 	return 0;
3483 }
3484 
3485 /**
3486  * btrfs_clone() - clone a range from inode file to another
3487  *
3488  * @src: Inode to clone from
3489  * @inode: Inode to clone to
3490  * @off: Offset within source to start clone from
3491  * @olen: Original length, passed by user, of range to clone
3492  * @olen_aligned: Block-aligned value of olen
3493  * @destoff: Offset within @inode to start clone
3494  * @no_time_update: Whether to update mtime/ctime on the target inode
3495  */
3496 static int btrfs_clone(struct inode *src, struct inode *inode,
3497 		       const u64 off, const u64 olen, const u64 olen_aligned,
3498 		       const u64 destoff, int no_time_update)
3499 {
3500 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3501 	struct btrfs_root *root = BTRFS_I(inode)->root;
3502 	struct btrfs_path *path = NULL;
3503 	struct extent_buffer *leaf;
3504 	struct btrfs_trans_handle *trans;
3505 	char *buf = NULL;
3506 	struct btrfs_key key;
3507 	u32 nritems;
3508 	int slot;
3509 	int ret;
3510 	const u64 len = olen_aligned;
3511 	u64 last_dest_end = destoff;
3512 
3513 	ret = -ENOMEM;
3514 	buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
3515 	if (!buf)
3516 		return ret;
3517 
3518 	path = btrfs_alloc_path();
3519 	if (!path) {
3520 		kvfree(buf);
3521 		return ret;
3522 	}
3523 
3524 	path->reada = READA_FORWARD;
3525 	/* clone data */
3526 	key.objectid = btrfs_ino(BTRFS_I(src));
3527 	key.type = BTRFS_EXTENT_DATA_KEY;
3528 	key.offset = off;
3529 
3530 	while (1) {
3531 		u64 next_key_min_offset = key.offset + 1;
3532 
3533 		/*
3534 		 * note the key will change type as we walk through the
3535 		 * tree.
3536 		 */
3537 		path->leave_spinning = 1;
3538 		ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3539 				0, 0);
3540 		if (ret < 0)
3541 			goto out;
3542 		/*
3543 		 * First search, if no extent item that starts at offset off was
3544 		 * found but the previous item is an extent item, it's possible
3545 		 * it might overlap our target range, therefore process it.
3546 		 */
3547 		if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3548 			btrfs_item_key_to_cpu(path->nodes[0], &key,
3549 					      path->slots[0] - 1);
3550 			if (key.type == BTRFS_EXTENT_DATA_KEY)
3551 				path->slots[0]--;
3552 		}
3553 
3554 		nritems = btrfs_header_nritems(path->nodes[0]);
3555 process_slot:
3556 		if (path->slots[0] >= nritems) {
3557 			ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3558 			if (ret < 0)
3559 				goto out;
3560 			if (ret > 0)
3561 				break;
3562 			nritems = btrfs_header_nritems(path->nodes[0]);
3563 		}
3564 		leaf = path->nodes[0];
3565 		slot = path->slots[0];
3566 
3567 		btrfs_item_key_to_cpu(leaf, &key, slot);
3568 		if (key.type > BTRFS_EXTENT_DATA_KEY ||
3569 		    key.objectid != btrfs_ino(BTRFS_I(src)))
3570 			break;
3571 
3572 		if (key.type == BTRFS_EXTENT_DATA_KEY) {
3573 			struct btrfs_file_extent_item *extent;
3574 			int type;
3575 			u32 size;
3576 			struct btrfs_key new_key;
3577 			u64 disko = 0, diskl = 0;
3578 			u64 datao = 0, datal = 0;
3579 			u8 comp;
3580 			u64 drop_start;
3581 
3582 			extent = btrfs_item_ptr(leaf, slot,
3583 						struct btrfs_file_extent_item);
3584 			comp = btrfs_file_extent_compression(leaf, extent);
3585 			type = btrfs_file_extent_type(leaf, extent);
3586 			if (type == BTRFS_FILE_EXTENT_REG ||
3587 			    type == BTRFS_FILE_EXTENT_PREALLOC) {
3588 				disko = btrfs_file_extent_disk_bytenr(leaf,
3589 								      extent);
3590 				diskl = btrfs_file_extent_disk_num_bytes(leaf,
3591 								 extent);
3592 				datao = btrfs_file_extent_offset(leaf, extent);
3593 				datal = btrfs_file_extent_num_bytes(leaf,
3594 								    extent);
3595 			} else if (type == BTRFS_FILE_EXTENT_INLINE) {
3596 				/* take upper bound, may be compressed */
3597 				datal = btrfs_file_extent_ram_bytes(leaf,
3598 								    extent);
3599 			}
3600 
3601 			/*
3602 			 * The first search might have left us at an extent
3603 			 * item that ends before our target range's start, can
3604 			 * happen if we have holes and NO_HOLES feature enabled.
3605 			 */
3606 			if (key.offset + datal <= off) {
3607 				path->slots[0]++;
3608 				goto process_slot;
3609 			} else if (key.offset >= off + len) {
3610 				break;
3611 			}
3612 			next_key_min_offset = key.offset + datal;
3613 			size = btrfs_item_size_nr(leaf, slot);
3614 			read_extent_buffer(leaf, buf,
3615 					   btrfs_item_ptr_offset(leaf, slot),
3616 					   size);
3617 
3618 			btrfs_release_path(path);
3619 			path->leave_spinning = 0;
3620 
3621 			memcpy(&new_key, &key, sizeof(new_key));
3622 			new_key.objectid = btrfs_ino(BTRFS_I(inode));
3623 			if (off <= key.offset)
3624 				new_key.offset = key.offset + destoff - off;
3625 			else
3626 				new_key.offset = destoff;
3627 
3628 			/*
3629 			 * Deal with a hole that doesn't have an extent item
3630 			 * that represents it (NO_HOLES feature enabled).
3631 			 * This hole is either in the middle of the cloning
3632 			 * range or at the beginning (fully overlaps it or
3633 			 * partially overlaps it).
3634 			 */
3635 			if (new_key.offset != last_dest_end)
3636 				drop_start = last_dest_end;
3637 			else
3638 				drop_start = new_key.offset;
3639 
3640 			/*
3641 			 * 1 - adjusting old extent (we may have to split it)
3642 			 * 1 - add new extent
3643 			 * 1 - inode update
3644 			 */
3645 			trans = btrfs_start_transaction(root, 3);
3646 			if (IS_ERR(trans)) {
3647 				ret = PTR_ERR(trans);
3648 				goto out;
3649 			}
3650 
3651 			if (type == BTRFS_FILE_EXTENT_REG ||
3652 			    type == BTRFS_FILE_EXTENT_PREALLOC) {
3653 				/*
3654 				 *    a  | --- range to clone ---|  b
3655 				 * | ------------- extent ------------- |
3656 				 */
3657 
3658 				/* subtract range b */
3659 				if (key.offset + datal > off + len)
3660 					datal = off + len - key.offset;
3661 
3662 				/* subtract range a */
3663 				if (off > key.offset) {
3664 					datao += off - key.offset;
3665 					datal -= off - key.offset;
3666 				}
3667 
3668 				ret = btrfs_drop_extents(trans, root, inode,
3669 							 drop_start,
3670 							 new_key.offset + datal,
3671 							 1);
3672 				if (ret) {
3673 					if (ret != -EOPNOTSUPP)
3674 						btrfs_abort_transaction(trans,
3675 									ret);
3676 					btrfs_end_transaction(trans);
3677 					goto out;
3678 				}
3679 
3680 				ret = btrfs_insert_empty_item(trans, root, path,
3681 							      &new_key, size);
3682 				if (ret) {
3683 					btrfs_abort_transaction(trans, ret);
3684 					btrfs_end_transaction(trans);
3685 					goto out;
3686 				}
3687 
3688 				leaf = path->nodes[0];
3689 				slot = path->slots[0];
3690 				write_extent_buffer(leaf, buf,
3691 					    btrfs_item_ptr_offset(leaf, slot),
3692 					    size);
3693 
3694 				extent = btrfs_item_ptr(leaf, slot,
3695 						struct btrfs_file_extent_item);
3696 
3697 				/* disko == 0 means it's a hole */
3698 				if (!disko)
3699 					datao = 0;
3700 
3701 				btrfs_set_file_extent_offset(leaf, extent,
3702 							     datao);
3703 				btrfs_set_file_extent_num_bytes(leaf, extent,
3704 								datal);
3705 
3706 				if (disko) {
3707 					inode_add_bytes(inode, datal);
3708 					ret = btrfs_inc_extent_ref(trans,
3709 							fs_info,
3710 							disko, diskl, 0,
3711 							root->root_key.objectid,
3712 							btrfs_ino(BTRFS_I(inode)),
3713 							new_key.offset - datao);
3714 					if (ret) {
3715 						btrfs_abort_transaction(trans,
3716 									ret);
3717 						btrfs_end_transaction(trans);
3718 						goto out;
3719 
3720 					}
3721 				}
3722 			} else if (type == BTRFS_FILE_EXTENT_INLINE) {
3723 				u64 skip = 0;
3724 				u64 trim = 0;
3725 
3726 				if (off > key.offset) {
3727 					skip = off - key.offset;
3728 					new_key.offset += skip;
3729 				}
3730 
3731 				if (key.offset + datal > off + len)
3732 					trim = key.offset + datal - (off + len);
3733 
3734 				if (comp && (skip || trim)) {
3735 					ret = -EINVAL;
3736 					btrfs_end_transaction(trans);
3737 					goto out;
3738 				}
3739 				size -= skip + trim;
3740 				datal -= skip + trim;
3741 
3742 				ret = clone_copy_inline_extent(inode,
3743 							       trans, path,
3744 							       &new_key,
3745 							       drop_start,
3746 							       datal,
3747 							       skip, size, buf);
3748 				if (ret) {
3749 					if (ret != -EOPNOTSUPP)
3750 						btrfs_abort_transaction(trans,
3751 									ret);
3752 					btrfs_end_transaction(trans);
3753 					goto out;
3754 				}
3755 				leaf = path->nodes[0];
3756 				slot = path->slots[0];
3757 			}
3758 
3759 			/* If we have an implicit hole (NO_HOLES feature). */
3760 			if (drop_start < new_key.offset)
3761 				clone_update_extent_map(BTRFS_I(inode), trans,
3762 						NULL, drop_start,
3763 						new_key.offset - drop_start);
3764 
3765 			clone_update_extent_map(BTRFS_I(inode), trans,
3766 					path, 0, 0);
3767 
3768 			btrfs_mark_buffer_dirty(leaf);
3769 			btrfs_release_path(path);
3770 
3771 			last_dest_end = ALIGN(new_key.offset + datal,
3772 					      fs_info->sectorsize);
3773 			ret = clone_finish_inode_update(trans, inode,
3774 							last_dest_end,
3775 							destoff, olen,
3776 							no_time_update);
3777 			if (ret)
3778 				goto out;
3779 			if (new_key.offset + datal >= destoff + len)
3780 				break;
3781 		}
3782 		btrfs_release_path(path);
3783 		key.offset = next_key_min_offset;
3784 
3785 		if (fatal_signal_pending(current)) {
3786 			ret = -EINTR;
3787 			goto out;
3788 		}
3789 	}
3790 	ret = 0;
3791 
3792 	if (last_dest_end < destoff + len) {
3793 		/*
3794 		 * We have an implicit hole (NO_HOLES feature is enabled) that
3795 		 * fully or partially overlaps our cloning range at its end.
3796 		 */
3797 		btrfs_release_path(path);
3798 
3799 		/*
3800 		 * 1 - remove extent(s)
3801 		 * 1 - inode update
3802 		 */
3803 		trans = btrfs_start_transaction(root, 2);
3804 		if (IS_ERR(trans)) {
3805 			ret = PTR_ERR(trans);
3806 			goto out;
3807 		}
3808 		ret = btrfs_drop_extents(trans, root, inode,
3809 					 last_dest_end, destoff + len, 1);
3810 		if (ret) {
3811 			if (ret != -EOPNOTSUPP)
3812 				btrfs_abort_transaction(trans, ret);
3813 			btrfs_end_transaction(trans);
3814 			goto out;
3815 		}
3816 		clone_update_extent_map(BTRFS_I(inode), trans, NULL,
3817 				last_dest_end,
3818 				destoff + len - last_dest_end);
3819 		ret = clone_finish_inode_update(trans, inode, destoff + len,
3820 						destoff, olen, no_time_update);
3821 	}
3822 
3823 out:
3824 	btrfs_free_path(path);
3825 	kvfree(buf);
3826 	return ret;
3827 }
3828 
3829 static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
3830 					u64 off, u64 olen, u64 destoff)
3831 {
3832 	struct inode *inode = file_inode(file);
3833 	struct inode *src = file_inode(file_src);
3834 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3835 	struct btrfs_root *root = BTRFS_I(inode)->root;
3836 	int ret;
3837 	u64 len = olen;
3838 	u64 bs = fs_info->sb->s_blocksize;
3839 	int same_inode = src == inode;
3840 
3841 	/*
3842 	 * TODO:
3843 	 * - split compressed inline extents.  annoying: we need to
3844 	 *   decompress into destination's address_space (the file offset
3845 	 *   may change, so source mapping won't do), then recompress (or
3846 	 *   otherwise reinsert) a subrange.
3847 	 *
3848 	 * - split destination inode's inline extents.  The inline extents can
3849 	 *   be either compressed or non-compressed.
3850 	 */
3851 
3852 	if (btrfs_root_readonly(root))
3853 		return -EROFS;
3854 
3855 	if (file_src->f_path.mnt != file->f_path.mnt ||
3856 	    src->i_sb != inode->i_sb)
3857 		return -EXDEV;
3858 
3859 	/* don't make the dst file partly checksummed */
3860 	if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3861 	    (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3862 		return -EINVAL;
3863 
3864 	if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3865 		return -EISDIR;
3866 
3867 	if (!same_inode) {
3868 		btrfs_double_inode_lock(src, inode);
3869 	} else {
3870 		inode_lock(src);
3871 	}
3872 
3873 	/* determine range to clone */
3874 	ret = -EINVAL;
3875 	if (off + len > src->i_size || off + len < off)
3876 		goto out_unlock;
3877 	if (len == 0)
3878 		olen = len = src->i_size - off;
3879 	/* if we extend to eof, continue to block boundary */
3880 	if (off + len == src->i_size)
3881 		len = ALIGN(src->i_size, bs) - off;
3882 
3883 	if (len == 0) {
3884 		ret = 0;
3885 		goto out_unlock;
3886 	}
3887 
3888 	/* verify the end result is block aligned */
3889 	if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3890 	    !IS_ALIGNED(destoff, bs))
3891 		goto out_unlock;
3892 
3893 	/* verify if ranges are overlapped within the same file */
3894 	if (same_inode) {
3895 		if (destoff + len > off && destoff < off + len)
3896 			goto out_unlock;
3897 	}
3898 
3899 	if (destoff > inode->i_size) {
3900 		ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3901 		if (ret)
3902 			goto out_unlock;
3903 	}
3904 
3905 	/*
3906 	 * Lock the target range too. Right after we replace the file extent
3907 	 * items in the fs tree (which now point to the cloned data), we might
3908 	 * have a worker replace them with extent items relative to a write
3909 	 * operation that was issued before this clone operation (i.e. confront
3910 	 * with inode.c:btrfs_finish_ordered_io).
3911 	 */
3912 	if (same_inode) {
3913 		u64 lock_start = min_t(u64, off, destoff);
3914 		u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
3915 
3916 		ret = lock_extent_range(src, lock_start, lock_len, true);
3917 	} else {
3918 		ret = btrfs_double_extent_lock(src, off, inode, destoff, len,
3919 					       true);
3920 	}
3921 	ASSERT(ret == 0);
3922 	if (WARN_ON(ret)) {
3923 		/* ranges in the io trees already unlocked */
3924 		goto out_unlock;
3925 	}
3926 
3927 	ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
3928 
3929 	if (same_inode) {
3930 		u64 lock_start = min_t(u64, off, destoff);
3931 		u64 lock_end = max_t(u64, off, destoff) + len - 1;
3932 
3933 		unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
3934 	} else {
3935 		btrfs_double_extent_unlock(src, off, inode, destoff, len);
3936 	}
3937 	/*
3938 	 * Truncate page cache pages so that future reads will see the cloned
3939 	 * data immediately and not the previous data.
3940 	 */
3941 	truncate_inode_pages_range(&inode->i_data,
3942 				round_down(destoff, PAGE_SIZE),
3943 				round_up(destoff + len, PAGE_SIZE) - 1);
3944 out_unlock:
3945 	if (!same_inode)
3946 		btrfs_double_inode_unlock(src, inode);
3947 	else
3948 		inode_unlock(src);
3949 	return ret;
3950 }
3951 
3952 int btrfs_clone_file_range(struct file *src_file, loff_t off,
3953 		struct file *dst_file, loff_t destoff, u64 len)
3954 {
3955 	return btrfs_clone_files(dst_file, src_file, off, len, destoff);
3956 }
3957 
3958 /*
3959  * there are many ways the trans_start and trans_end ioctls can lead
3960  * to deadlocks.  They should only be used by applications that
3961  * basically own the machine, and have a very in depth understanding
3962  * of all the possible deadlocks and enospc problems.
3963  */
3964 static long btrfs_ioctl_trans_start(struct file *file)
3965 {
3966 	struct inode *inode = file_inode(file);
3967 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3968 	struct btrfs_root *root = BTRFS_I(inode)->root;
3969 	struct btrfs_trans_handle *trans;
3970 	struct btrfs_file_private *private;
3971 	int ret;
3972 	static bool warned = false;
3973 
3974 	ret = -EPERM;
3975 	if (!capable(CAP_SYS_ADMIN))
3976 		goto out;
3977 
3978 	if (!warned) {
3979 		btrfs_warn(fs_info,
3980 			"Userspace transaction mechanism is considered "
3981 			"deprecated and slated to be removed in 4.17. "
3982 			"If you have a valid use case please "
3983 			"speak up on the mailing list");
3984 		WARN_ON(1);
3985 		warned = true;
3986 	}
3987 
3988 	ret = -EINPROGRESS;
3989 	private = file->private_data;
3990 	if (private && private->trans)
3991 		goto out;
3992 	if (!private) {
3993 		private = kzalloc(sizeof(struct btrfs_file_private),
3994 				  GFP_KERNEL);
3995 		if (!private)
3996 			return -ENOMEM;
3997 		file->private_data = private;
3998 	}
3999 
4000 	ret = -EROFS;
4001 	if (btrfs_root_readonly(root))
4002 		goto out;
4003 
4004 	ret = mnt_want_write_file(file);
4005 	if (ret)
4006 		goto out;
4007 
4008 	atomic_inc(&fs_info->open_ioctl_trans);
4009 
4010 	ret = -ENOMEM;
4011 	trans = btrfs_start_ioctl_transaction(root);
4012 	if (IS_ERR(trans))
4013 		goto out_drop;
4014 
4015 	private->trans = trans;
4016 	return 0;
4017 
4018 out_drop:
4019 	atomic_dec(&fs_info->open_ioctl_trans);
4020 	mnt_drop_write_file(file);
4021 out:
4022 	return ret;
4023 }
4024 
4025 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
4026 {
4027 	struct inode *inode = file_inode(file);
4028 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4029 	struct btrfs_root *root = BTRFS_I(inode)->root;
4030 	struct btrfs_root *new_root;
4031 	struct btrfs_dir_item *di;
4032 	struct btrfs_trans_handle *trans;
4033 	struct btrfs_path *path;
4034 	struct btrfs_key location;
4035 	struct btrfs_disk_key disk_key;
4036 	u64 objectid = 0;
4037 	u64 dir_id;
4038 	int ret;
4039 
4040 	if (!capable(CAP_SYS_ADMIN))
4041 		return -EPERM;
4042 
4043 	ret = mnt_want_write_file(file);
4044 	if (ret)
4045 		return ret;
4046 
4047 	if (copy_from_user(&objectid, argp, sizeof(objectid))) {
4048 		ret = -EFAULT;
4049 		goto out;
4050 	}
4051 
4052 	if (!objectid)
4053 		objectid = BTRFS_FS_TREE_OBJECTID;
4054 
4055 	location.objectid = objectid;
4056 	location.type = BTRFS_ROOT_ITEM_KEY;
4057 	location.offset = (u64)-1;
4058 
4059 	new_root = btrfs_read_fs_root_no_name(fs_info, &location);
4060 	if (IS_ERR(new_root)) {
4061 		ret = PTR_ERR(new_root);
4062 		goto out;
4063 	}
4064 	if (!is_fstree(new_root->objectid)) {
4065 		ret = -ENOENT;
4066 		goto out;
4067 	}
4068 
4069 	path = btrfs_alloc_path();
4070 	if (!path) {
4071 		ret = -ENOMEM;
4072 		goto out;
4073 	}
4074 	path->leave_spinning = 1;
4075 
4076 	trans = btrfs_start_transaction(root, 1);
4077 	if (IS_ERR(trans)) {
4078 		btrfs_free_path(path);
4079 		ret = PTR_ERR(trans);
4080 		goto out;
4081 	}
4082 
4083 	dir_id = btrfs_super_root_dir(fs_info->super_copy);
4084 	di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
4085 				   dir_id, "default", 7, 1);
4086 	if (IS_ERR_OR_NULL(di)) {
4087 		btrfs_free_path(path);
4088 		btrfs_end_transaction(trans);
4089 		btrfs_err(fs_info,
4090 			  "Umm, you don't have the default diritem, this isn't going to work");
4091 		ret = -ENOENT;
4092 		goto out;
4093 	}
4094 
4095 	btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
4096 	btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
4097 	btrfs_mark_buffer_dirty(path->nodes[0]);
4098 	btrfs_free_path(path);
4099 
4100 	btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
4101 	btrfs_end_transaction(trans);
4102 out:
4103 	mnt_drop_write_file(file);
4104 	return ret;
4105 }
4106 
4107 void btrfs_get_block_group_info(struct list_head *groups_list,
4108 				struct btrfs_ioctl_space_info *space)
4109 {
4110 	struct btrfs_block_group_cache *block_group;
4111 
4112 	space->total_bytes = 0;
4113 	space->used_bytes = 0;
4114 	space->flags = 0;
4115 	list_for_each_entry(block_group, groups_list, list) {
4116 		space->flags = block_group->flags;
4117 		space->total_bytes += block_group->key.offset;
4118 		space->used_bytes +=
4119 			btrfs_block_group_used(&block_group->item);
4120 	}
4121 }
4122 
4123 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
4124 				   void __user *arg)
4125 {
4126 	struct btrfs_ioctl_space_args space_args;
4127 	struct btrfs_ioctl_space_info space;
4128 	struct btrfs_ioctl_space_info *dest;
4129 	struct btrfs_ioctl_space_info *dest_orig;
4130 	struct btrfs_ioctl_space_info __user *user_dest;
4131 	struct btrfs_space_info *info;
4132 	u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
4133 		       BTRFS_BLOCK_GROUP_SYSTEM,
4134 		       BTRFS_BLOCK_GROUP_METADATA,
4135 		       BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
4136 	int num_types = 4;
4137 	int alloc_size;
4138 	int ret = 0;
4139 	u64 slot_count = 0;
4140 	int i, c;
4141 
4142 	if (copy_from_user(&space_args,
4143 			   (struct btrfs_ioctl_space_args __user *)arg,
4144 			   sizeof(space_args)))
4145 		return -EFAULT;
4146 
4147 	for (i = 0; i < num_types; i++) {
4148 		struct btrfs_space_info *tmp;
4149 
4150 		info = NULL;
4151 		rcu_read_lock();
4152 		list_for_each_entry_rcu(tmp, &fs_info->space_info,
4153 					list) {
4154 			if (tmp->flags == types[i]) {
4155 				info = tmp;
4156 				break;
4157 			}
4158 		}
4159 		rcu_read_unlock();
4160 
4161 		if (!info)
4162 			continue;
4163 
4164 		down_read(&info->groups_sem);
4165 		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4166 			if (!list_empty(&info->block_groups[c]))
4167 				slot_count++;
4168 		}
4169 		up_read(&info->groups_sem);
4170 	}
4171 
4172 	/*
4173 	 * Global block reserve, exported as a space_info
4174 	 */
4175 	slot_count++;
4176 
4177 	/* space_slots == 0 means they are asking for a count */
4178 	if (space_args.space_slots == 0) {
4179 		space_args.total_spaces = slot_count;
4180 		goto out;
4181 	}
4182 
4183 	slot_count = min_t(u64, space_args.space_slots, slot_count);
4184 
4185 	alloc_size = sizeof(*dest) * slot_count;
4186 
4187 	/* we generally have at most 6 or so space infos, one for each raid
4188 	 * level.  So, a whole page should be more than enough for everyone
4189 	 */
4190 	if (alloc_size > PAGE_SIZE)
4191 		return -ENOMEM;
4192 
4193 	space_args.total_spaces = 0;
4194 	dest = kmalloc(alloc_size, GFP_KERNEL);
4195 	if (!dest)
4196 		return -ENOMEM;
4197 	dest_orig = dest;
4198 
4199 	/* now we have a buffer to copy into */
4200 	for (i = 0; i < num_types; i++) {
4201 		struct btrfs_space_info *tmp;
4202 
4203 		if (!slot_count)
4204 			break;
4205 
4206 		info = NULL;
4207 		rcu_read_lock();
4208 		list_for_each_entry_rcu(tmp, &fs_info->space_info,
4209 					list) {
4210 			if (tmp->flags == types[i]) {
4211 				info = tmp;
4212 				break;
4213 			}
4214 		}
4215 		rcu_read_unlock();
4216 
4217 		if (!info)
4218 			continue;
4219 		down_read(&info->groups_sem);
4220 		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4221 			if (!list_empty(&info->block_groups[c])) {
4222 				btrfs_get_block_group_info(
4223 					&info->block_groups[c], &space);
4224 				memcpy(dest, &space, sizeof(space));
4225 				dest++;
4226 				space_args.total_spaces++;
4227 				slot_count--;
4228 			}
4229 			if (!slot_count)
4230 				break;
4231 		}
4232 		up_read(&info->groups_sem);
4233 	}
4234 
4235 	/*
4236 	 * Add global block reserve
4237 	 */
4238 	if (slot_count) {
4239 		struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4240 
4241 		spin_lock(&block_rsv->lock);
4242 		space.total_bytes = block_rsv->size;
4243 		space.used_bytes = block_rsv->size - block_rsv->reserved;
4244 		spin_unlock(&block_rsv->lock);
4245 		space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4246 		memcpy(dest, &space, sizeof(space));
4247 		space_args.total_spaces++;
4248 	}
4249 
4250 	user_dest = (struct btrfs_ioctl_space_info __user *)
4251 		(arg + sizeof(struct btrfs_ioctl_space_args));
4252 
4253 	if (copy_to_user(user_dest, dest_orig, alloc_size))
4254 		ret = -EFAULT;
4255 
4256 	kfree(dest_orig);
4257 out:
4258 	if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4259 		ret = -EFAULT;
4260 
4261 	return ret;
4262 }
4263 
4264 /*
4265  * there are many ways the trans_start and trans_end ioctls can lead
4266  * to deadlocks.  They should only be used by applications that
4267  * basically own the machine, and have a very in depth understanding
4268  * of all the possible deadlocks and enospc problems.
4269  */
4270 long btrfs_ioctl_trans_end(struct file *file)
4271 {
4272 	struct inode *inode = file_inode(file);
4273 	struct btrfs_root *root = BTRFS_I(inode)->root;
4274 	struct btrfs_file_private *private = file->private_data;
4275 
4276 	if (!private || !private->trans)
4277 		return -EINVAL;
4278 
4279 	btrfs_end_transaction(private->trans);
4280 	private->trans = NULL;
4281 
4282 	atomic_dec(&root->fs_info->open_ioctl_trans);
4283 
4284 	mnt_drop_write_file(file);
4285 	return 0;
4286 }
4287 
4288 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4289 					    void __user *argp)
4290 {
4291 	struct btrfs_trans_handle *trans;
4292 	u64 transid;
4293 	int ret;
4294 
4295 	trans = btrfs_attach_transaction_barrier(root);
4296 	if (IS_ERR(trans)) {
4297 		if (PTR_ERR(trans) != -ENOENT)
4298 			return PTR_ERR(trans);
4299 
4300 		/* No running transaction, don't bother */
4301 		transid = root->fs_info->last_trans_committed;
4302 		goto out;
4303 	}
4304 	transid = trans->transid;
4305 	ret = btrfs_commit_transaction_async(trans, 0);
4306 	if (ret) {
4307 		btrfs_end_transaction(trans);
4308 		return ret;
4309 	}
4310 out:
4311 	if (argp)
4312 		if (copy_to_user(argp, &transid, sizeof(transid)))
4313 			return -EFAULT;
4314 	return 0;
4315 }
4316 
4317 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
4318 					   void __user *argp)
4319 {
4320 	u64 transid;
4321 
4322 	if (argp) {
4323 		if (copy_from_user(&transid, argp, sizeof(transid)))
4324 			return -EFAULT;
4325 	} else {
4326 		transid = 0;  /* current trans */
4327 	}
4328 	return btrfs_wait_for_commit(fs_info, transid);
4329 }
4330 
4331 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4332 {
4333 	struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
4334 	struct btrfs_ioctl_scrub_args *sa;
4335 	int ret;
4336 
4337 	if (!capable(CAP_SYS_ADMIN))
4338 		return -EPERM;
4339 
4340 	sa = memdup_user(arg, sizeof(*sa));
4341 	if (IS_ERR(sa))
4342 		return PTR_ERR(sa);
4343 
4344 	if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4345 		ret = mnt_want_write_file(file);
4346 		if (ret)
4347 			goto out;
4348 	}
4349 
4350 	ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
4351 			      &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4352 			      0);
4353 
4354 	if (copy_to_user(arg, sa, sizeof(*sa)))
4355 		ret = -EFAULT;
4356 
4357 	if (!(sa->flags & BTRFS_SCRUB_READONLY))
4358 		mnt_drop_write_file(file);
4359 out:
4360 	kfree(sa);
4361 	return ret;
4362 }
4363 
4364 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
4365 {
4366 	if (!capable(CAP_SYS_ADMIN))
4367 		return -EPERM;
4368 
4369 	return btrfs_scrub_cancel(fs_info);
4370 }
4371 
4372 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
4373 				       void __user *arg)
4374 {
4375 	struct btrfs_ioctl_scrub_args *sa;
4376 	int ret;
4377 
4378 	if (!capable(CAP_SYS_ADMIN))
4379 		return -EPERM;
4380 
4381 	sa = memdup_user(arg, sizeof(*sa));
4382 	if (IS_ERR(sa))
4383 		return PTR_ERR(sa);
4384 
4385 	ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
4386 
4387 	if (copy_to_user(arg, sa, sizeof(*sa)))
4388 		ret = -EFAULT;
4389 
4390 	kfree(sa);
4391 	return ret;
4392 }
4393 
4394 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
4395 				      void __user *arg)
4396 {
4397 	struct btrfs_ioctl_get_dev_stats *sa;
4398 	int ret;
4399 
4400 	sa = memdup_user(arg, sizeof(*sa));
4401 	if (IS_ERR(sa))
4402 		return PTR_ERR(sa);
4403 
4404 	if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4405 		kfree(sa);
4406 		return -EPERM;
4407 	}
4408 
4409 	ret = btrfs_get_dev_stats(fs_info, sa);
4410 
4411 	if (copy_to_user(arg, sa, sizeof(*sa)))
4412 		ret = -EFAULT;
4413 
4414 	kfree(sa);
4415 	return ret;
4416 }
4417 
4418 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
4419 				    void __user *arg)
4420 {
4421 	struct btrfs_ioctl_dev_replace_args *p;
4422 	int ret;
4423 
4424 	if (!capable(CAP_SYS_ADMIN))
4425 		return -EPERM;
4426 
4427 	p = memdup_user(arg, sizeof(*p));
4428 	if (IS_ERR(p))
4429 		return PTR_ERR(p);
4430 
4431 	switch (p->cmd) {
4432 	case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4433 		if (sb_rdonly(fs_info->sb)) {
4434 			ret = -EROFS;
4435 			goto out;
4436 		}
4437 		if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4438 			ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4439 		} else {
4440 			ret = btrfs_dev_replace_by_ioctl(fs_info, p);
4441 			clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4442 		}
4443 		break;
4444 	case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4445 		btrfs_dev_replace_status(fs_info, p);
4446 		ret = 0;
4447 		break;
4448 	case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4449 		ret = btrfs_dev_replace_cancel(fs_info, p);
4450 		break;
4451 	default:
4452 		ret = -EINVAL;
4453 		break;
4454 	}
4455 
4456 	if (copy_to_user(arg, p, sizeof(*p)))
4457 		ret = -EFAULT;
4458 out:
4459 	kfree(p);
4460 	return ret;
4461 }
4462 
4463 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4464 {
4465 	int ret = 0;
4466 	int i;
4467 	u64 rel_ptr;
4468 	int size;
4469 	struct btrfs_ioctl_ino_path_args *ipa = NULL;
4470 	struct inode_fs_paths *ipath = NULL;
4471 	struct btrfs_path *path;
4472 
4473 	if (!capable(CAP_DAC_READ_SEARCH))
4474 		return -EPERM;
4475 
4476 	path = btrfs_alloc_path();
4477 	if (!path) {
4478 		ret = -ENOMEM;
4479 		goto out;
4480 	}
4481 
4482 	ipa = memdup_user(arg, sizeof(*ipa));
4483 	if (IS_ERR(ipa)) {
4484 		ret = PTR_ERR(ipa);
4485 		ipa = NULL;
4486 		goto out;
4487 	}
4488 
4489 	size = min_t(u32, ipa->size, 4096);
4490 	ipath = init_ipath(size, root, path);
4491 	if (IS_ERR(ipath)) {
4492 		ret = PTR_ERR(ipath);
4493 		ipath = NULL;
4494 		goto out;
4495 	}
4496 
4497 	ret = paths_from_inode(ipa->inum, ipath);
4498 	if (ret < 0)
4499 		goto out;
4500 
4501 	for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4502 		rel_ptr = ipath->fspath->val[i] -
4503 			  (u64)(unsigned long)ipath->fspath->val;
4504 		ipath->fspath->val[i] = rel_ptr;
4505 	}
4506 
4507 	ret = copy_to_user((void *)(unsigned long)ipa->fspath,
4508 			   (void *)(unsigned long)ipath->fspath, size);
4509 	if (ret) {
4510 		ret = -EFAULT;
4511 		goto out;
4512 	}
4513 
4514 out:
4515 	btrfs_free_path(path);
4516 	free_ipath(ipath);
4517 	kfree(ipa);
4518 
4519 	return ret;
4520 }
4521 
4522 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4523 {
4524 	struct btrfs_data_container *inodes = ctx;
4525 	const size_t c = 3 * sizeof(u64);
4526 
4527 	if (inodes->bytes_left >= c) {
4528 		inodes->bytes_left -= c;
4529 		inodes->val[inodes->elem_cnt] = inum;
4530 		inodes->val[inodes->elem_cnt + 1] = offset;
4531 		inodes->val[inodes->elem_cnt + 2] = root;
4532 		inodes->elem_cnt += 3;
4533 	} else {
4534 		inodes->bytes_missing += c - inodes->bytes_left;
4535 		inodes->bytes_left = 0;
4536 		inodes->elem_missed += 3;
4537 	}
4538 
4539 	return 0;
4540 }
4541 
4542 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
4543 					void __user *arg)
4544 {
4545 	int ret = 0;
4546 	int size;
4547 	struct btrfs_ioctl_logical_ino_args *loi;
4548 	struct btrfs_data_container *inodes = NULL;
4549 	struct btrfs_path *path = NULL;
4550 
4551 	if (!capable(CAP_SYS_ADMIN))
4552 		return -EPERM;
4553 
4554 	loi = memdup_user(arg, sizeof(*loi));
4555 	if (IS_ERR(loi))
4556 		return PTR_ERR(loi);
4557 
4558 	path = btrfs_alloc_path();
4559 	if (!path) {
4560 		ret = -ENOMEM;
4561 		goto out;
4562 	}
4563 
4564 	size = min_t(u32, loi->size, SZ_64K);
4565 	inodes = init_data_container(size);
4566 	if (IS_ERR(inodes)) {
4567 		ret = PTR_ERR(inodes);
4568 		inodes = NULL;
4569 		goto out;
4570 	}
4571 
4572 	ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
4573 					  build_ino_list, inodes);
4574 	if (ret == -EINVAL)
4575 		ret = -ENOENT;
4576 	if (ret < 0)
4577 		goto out;
4578 
4579 	ret = copy_to_user((void *)(unsigned long)loi->inodes,
4580 			   (void *)(unsigned long)inodes, size);
4581 	if (ret)
4582 		ret = -EFAULT;
4583 
4584 out:
4585 	btrfs_free_path(path);
4586 	kvfree(inodes);
4587 	kfree(loi);
4588 
4589 	return ret;
4590 }
4591 
4592 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4593 			       struct btrfs_ioctl_balance_args *bargs)
4594 {
4595 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4596 
4597 	bargs->flags = bctl->flags;
4598 
4599 	if (atomic_read(&fs_info->balance_running))
4600 		bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4601 	if (atomic_read(&fs_info->balance_pause_req))
4602 		bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4603 	if (atomic_read(&fs_info->balance_cancel_req))
4604 		bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4605 
4606 	memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4607 	memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4608 	memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4609 
4610 	if (lock) {
4611 		spin_lock(&fs_info->balance_lock);
4612 		memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4613 		spin_unlock(&fs_info->balance_lock);
4614 	} else {
4615 		memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4616 	}
4617 }
4618 
4619 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4620 {
4621 	struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4622 	struct btrfs_fs_info *fs_info = root->fs_info;
4623 	struct btrfs_ioctl_balance_args *bargs;
4624 	struct btrfs_balance_control *bctl;
4625 	bool need_unlock; /* for mut. excl. ops lock */
4626 	int ret;
4627 
4628 	if (!capable(CAP_SYS_ADMIN))
4629 		return -EPERM;
4630 
4631 	ret = mnt_want_write_file(file);
4632 	if (ret)
4633 		return ret;
4634 
4635 again:
4636 	if (!test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4637 		mutex_lock(&fs_info->volume_mutex);
4638 		mutex_lock(&fs_info->balance_mutex);
4639 		need_unlock = true;
4640 		goto locked;
4641 	}
4642 
4643 	/*
4644 	 * mut. excl. ops lock is locked.  Three possibilities:
4645 	 *   (1) some other op is running
4646 	 *   (2) balance is running
4647 	 *   (3) balance is paused -- special case (think resume)
4648 	 */
4649 	mutex_lock(&fs_info->balance_mutex);
4650 	if (fs_info->balance_ctl) {
4651 		/* this is either (2) or (3) */
4652 		if (!atomic_read(&fs_info->balance_running)) {
4653 			mutex_unlock(&fs_info->balance_mutex);
4654 			if (!mutex_trylock(&fs_info->volume_mutex))
4655 				goto again;
4656 			mutex_lock(&fs_info->balance_mutex);
4657 
4658 			if (fs_info->balance_ctl &&
4659 			    !atomic_read(&fs_info->balance_running)) {
4660 				/* this is (3) */
4661 				need_unlock = false;
4662 				goto locked;
4663 			}
4664 
4665 			mutex_unlock(&fs_info->balance_mutex);
4666 			mutex_unlock(&fs_info->volume_mutex);
4667 			goto again;
4668 		} else {
4669 			/* this is (2) */
4670 			mutex_unlock(&fs_info->balance_mutex);
4671 			ret = -EINPROGRESS;
4672 			goto out;
4673 		}
4674 	} else {
4675 		/* this is (1) */
4676 		mutex_unlock(&fs_info->balance_mutex);
4677 		ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4678 		goto out;
4679 	}
4680 
4681 locked:
4682 	BUG_ON(!test_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4683 
4684 	if (arg) {
4685 		bargs = memdup_user(arg, sizeof(*bargs));
4686 		if (IS_ERR(bargs)) {
4687 			ret = PTR_ERR(bargs);
4688 			goto out_unlock;
4689 		}
4690 
4691 		if (bargs->flags & BTRFS_BALANCE_RESUME) {
4692 			if (!fs_info->balance_ctl) {
4693 				ret = -ENOTCONN;
4694 				goto out_bargs;
4695 			}
4696 
4697 			bctl = fs_info->balance_ctl;
4698 			spin_lock(&fs_info->balance_lock);
4699 			bctl->flags |= BTRFS_BALANCE_RESUME;
4700 			spin_unlock(&fs_info->balance_lock);
4701 
4702 			goto do_balance;
4703 		}
4704 	} else {
4705 		bargs = NULL;
4706 	}
4707 
4708 	if (fs_info->balance_ctl) {
4709 		ret = -EINPROGRESS;
4710 		goto out_bargs;
4711 	}
4712 
4713 	bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4714 	if (!bctl) {
4715 		ret = -ENOMEM;
4716 		goto out_bargs;
4717 	}
4718 
4719 	bctl->fs_info = fs_info;
4720 	if (arg) {
4721 		memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4722 		memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4723 		memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4724 
4725 		bctl->flags = bargs->flags;
4726 	} else {
4727 		/* balance everything - no filters */
4728 		bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4729 	}
4730 
4731 	if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4732 		ret = -EINVAL;
4733 		goto out_bctl;
4734 	}
4735 
4736 do_balance:
4737 	/*
4738 	 * Ownership of bctl and filesystem flag BTRFS_FS_EXCL_OP
4739 	 * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4740 	 * or, if restriper was paused all the way until unmount, in
4741 	 * free_fs_info.  The flag is cleared in __cancel_balance.
4742 	 */
4743 	need_unlock = false;
4744 
4745 	ret = btrfs_balance(bctl, bargs);
4746 	bctl = NULL;
4747 
4748 	if (arg) {
4749 		if (copy_to_user(arg, bargs, sizeof(*bargs)))
4750 			ret = -EFAULT;
4751 	}
4752 
4753 out_bctl:
4754 	kfree(bctl);
4755 out_bargs:
4756 	kfree(bargs);
4757 out_unlock:
4758 	mutex_unlock(&fs_info->balance_mutex);
4759 	mutex_unlock(&fs_info->volume_mutex);
4760 	if (need_unlock)
4761 		clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4762 out:
4763 	mnt_drop_write_file(file);
4764 	return ret;
4765 }
4766 
4767 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4768 {
4769 	if (!capable(CAP_SYS_ADMIN))
4770 		return -EPERM;
4771 
4772 	switch (cmd) {
4773 	case BTRFS_BALANCE_CTL_PAUSE:
4774 		return btrfs_pause_balance(fs_info);
4775 	case BTRFS_BALANCE_CTL_CANCEL:
4776 		return btrfs_cancel_balance(fs_info);
4777 	}
4778 
4779 	return -EINVAL;
4780 }
4781 
4782 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4783 					 void __user *arg)
4784 {
4785 	struct btrfs_ioctl_balance_args *bargs;
4786 	int ret = 0;
4787 
4788 	if (!capable(CAP_SYS_ADMIN))
4789 		return -EPERM;
4790 
4791 	mutex_lock(&fs_info->balance_mutex);
4792 	if (!fs_info->balance_ctl) {
4793 		ret = -ENOTCONN;
4794 		goto out;
4795 	}
4796 
4797 	bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4798 	if (!bargs) {
4799 		ret = -ENOMEM;
4800 		goto out;
4801 	}
4802 
4803 	update_ioctl_balance_args(fs_info, 1, bargs);
4804 
4805 	if (copy_to_user(arg, bargs, sizeof(*bargs)))
4806 		ret = -EFAULT;
4807 
4808 	kfree(bargs);
4809 out:
4810 	mutex_unlock(&fs_info->balance_mutex);
4811 	return ret;
4812 }
4813 
4814 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4815 {
4816 	struct inode *inode = file_inode(file);
4817 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4818 	struct btrfs_ioctl_quota_ctl_args *sa;
4819 	struct btrfs_trans_handle *trans = NULL;
4820 	int ret;
4821 	int err;
4822 
4823 	if (!capable(CAP_SYS_ADMIN))
4824 		return -EPERM;
4825 
4826 	ret = mnt_want_write_file(file);
4827 	if (ret)
4828 		return ret;
4829 
4830 	sa = memdup_user(arg, sizeof(*sa));
4831 	if (IS_ERR(sa)) {
4832 		ret = PTR_ERR(sa);
4833 		goto drop_write;
4834 	}
4835 
4836 	down_write(&fs_info->subvol_sem);
4837 	trans = btrfs_start_transaction(fs_info->tree_root, 2);
4838 	if (IS_ERR(trans)) {
4839 		ret = PTR_ERR(trans);
4840 		goto out;
4841 	}
4842 
4843 	switch (sa->cmd) {
4844 	case BTRFS_QUOTA_CTL_ENABLE:
4845 		ret = btrfs_quota_enable(trans, fs_info);
4846 		break;
4847 	case BTRFS_QUOTA_CTL_DISABLE:
4848 		ret = btrfs_quota_disable(trans, fs_info);
4849 		break;
4850 	default:
4851 		ret = -EINVAL;
4852 		break;
4853 	}
4854 
4855 	err = btrfs_commit_transaction(trans);
4856 	if (err && !ret)
4857 		ret = err;
4858 out:
4859 	kfree(sa);
4860 	up_write(&fs_info->subvol_sem);
4861 drop_write:
4862 	mnt_drop_write_file(file);
4863 	return ret;
4864 }
4865 
4866 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4867 {
4868 	struct inode *inode = file_inode(file);
4869 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4870 	struct btrfs_root *root = BTRFS_I(inode)->root;
4871 	struct btrfs_ioctl_qgroup_assign_args *sa;
4872 	struct btrfs_trans_handle *trans;
4873 	int ret;
4874 	int err;
4875 
4876 	if (!capable(CAP_SYS_ADMIN))
4877 		return -EPERM;
4878 
4879 	ret = mnt_want_write_file(file);
4880 	if (ret)
4881 		return ret;
4882 
4883 	sa = memdup_user(arg, sizeof(*sa));
4884 	if (IS_ERR(sa)) {
4885 		ret = PTR_ERR(sa);
4886 		goto drop_write;
4887 	}
4888 
4889 	trans = btrfs_join_transaction(root);
4890 	if (IS_ERR(trans)) {
4891 		ret = PTR_ERR(trans);
4892 		goto out;
4893 	}
4894 
4895 	if (sa->assign) {
4896 		ret = btrfs_add_qgroup_relation(trans, fs_info,
4897 						sa->src, sa->dst);
4898 	} else {
4899 		ret = btrfs_del_qgroup_relation(trans, fs_info,
4900 						sa->src, sa->dst);
4901 	}
4902 
4903 	/* update qgroup status and info */
4904 	err = btrfs_run_qgroups(trans, fs_info);
4905 	if (err < 0)
4906 		btrfs_handle_fs_error(fs_info, err,
4907 				      "failed to update qgroup status and info");
4908 	err = btrfs_end_transaction(trans);
4909 	if (err && !ret)
4910 		ret = err;
4911 
4912 out:
4913 	kfree(sa);
4914 drop_write:
4915 	mnt_drop_write_file(file);
4916 	return ret;
4917 }
4918 
4919 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4920 {
4921 	struct inode *inode = file_inode(file);
4922 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4923 	struct btrfs_root *root = BTRFS_I(inode)->root;
4924 	struct btrfs_ioctl_qgroup_create_args *sa;
4925 	struct btrfs_trans_handle *trans;
4926 	int ret;
4927 	int err;
4928 
4929 	if (!capable(CAP_SYS_ADMIN))
4930 		return -EPERM;
4931 
4932 	ret = mnt_want_write_file(file);
4933 	if (ret)
4934 		return ret;
4935 
4936 	sa = memdup_user(arg, sizeof(*sa));
4937 	if (IS_ERR(sa)) {
4938 		ret = PTR_ERR(sa);
4939 		goto drop_write;
4940 	}
4941 
4942 	if (!sa->qgroupid) {
4943 		ret = -EINVAL;
4944 		goto out;
4945 	}
4946 
4947 	trans = btrfs_join_transaction(root);
4948 	if (IS_ERR(trans)) {
4949 		ret = PTR_ERR(trans);
4950 		goto out;
4951 	}
4952 
4953 	if (sa->create) {
4954 		ret = btrfs_create_qgroup(trans, fs_info, sa->qgroupid);
4955 	} else {
4956 		ret = btrfs_remove_qgroup(trans, fs_info, sa->qgroupid);
4957 	}
4958 
4959 	err = btrfs_end_transaction(trans);
4960 	if (err && !ret)
4961 		ret = err;
4962 
4963 out:
4964 	kfree(sa);
4965 drop_write:
4966 	mnt_drop_write_file(file);
4967 	return ret;
4968 }
4969 
4970 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4971 {
4972 	struct inode *inode = file_inode(file);
4973 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4974 	struct btrfs_root *root = BTRFS_I(inode)->root;
4975 	struct btrfs_ioctl_qgroup_limit_args *sa;
4976 	struct btrfs_trans_handle *trans;
4977 	int ret;
4978 	int err;
4979 	u64 qgroupid;
4980 
4981 	if (!capable(CAP_SYS_ADMIN))
4982 		return -EPERM;
4983 
4984 	ret = mnt_want_write_file(file);
4985 	if (ret)
4986 		return ret;
4987 
4988 	sa = memdup_user(arg, sizeof(*sa));
4989 	if (IS_ERR(sa)) {
4990 		ret = PTR_ERR(sa);
4991 		goto drop_write;
4992 	}
4993 
4994 	trans = btrfs_join_transaction(root);
4995 	if (IS_ERR(trans)) {
4996 		ret = PTR_ERR(trans);
4997 		goto out;
4998 	}
4999 
5000 	qgroupid = sa->qgroupid;
5001 	if (!qgroupid) {
5002 		/* take the current subvol as qgroup */
5003 		qgroupid = root->root_key.objectid;
5004 	}
5005 
5006 	ret = btrfs_limit_qgroup(trans, fs_info, qgroupid, &sa->lim);
5007 
5008 	err = btrfs_end_transaction(trans);
5009 	if (err && !ret)
5010 		ret = err;
5011 
5012 out:
5013 	kfree(sa);
5014 drop_write:
5015 	mnt_drop_write_file(file);
5016 	return ret;
5017 }
5018 
5019 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
5020 {
5021 	struct inode *inode = file_inode(file);
5022 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5023 	struct btrfs_ioctl_quota_rescan_args *qsa;
5024 	int ret;
5025 
5026 	if (!capable(CAP_SYS_ADMIN))
5027 		return -EPERM;
5028 
5029 	ret = mnt_want_write_file(file);
5030 	if (ret)
5031 		return ret;
5032 
5033 	qsa = memdup_user(arg, sizeof(*qsa));
5034 	if (IS_ERR(qsa)) {
5035 		ret = PTR_ERR(qsa);
5036 		goto drop_write;
5037 	}
5038 
5039 	if (qsa->flags) {
5040 		ret = -EINVAL;
5041 		goto out;
5042 	}
5043 
5044 	ret = btrfs_qgroup_rescan(fs_info);
5045 
5046 out:
5047 	kfree(qsa);
5048 drop_write:
5049 	mnt_drop_write_file(file);
5050 	return ret;
5051 }
5052 
5053 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
5054 {
5055 	struct inode *inode = file_inode(file);
5056 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5057 	struct btrfs_ioctl_quota_rescan_args *qsa;
5058 	int ret = 0;
5059 
5060 	if (!capable(CAP_SYS_ADMIN))
5061 		return -EPERM;
5062 
5063 	qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
5064 	if (!qsa)
5065 		return -ENOMEM;
5066 
5067 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
5068 		qsa->flags = 1;
5069 		qsa->progress = fs_info->qgroup_rescan_progress.objectid;
5070 	}
5071 
5072 	if (copy_to_user(arg, qsa, sizeof(*qsa)))
5073 		ret = -EFAULT;
5074 
5075 	kfree(qsa);
5076 	return ret;
5077 }
5078 
5079 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
5080 {
5081 	struct inode *inode = file_inode(file);
5082 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5083 
5084 	if (!capable(CAP_SYS_ADMIN))
5085 		return -EPERM;
5086 
5087 	return btrfs_qgroup_wait_for_completion(fs_info, true);
5088 }
5089 
5090 static long _btrfs_ioctl_set_received_subvol(struct file *file,
5091 					    struct btrfs_ioctl_received_subvol_args *sa)
5092 {
5093 	struct inode *inode = file_inode(file);
5094 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5095 	struct btrfs_root *root = BTRFS_I(inode)->root;
5096 	struct btrfs_root_item *root_item = &root->root_item;
5097 	struct btrfs_trans_handle *trans;
5098 	struct timespec ct = current_time(inode);
5099 	int ret = 0;
5100 	int received_uuid_changed;
5101 
5102 	if (!inode_owner_or_capable(inode))
5103 		return -EPERM;
5104 
5105 	ret = mnt_want_write_file(file);
5106 	if (ret < 0)
5107 		return ret;
5108 
5109 	down_write(&fs_info->subvol_sem);
5110 
5111 	if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
5112 		ret = -EINVAL;
5113 		goto out;
5114 	}
5115 
5116 	if (btrfs_root_readonly(root)) {
5117 		ret = -EROFS;
5118 		goto out;
5119 	}
5120 
5121 	/*
5122 	 * 1 - root item
5123 	 * 2 - uuid items (received uuid + subvol uuid)
5124 	 */
5125 	trans = btrfs_start_transaction(root, 3);
5126 	if (IS_ERR(trans)) {
5127 		ret = PTR_ERR(trans);
5128 		trans = NULL;
5129 		goto out;
5130 	}
5131 
5132 	sa->rtransid = trans->transid;
5133 	sa->rtime.sec = ct.tv_sec;
5134 	sa->rtime.nsec = ct.tv_nsec;
5135 
5136 	received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5137 				       BTRFS_UUID_SIZE);
5138 	if (received_uuid_changed &&
5139 	    !btrfs_is_empty_uuid(root_item->received_uuid))
5140 		btrfs_uuid_tree_rem(trans, fs_info, root_item->received_uuid,
5141 				    BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5142 				    root->root_key.objectid);
5143 	memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5144 	btrfs_set_root_stransid(root_item, sa->stransid);
5145 	btrfs_set_root_rtransid(root_item, sa->rtransid);
5146 	btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5147 	btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5148 	btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5149 	btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5150 
5151 	ret = btrfs_update_root(trans, fs_info->tree_root,
5152 				&root->root_key, &root->root_item);
5153 	if (ret < 0) {
5154 		btrfs_end_transaction(trans);
5155 		goto out;
5156 	}
5157 	if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5158 		ret = btrfs_uuid_tree_add(trans, fs_info, sa->uuid,
5159 					  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5160 					  root->root_key.objectid);
5161 		if (ret < 0 && ret != -EEXIST) {
5162 			btrfs_abort_transaction(trans, ret);
5163 			goto out;
5164 		}
5165 	}
5166 	ret = btrfs_commit_transaction(trans);
5167 	if (ret < 0) {
5168 		btrfs_abort_transaction(trans, ret);
5169 		goto out;
5170 	}
5171 
5172 out:
5173 	up_write(&fs_info->subvol_sem);
5174 	mnt_drop_write_file(file);
5175 	return ret;
5176 }
5177 
5178 #ifdef CONFIG_64BIT
5179 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5180 						void __user *arg)
5181 {
5182 	struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5183 	struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5184 	int ret = 0;
5185 
5186 	args32 = memdup_user(arg, sizeof(*args32));
5187 	if (IS_ERR(args32))
5188 		return PTR_ERR(args32);
5189 
5190 	args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
5191 	if (!args64) {
5192 		ret = -ENOMEM;
5193 		goto out;
5194 	}
5195 
5196 	memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5197 	args64->stransid = args32->stransid;
5198 	args64->rtransid = args32->rtransid;
5199 	args64->stime.sec = args32->stime.sec;
5200 	args64->stime.nsec = args32->stime.nsec;
5201 	args64->rtime.sec = args32->rtime.sec;
5202 	args64->rtime.nsec = args32->rtime.nsec;
5203 	args64->flags = args32->flags;
5204 
5205 	ret = _btrfs_ioctl_set_received_subvol(file, args64);
5206 	if (ret)
5207 		goto out;
5208 
5209 	memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5210 	args32->stransid = args64->stransid;
5211 	args32->rtransid = args64->rtransid;
5212 	args32->stime.sec = args64->stime.sec;
5213 	args32->stime.nsec = args64->stime.nsec;
5214 	args32->rtime.sec = args64->rtime.sec;
5215 	args32->rtime.nsec = args64->rtime.nsec;
5216 	args32->flags = args64->flags;
5217 
5218 	ret = copy_to_user(arg, args32, sizeof(*args32));
5219 	if (ret)
5220 		ret = -EFAULT;
5221 
5222 out:
5223 	kfree(args32);
5224 	kfree(args64);
5225 	return ret;
5226 }
5227 #endif
5228 
5229 static long btrfs_ioctl_set_received_subvol(struct file *file,
5230 					    void __user *arg)
5231 {
5232 	struct btrfs_ioctl_received_subvol_args *sa = NULL;
5233 	int ret = 0;
5234 
5235 	sa = memdup_user(arg, sizeof(*sa));
5236 	if (IS_ERR(sa))
5237 		return PTR_ERR(sa);
5238 
5239 	ret = _btrfs_ioctl_set_received_subvol(file, sa);
5240 
5241 	if (ret)
5242 		goto out;
5243 
5244 	ret = copy_to_user(arg, sa, sizeof(*sa));
5245 	if (ret)
5246 		ret = -EFAULT;
5247 
5248 out:
5249 	kfree(sa);
5250 	return ret;
5251 }
5252 
5253 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5254 {
5255 	struct inode *inode = file_inode(file);
5256 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5257 	size_t len;
5258 	int ret;
5259 	char label[BTRFS_LABEL_SIZE];
5260 
5261 	spin_lock(&fs_info->super_lock);
5262 	memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5263 	spin_unlock(&fs_info->super_lock);
5264 
5265 	len = strnlen(label, BTRFS_LABEL_SIZE);
5266 
5267 	if (len == BTRFS_LABEL_SIZE) {
5268 		btrfs_warn(fs_info,
5269 			   "label is too long, return the first %zu bytes",
5270 			   --len);
5271 	}
5272 
5273 	ret = copy_to_user(arg, label, len);
5274 
5275 	return ret ? -EFAULT : 0;
5276 }
5277 
5278 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5279 {
5280 	struct inode *inode = file_inode(file);
5281 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5282 	struct btrfs_root *root = BTRFS_I(inode)->root;
5283 	struct btrfs_super_block *super_block = fs_info->super_copy;
5284 	struct btrfs_trans_handle *trans;
5285 	char label[BTRFS_LABEL_SIZE];
5286 	int ret;
5287 
5288 	if (!capable(CAP_SYS_ADMIN))
5289 		return -EPERM;
5290 
5291 	if (copy_from_user(label, arg, sizeof(label)))
5292 		return -EFAULT;
5293 
5294 	if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5295 		btrfs_err(fs_info,
5296 			  "unable to set label with more than %d bytes",
5297 			  BTRFS_LABEL_SIZE - 1);
5298 		return -EINVAL;
5299 	}
5300 
5301 	ret = mnt_want_write_file(file);
5302 	if (ret)
5303 		return ret;
5304 
5305 	trans = btrfs_start_transaction(root, 0);
5306 	if (IS_ERR(trans)) {
5307 		ret = PTR_ERR(trans);
5308 		goto out_unlock;
5309 	}
5310 
5311 	spin_lock(&fs_info->super_lock);
5312 	strcpy(super_block->label, label);
5313 	spin_unlock(&fs_info->super_lock);
5314 	ret = btrfs_commit_transaction(trans);
5315 
5316 out_unlock:
5317 	mnt_drop_write_file(file);
5318 	return ret;
5319 }
5320 
5321 #define INIT_FEATURE_FLAGS(suffix) \
5322 	{ .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5323 	  .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5324 	  .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5325 
5326 int btrfs_ioctl_get_supported_features(void __user *arg)
5327 {
5328 	static const struct btrfs_ioctl_feature_flags features[3] = {
5329 		INIT_FEATURE_FLAGS(SUPP),
5330 		INIT_FEATURE_FLAGS(SAFE_SET),
5331 		INIT_FEATURE_FLAGS(SAFE_CLEAR)
5332 	};
5333 
5334 	if (copy_to_user(arg, &features, sizeof(features)))
5335 		return -EFAULT;
5336 
5337 	return 0;
5338 }
5339 
5340 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5341 {
5342 	struct inode *inode = file_inode(file);
5343 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5344 	struct btrfs_super_block *super_block = fs_info->super_copy;
5345 	struct btrfs_ioctl_feature_flags features;
5346 
5347 	features.compat_flags = btrfs_super_compat_flags(super_block);
5348 	features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5349 	features.incompat_flags = btrfs_super_incompat_flags(super_block);
5350 
5351 	if (copy_to_user(arg, &features, sizeof(features)))
5352 		return -EFAULT;
5353 
5354 	return 0;
5355 }
5356 
5357 static int check_feature_bits(struct btrfs_fs_info *fs_info,
5358 			      enum btrfs_feature_set set,
5359 			      u64 change_mask, u64 flags, u64 supported_flags,
5360 			      u64 safe_set, u64 safe_clear)
5361 {
5362 	const char *type = btrfs_feature_set_names[set];
5363 	char *names;
5364 	u64 disallowed, unsupported;
5365 	u64 set_mask = flags & change_mask;
5366 	u64 clear_mask = ~flags & change_mask;
5367 
5368 	unsupported = set_mask & ~supported_flags;
5369 	if (unsupported) {
5370 		names = btrfs_printable_features(set, unsupported);
5371 		if (names) {
5372 			btrfs_warn(fs_info,
5373 				   "this kernel does not support the %s feature bit%s",
5374 				   names, strchr(names, ',') ? "s" : "");
5375 			kfree(names);
5376 		} else
5377 			btrfs_warn(fs_info,
5378 				   "this kernel does not support %s bits 0x%llx",
5379 				   type, unsupported);
5380 		return -EOPNOTSUPP;
5381 	}
5382 
5383 	disallowed = set_mask & ~safe_set;
5384 	if (disallowed) {
5385 		names = btrfs_printable_features(set, disallowed);
5386 		if (names) {
5387 			btrfs_warn(fs_info,
5388 				   "can't set the %s feature bit%s while mounted",
5389 				   names, strchr(names, ',') ? "s" : "");
5390 			kfree(names);
5391 		} else
5392 			btrfs_warn(fs_info,
5393 				   "can't set %s bits 0x%llx while mounted",
5394 				   type, disallowed);
5395 		return -EPERM;
5396 	}
5397 
5398 	disallowed = clear_mask & ~safe_clear;
5399 	if (disallowed) {
5400 		names = btrfs_printable_features(set, disallowed);
5401 		if (names) {
5402 			btrfs_warn(fs_info,
5403 				   "can't clear the %s feature bit%s while mounted",
5404 				   names, strchr(names, ',') ? "s" : "");
5405 			kfree(names);
5406 		} else
5407 			btrfs_warn(fs_info,
5408 				   "can't clear %s bits 0x%llx while mounted",
5409 				   type, disallowed);
5410 		return -EPERM;
5411 	}
5412 
5413 	return 0;
5414 }
5415 
5416 #define check_feature(fs_info, change_mask, flags, mask_base)	\
5417 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,	\
5418 		   BTRFS_FEATURE_ ## mask_base ## _SUPP,	\
5419 		   BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,	\
5420 		   BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5421 
5422 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5423 {
5424 	struct inode *inode = file_inode(file);
5425 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5426 	struct btrfs_root *root = BTRFS_I(inode)->root;
5427 	struct btrfs_super_block *super_block = fs_info->super_copy;
5428 	struct btrfs_ioctl_feature_flags flags[2];
5429 	struct btrfs_trans_handle *trans;
5430 	u64 newflags;
5431 	int ret;
5432 
5433 	if (!capable(CAP_SYS_ADMIN))
5434 		return -EPERM;
5435 
5436 	if (copy_from_user(flags, arg, sizeof(flags)))
5437 		return -EFAULT;
5438 
5439 	/* Nothing to do */
5440 	if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5441 	    !flags[0].incompat_flags)
5442 		return 0;
5443 
5444 	ret = check_feature(fs_info, flags[0].compat_flags,
5445 			    flags[1].compat_flags, COMPAT);
5446 	if (ret)
5447 		return ret;
5448 
5449 	ret = check_feature(fs_info, flags[0].compat_ro_flags,
5450 			    flags[1].compat_ro_flags, COMPAT_RO);
5451 	if (ret)
5452 		return ret;
5453 
5454 	ret = check_feature(fs_info, flags[0].incompat_flags,
5455 			    flags[1].incompat_flags, INCOMPAT);
5456 	if (ret)
5457 		return ret;
5458 
5459 	ret = mnt_want_write_file(file);
5460 	if (ret)
5461 		return ret;
5462 
5463 	trans = btrfs_start_transaction(root, 0);
5464 	if (IS_ERR(trans)) {
5465 		ret = PTR_ERR(trans);
5466 		goto out_drop_write;
5467 	}
5468 
5469 	spin_lock(&fs_info->super_lock);
5470 	newflags = btrfs_super_compat_flags(super_block);
5471 	newflags |= flags[0].compat_flags & flags[1].compat_flags;
5472 	newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5473 	btrfs_set_super_compat_flags(super_block, newflags);
5474 
5475 	newflags = btrfs_super_compat_ro_flags(super_block);
5476 	newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5477 	newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5478 	btrfs_set_super_compat_ro_flags(super_block, newflags);
5479 
5480 	newflags = btrfs_super_incompat_flags(super_block);
5481 	newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5482 	newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5483 	btrfs_set_super_incompat_flags(super_block, newflags);
5484 	spin_unlock(&fs_info->super_lock);
5485 
5486 	ret = btrfs_commit_transaction(trans);
5487 out_drop_write:
5488 	mnt_drop_write_file(file);
5489 
5490 	return ret;
5491 }
5492 
5493 long btrfs_ioctl(struct file *file, unsigned int
5494 		cmd, unsigned long arg)
5495 {
5496 	struct inode *inode = file_inode(file);
5497 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5498 	struct btrfs_root *root = BTRFS_I(inode)->root;
5499 	void __user *argp = (void __user *)arg;
5500 
5501 	switch (cmd) {
5502 	case FS_IOC_GETFLAGS:
5503 		return btrfs_ioctl_getflags(file, argp);
5504 	case FS_IOC_SETFLAGS:
5505 		return btrfs_ioctl_setflags(file, argp);
5506 	case FS_IOC_GETVERSION:
5507 		return btrfs_ioctl_getversion(file, argp);
5508 	case FITRIM:
5509 		return btrfs_ioctl_fitrim(file, argp);
5510 	case BTRFS_IOC_SNAP_CREATE:
5511 		return btrfs_ioctl_snap_create(file, argp, 0);
5512 	case BTRFS_IOC_SNAP_CREATE_V2:
5513 		return btrfs_ioctl_snap_create_v2(file, argp, 0);
5514 	case BTRFS_IOC_SUBVOL_CREATE:
5515 		return btrfs_ioctl_snap_create(file, argp, 1);
5516 	case BTRFS_IOC_SUBVOL_CREATE_V2:
5517 		return btrfs_ioctl_snap_create_v2(file, argp, 1);
5518 	case BTRFS_IOC_SNAP_DESTROY:
5519 		return btrfs_ioctl_snap_destroy(file, argp);
5520 	case BTRFS_IOC_SUBVOL_GETFLAGS:
5521 		return btrfs_ioctl_subvol_getflags(file, argp);
5522 	case BTRFS_IOC_SUBVOL_SETFLAGS:
5523 		return btrfs_ioctl_subvol_setflags(file, argp);
5524 	case BTRFS_IOC_DEFAULT_SUBVOL:
5525 		return btrfs_ioctl_default_subvol(file, argp);
5526 	case BTRFS_IOC_DEFRAG:
5527 		return btrfs_ioctl_defrag(file, NULL);
5528 	case BTRFS_IOC_DEFRAG_RANGE:
5529 		return btrfs_ioctl_defrag(file, argp);
5530 	case BTRFS_IOC_RESIZE:
5531 		return btrfs_ioctl_resize(file, argp);
5532 	case BTRFS_IOC_ADD_DEV:
5533 		return btrfs_ioctl_add_dev(fs_info, argp);
5534 	case BTRFS_IOC_RM_DEV:
5535 		return btrfs_ioctl_rm_dev(file, argp);
5536 	case BTRFS_IOC_RM_DEV_V2:
5537 		return btrfs_ioctl_rm_dev_v2(file, argp);
5538 	case BTRFS_IOC_FS_INFO:
5539 		return btrfs_ioctl_fs_info(fs_info, argp);
5540 	case BTRFS_IOC_DEV_INFO:
5541 		return btrfs_ioctl_dev_info(fs_info, argp);
5542 	case BTRFS_IOC_BALANCE:
5543 		return btrfs_ioctl_balance(file, NULL);
5544 	case BTRFS_IOC_TRANS_START:
5545 		return btrfs_ioctl_trans_start(file);
5546 	case BTRFS_IOC_TRANS_END:
5547 		return btrfs_ioctl_trans_end(file);
5548 	case BTRFS_IOC_TREE_SEARCH:
5549 		return btrfs_ioctl_tree_search(file, argp);
5550 	case BTRFS_IOC_TREE_SEARCH_V2:
5551 		return btrfs_ioctl_tree_search_v2(file, argp);
5552 	case BTRFS_IOC_INO_LOOKUP:
5553 		return btrfs_ioctl_ino_lookup(file, argp);
5554 	case BTRFS_IOC_INO_PATHS:
5555 		return btrfs_ioctl_ino_to_path(root, argp);
5556 	case BTRFS_IOC_LOGICAL_INO:
5557 		return btrfs_ioctl_logical_to_ino(fs_info, argp);
5558 	case BTRFS_IOC_SPACE_INFO:
5559 		return btrfs_ioctl_space_info(fs_info, argp);
5560 	case BTRFS_IOC_SYNC: {
5561 		int ret;
5562 
5563 		ret = btrfs_start_delalloc_roots(fs_info, 0, -1);
5564 		if (ret)
5565 			return ret;
5566 		ret = btrfs_sync_fs(inode->i_sb, 1);
5567 		/*
5568 		 * The transaction thread may want to do more work,
5569 		 * namely it pokes the cleaner kthread that will start
5570 		 * processing uncleaned subvols.
5571 		 */
5572 		wake_up_process(fs_info->transaction_kthread);
5573 		return ret;
5574 	}
5575 	case BTRFS_IOC_START_SYNC:
5576 		return btrfs_ioctl_start_sync(root, argp);
5577 	case BTRFS_IOC_WAIT_SYNC:
5578 		return btrfs_ioctl_wait_sync(fs_info, argp);
5579 	case BTRFS_IOC_SCRUB:
5580 		return btrfs_ioctl_scrub(file, argp);
5581 	case BTRFS_IOC_SCRUB_CANCEL:
5582 		return btrfs_ioctl_scrub_cancel(fs_info);
5583 	case BTRFS_IOC_SCRUB_PROGRESS:
5584 		return btrfs_ioctl_scrub_progress(fs_info, argp);
5585 	case BTRFS_IOC_BALANCE_V2:
5586 		return btrfs_ioctl_balance(file, argp);
5587 	case BTRFS_IOC_BALANCE_CTL:
5588 		return btrfs_ioctl_balance_ctl(fs_info, arg);
5589 	case BTRFS_IOC_BALANCE_PROGRESS:
5590 		return btrfs_ioctl_balance_progress(fs_info, argp);
5591 	case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5592 		return btrfs_ioctl_set_received_subvol(file, argp);
5593 #ifdef CONFIG_64BIT
5594 	case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5595 		return btrfs_ioctl_set_received_subvol_32(file, argp);
5596 #endif
5597 	case BTRFS_IOC_SEND:
5598 		return btrfs_ioctl_send(file, argp);
5599 	case BTRFS_IOC_GET_DEV_STATS:
5600 		return btrfs_ioctl_get_dev_stats(fs_info, argp);
5601 	case BTRFS_IOC_QUOTA_CTL:
5602 		return btrfs_ioctl_quota_ctl(file, argp);
5603 	case BTRFS_IOC_QGROUP_ASSIGN:
5604 		return btrfs_ioctl_qgroup_assign(file, argp);
5605 	case BTRFS_IOC_QGROUP_CREATE:
5606 		return btrfs_ioctl_qgroup_create(file, argp);
5607 	case BTRFS_IOC_QGROUP_LIMIT:
5608 		return btrfs_ioctl_qgroup_limit(file, argp);
5609 	case BTRFS_IOC_QUOTA_RESCAN:
5610 		return btrfs_ioctl_quota_rescan(file, argp);
5611 	case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5612 		return btrfs_ioctl_quota_rescan_status(file, argp);
5613 	case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5614 		return btrfs_ioctl_quota_rescan_wait(file, argp);
5615 	case BTRFS_IOC_DEV_REPLACE:
5616 		return btrfs_ioctl_dev_replace(fs_info, argp);
5617 	case BTRFS_IOC_GET_FSLABEL:
5618 		return btrfs_ioctl_get_fslabel(file, argp);
5619 	case BTRFS_IOC_SET_FSLABEL:
5620 		return btrfs_ioctl_set_fslabel(file, argp);
5621 	case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5622 		return btrfs_ioctl_get_supported_features(argp);
5623 	case BTRFS_IOC_GET_FEATURES:
5624 		return btrfs_ioctl_get_features(file, argp);
5625 	case BTRFS_IOC_SET_FEATURES:
5626 		return btrfs_ioctl_set_features(file, argp);
5627 	}
5628 
5629 	return -ENOTTY;
5630 }
5631 
5632 #ifdef CONFIG_COMPAT
5633 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5634 {
5635 	/*
5636 	 * These all access 32-bit values anyway so no further
5637 	 * handling is necessary.
5638 	 */
5639 	switch (cmd) {
5640 	case FS_IOC32_GETFLAGS:
5641 		cmd = FS_IOC_GETFLAGS;
5642 		break;
5643 	case FS_IOC32_SETFLAGS:
5644 		cmd = FS_IOC_SETFLAGS;
5645 		break;
5646 	case FS_IOC32_GETVERSION:
5647 		cmd = FS_IOC_GETVERSION;
5648 		break;
5649 	}
5650 
5651 	return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5652 }
5653 #endif
5654