xref: /linux/fs/btrfs/ioctl.c (revision a4cc96d1f0170b779c32c6b2cc58764f5d2cdef0)
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/fsnotify.h>
25 #include <linux/pagemap.h>
26 #include <linux/highmem.h>
27 #include <linux/time.h>
28 #include <linux/init.h>
29 #include <linux/string.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mount.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/swap.h>
35 #include <linux/writeback.h>
36 #include <linux/statfs.h>
37 #include <linux/compat.h>
38 #include <linux/bit_spinlock.h>
39 #include <linux/security.h>
40 #include <linux/xattr.h>
41 #include <linux/vmalloc.h>
42 #include <linux/slab.h>
43 #include <linux/blkdev.h>
44 #include <linux/uuid.h>
45 #include <linux/btrfs.h>
46 #include <linux/uaccess.h>
47 #include "ctree.h"
48 #include "disk-io.h"
49 #include "transaction.h"
50 #include "btrfs_inode.h"
51 #include "print-tree.h"
52 #include "volumes.h"
53 #include "locking.h"
54 #include "inode-map.h"
55 #include "backref.h"
56 #include "rcu-string.h"
57 #include "send.h"
58 #include "dev-replace.h"
59 #include "props.h"
60 #include "sysfs.h"
61 #include "qgroup.h"
62 #include "tree-log.h"
63 #include "compression.h"
64 
65 #ifdef CONFIG_64BIT
66 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
67  * structures are incorrect, as the timespec structure from userspace
68  * is 4 bytes too small. We define these alternatives here to teach
69  * the kernel about the 32-bit struct packing.
70  */
71 struct btrfs_ioctl_timespec_32 {
72 	__u64 sec;
73 	__u32 nsec;
74 } __attribute__ ((__packed__));
75 
76 struct btrfs_ioctl_received_subvol_args_32 {
77 	char	uuid[BTRFS_UUID_SIZE];	/* in */
78 	__u64	stransid;		/* in */
79 	__u64	rtransid;		/* out */
80 	struct btrfs_ioctl_timespec_32 stime; /* in */
81 	struct btrfs_ioctl_timespec_32 rtime; /* out */
82 	__u64	flags;			/* in */
83 	__u64	reserved[16];		/* in */
84 } __attribute__ ((__packed__));
85 
86 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
87 				struct btrfs_ioctl_received_subvol_args_32)
88 #endif
89 
90 
91 static int btrfs_clone(struct inode *src, struct inode *inode,
92 		       u64 off, u64 olen, u64 olen_aligned, u64 destoff,
93 		       int no_time_update);
94 
95 /* Mask out flags that are inappropriate for the given type of inode. */
96 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
97 {
98 	if (S_ISDIR(mode))
99 		return flags;
100 	else if (S_ISREG(mode))
101 		return flags & ~FS_DIRSYNC_FL;
102 	else
103 		return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
104 }
105 
106 /*
107  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
108  */
109 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
110 {
111 	unsigned int iflags = 0;
112 
113 	if (flags & BTRFS_INODE_SYNC)
114 		iflags |= FS_SYNC_FL;
115 	if (flags & BTRFS_INODE_IMMUTABLE)
116 		iflags |= FS_IMMUTABLE_FL;
117 	if (flags & BTRFS_INODE_APPEND)
118 		iflags |= FS_APPEND_FL;
119 	if (flags & BTRFS_INODE_NODUMP)
120 		iflags |= FS_NODUMP_FL;
121 	if (flags & BTRFS_INODE_NOATIME)
122 		iflags |= FS_NOATIME_FL;
123 	if (flags & BTRFS_INODE_DIRSYNC)
124 		iflags |= FS_DIRSYNC_FL;
125 	if (flags & BTRFS_INODE_NODATACOW)
126 		iflags |= FS_NOCOW_FL;
127 
128 	if (flags & BTRFS_INODE_NOCOMPRESS)
129 		iflags |= FS_NOCOMP_FL;
130 	else if (flags & BTRFS_INODE_COMPRESS)
131 		iflags |= FS_COMPR_FL;
132 
133 	return iflags;
134 }
135 
136 /*
137  * Update inode->i_flags based on the btrfs internal flags.
138  */
139 void btrfs_update_iflags(struct inode *inode)
140 {
141 	struct btrfs_inode *ip = BTRFS_I(inode);
142 	unsigned int new_fl = 0;
143 
144 	if (ip->flags & BTRFS_INODE_SYNC)
145 		new_fl |= S_SYNC;
146 	if (ip->flags & BTRFS_INODE_IMMUTABLE)
147 		new_fl |= S_IMMUTABLE;
148 	if (ip->flags & BTRFS_INODE_APPEND)
149 		new_fl |= S_APPEND;
150 	if (ip->flags & BTRFS_INODE_NOATIME)
151 		new_fl |= S_NOATIME;
152 	if (ip->flags & BTRFS_INODE_DIRSYNC)
153 		new_fl |= S_DIRSYNC;
154 
155 	set_mask_bits(&inode->i_flags,
156 		      S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
157 		      new_fl);
158 }
159 
160 /*
161  * Inherit flags from the parent inode.
162  *
163  * Currently only the compression flags and the cow flags are inherited.
164  */
165 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
166 {
167 	unsigned int flags;
168 
169 	if (!dir)
170 		return;
171 
172 	flags = BTRFS_I(dir)->flags;
173 
174 	if (flags & BTRFS_INODE_NOCOMPRESS) {
175 		BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
176 		BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
177 	} else if (flags & BTRFS_INODE_COMPRESS) {
178 		BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
179 		BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
180 	}
181 
182 	if (flags & BTRFS_INODE_NODATACOW) {
183 		BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
184 		if (S_ISREG(inode->i_mode))
185 			BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
186 	}
187 
188 	btrfs_update_iflags(inode);
189 }
190 
191 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
192 {
193 	struct btrfs_inode *ip = BTRFS_I(file_inode(file));
194 	unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
195 
196 	if (copy_to_user(arg, &flags, sizeof(flags)))
197 		return -EFAULT;
198 	return 0;
199 }
200 
201 static int check_flags(unsigned int flags)
202 {
203 	if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
204 		      FS_NOATIME_FL | FS_NODUMP_FL | \
205 		      FS_SYNC_FL | FS_DIRSYNC_FL | \
206 		      FS_NOCOMP_FL | FS_COMPR_FL |
207 		      FS_NOCOW_FL))
208 		return -EOPNOTSUPP;
209 
210 	if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
211 		return -EINVAL;
212 
213 	return 0;
214 }
215 
216 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
217 {
218 	struct inode *inode = file_inode(file);
219 	struct btrfs_inode *ip = BTRFS_I(inode);
220 	struct btrfs_root *root = ip->root;
221 	struct btrfs_trans_handle *trans;
222 	unsigned int flags, oldflags;
223 	int ret;
224 	u64 ip_oldflags;
225 	unsigned int i_oldflags;
226 	umode_t mode;
227 
228 	if (!inode_owner_or_capable(inode))
229 		return -EPERM;
230 
231 	if (btrfs_root_readonly(root))
232 		return -EROFS;
233 
234 	if (copy_from_user(&flags, arg, sizeof(flags)))
235 		return -EFAULT;
236 
237 	ret = check_flags(flags);
238 	if (ret)
239 		return ret;
240 
241 	ret = mnt_want_write_file(file);
242 	if (ret)
243 		return ret;
244 
245 	inode_lock(inode);
246 
247 	ip_oldflags = ip->flags;
248 	i_oldflags = inode->i_flags;
249 	mode = inode->i_mode;
250 
251 	flags = btrfs_mask_flags(inode->i_mode, flags);
252 	oldflags = btrfs_flags_to_ioctl(ip->flags);
253 	if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
254 		if (!capable(CAP_LINUX_IMMUTABLE)) {
255 			ret = -EPERM;
256 			goto out_unlock;
257 		}
258 	}
259 
260 	if (flags & FS_SYNC_FL)
261 		ip->flags |= BTRFS_INODE_SYNC;
262 	else
263 		ip->flags &= ~BTRFS_INODE_SYNC;
264 	if (flags & FS_IMMUTABLE_FL)
265 		ip->flags |= BTRFS_INODE_IMMUTABLE;
266 	else
267 		ip->flags &= ~BTRFS_INODE_IMMUTABLE;
268 	if (flags & FS_APPEND_FL)
269 		ip->flags |= BTRFS_INODE_APPEND;
270 	else
271 		ip->flags &= ~BTRFS_INODE_APPEND;
272 	if (flags & FS_NODUMP_FL)
273 		ip->flags |= BTRFS_INODE_NODUMP;
274 	else
275 		ip->flags &= ~BTRFS_INODE_NODUMP;
276 	if (flags & FS_NOATIME_FL)
277 		ip->flags |= BTRFS_INODE_NOATIME;
278 	else
279 		ip->flags &= ~BTRFS_INODE_NOATIME;
280 	if (flags & FS_DIRSYNC_FL)
281 		ip->flags |= BTRFS_INODE_DIRSYNC;
282 	else
283 		ip->flags &= ~BTRFS_INODE_DIRSYNC;
284 	if (flags & FS_NOCOW_FL) {
285 		if (S_ISREG(mode)) {
286 			/*
287 			 * It's safe to turn csums off here, no extents exist.
288 			 * Otherwise we want the flag to reflect the real COW
289 			 * status of the file and will not set it.
290 			 */
291 			if (inode->i_size == 0)
292 				ip->flags |= BTRFS_INODE_NODATACOW
293 					   | BTRFS_INODE_NODATASUM;
294 		} else {
295 			ip->flags |= BTRFS_INODE_NODATACOW;
296 		}
297 	} else {
298 		/*
299 		 * Revert back under same assumptions as above
300 		 */
301 		if (S_ISREG(mode)) {
302 			if (inode->i_size == 0)
303 				ip->flags &= ~(BTRFS_INODE_NODATACOW
304 				             | BTRFS_INODE_NODATASUM);
305 		} else {
306 			ip->flags &= ~BTRFS_INODE_NODATACOW;
307 		}
308 	}
309 
310 	/*
311 	 * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
312 	 * flag may be changed automatically if compression code won't make
313 	 * things smaller.
314 	 */
315 	if (flags & FS_NOCOMP_FL) {
316 		ip->flags &= ~BTRFS_INODE_COMPRESS;
317 		ip->flags |= BTRFS_INODE_NOCOMPRESS;
318 
319 		ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
320 		if (ret && ret != -ENODATA)
321 			goto out_drop;
322 	} else if (flags & FS_COMPR_FL) {
323 		const char *comp;
324 
325 		ip->flags |= BTRFS_INODE_COMPRESS;
326 		ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
327 
328 		if (root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
329 			comp = "lzo";
330 		else
331 			comp = "zlib";
332 		ret = btrfs_set_prop(inode, "btrfs.compression",
333 				     comp, strlen(comp), 0);
334 		if (ret)
335 			goto out_drop;
336 
337 	} else {
338 		ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
339 		if (ret && ret != -ENODATA)
340 			goto out_drop;
341 		ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
342 	}
343 
344 	trans = btrfs_start_transaction(root, 1);
345 	if (IS_ERR(trans)) {
346 		ret = PTR_ERR(trans);
347 		goto out_drop;
348 	}
349 
350 	btrfs_update_iflags(inode);
351 	inode_inc_iversion(inode);
352 	inode->i_ctime = current_fs_time(inode->i_sb);
353 	ret = btrfs_update_inode(trans, root, inode);
354 
355 	btrfs_end_transaction(trans, root);
356  out_drop:
357 	if (ret) {
358 		ip->flags = ip_oldflags;
359 		inode->i_flags = i_oldflags;
360 	}
361 
362  out_unlock:
363 	inode_unlock(inode);
364 	mnt_drop_write_file(file);
365 	return ret;
366 }
367 
368 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
369 {
370 	struct inode *inode = file_inode(file);
371 
372 	return put_user(inode->i_generation, arg);
373 }
374 
375 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
376 {
377 	struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
378 	struct btrfs_device *device;
379 	struct request_queue *q;
380 	struct fstrim_range range;
381 	u64 minlen = ULLONG_MAX;
382 	u64 num_devices = 0;
383 	u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
384 	int ret;
385 
386 	if (!capable(CAP_SYS_ADMIN))
387 		return -EPERM;
388 
389 	rcu_read_lock();
390 	list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
391 				dev_list) {
392 		if (!device->bdev)
393 			continue;
394 		q = bdev_get_queue(device->bdev);
395 		if (blk_queue_discard(q)) {
396 			num_devices++;
397 			minlen = min((u64)q->limits.discard_granularity,
398 				     minlen);
399 		}
400 	}
401 	rcu_read_unlock();
402 
403 	if (!num_devices)
404 		return -EOPNOTSUPP;
405 	if (copy_from_user(&range, arg, sizeof(range)))
406 		return -EFAULT;
407 	if (range.start > total_bytes ||
408 	    range.len < fs_info->sb->s_blocksize)
409 		return -EINVAL;
410 
411 	range.len = min(range.len, total_bytes - range.start);
412 	range.minlen = max(range.minlen, minlen);
413 	ret = btrfs_trim_fs(fs_info->tree_root, &range);
414 	if (ret < 0)
415 		return ret;
416 
417 	if (copy_to_user(arg, &range, sizeof(range)))
418 		return -EFAULT;
419 
420 	return 0;
421 }
422 
423 int btrfs_is_empty_uuid(u8 *uuid)
424 {
425 	int i;
426 
427 	for (i = 0; i < BTRFS_UUID_SIZE; i++) {
428 		if (uuid[i])
429 			return 0;
430 	}
431 	return 1;
432 }
433 
434 static noinline int create_subvol(struct inode *dir,
435 				  struct dentry *dentry,
436 				  char *name, int namelen,
437 				  u64 *async_transid,
438 				  struct btrfs_qgroup_inherit *inherit)
439 {
440 	struct btrfs_trans_handle *trans;
441 	struct btrfs_key key;
442 	struct btrfs_root_item *root_item;
443 	struct btrfs_inode_item *inode_item;
444 	struct extent_buffer *leaf;
445 	struct btrfs_root *root = BTRFS_I(dir)->root;
446 	struct btrfs_root *new_root;
447 	struct btrfs_block_rsv block_rsv;
448 	struct timespec cur_time = current_fs_time(dir->i_sb);
449 	struct inode *inode;
450 	int ret;
451 	int err;
452 	u64 objectid;
453 	u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
454 	u64 index = 0;
455 	u64 qgroup_reserved;
456 	uuid_le new_uuid;
457 
458 	root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
459 	if (!root_item)
460 		return -ENOMEM;
461 
462 	ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
463 	if (ret)
464 		goto fail_free;
465 
466 	/*
467 	 * Don't create subvolume whose level is not zero. Or qgroup will be
468 	 * screwed up since it assumes subvolume qgroup's level to be 0.
469 	 */
470 	if (btrfs_qgroup_level(objectid)) {
471 		ret = -ENOSPC;
472 		goto fail_free;
473 	}
474 
475 	btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
476 	/*
477 	 * The same as the snapshot creation, please see the comment
478 	 * of create_snapshot().
479 	 */
480 	ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
481 					       8, &qgroup_reserved, false);
482 	if (ret)
483 		goto fail_free;
484 
485 	trans = btrfs_start_transaction(root, 0);
486 	if (IS_ERR(trans)) {
487 		ret = PTR_ERR(trans);
488 		btrfs_subvolume_release_metadata(root, &block_rsv,
489 						 qgroup_reserved);
490 		goto fail_free;
491 	}
492 	trans->block_rsv = &block_rsv;
493 	trans->bytes_reserved = block_rsv.size;
494 
495 	ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, inherit);
496 	if (ret)
497 		goto fail;
498 
499 	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
500 	if (IS_ERR(leaf)) {
501 		ret = PTR_ERR(leaf);
502 		goto fail;
503 	}
504 
505 	memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
506 	btrfs_set_header_bytenr(leaf, leaf->start);
507 	btrfs_set_header_generation(leaf, trans->transid);
508 	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
509 	btrfs_set_header_owner(leaf, objectid);
510 
511 	write_extent_buffer(leaf, root->fs_info->fsid, btrfs_header_fsid(),
512 			    BTRFS_FSID_SIZE);
513 	write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
514 			    btrfs_header_chunk_tree_uuid(leaf),
515 			    BTRFS_UUID_SIZE);
516 	btrfs_mark_buffer_dirty(leaf);
517 
518 	inode_item = &root_item->inode;
519 	btrfs_set_stack_inode_generation(inode_item, 1);
520 	btrfs_set_stack_inode_size(inode_item, 3);
521 	btrfs_set_stack_inode_nlink(inode_item, 1);
522 	btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
523 	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
524 
525 	btrfs_set_root_flags(root_item, 0);
526 	btrfs_set_root_limit(root_item, 0);
527 	btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
528 
529 	btrfs_set_root_bytenr(root_item, leaf->start);
530 	btrfs_set_root_generation(root_item, trans->transid);
531 	btrfs_set_root_level(root_item, 0);
532 	btrfs_set_root_refs(root_item, 1);
533 	btrfs_set_root_used(root_item, leaf->len);
534 	btrfs_set_root_last_snapshot(root_item, 0);
535 
536 	btrfs_set_root_generation_v2(root_item,
537 			btrfs_root_generation(root_item));
538 	uuid_le_gen(&new_uuid);
539 	memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
540 	btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
541 	btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
542 	root_item->ctime = root_item->otime;
543 	btrfs_set_root_ctransid(root_item, trans->transid);
544 	btrfs_set_root_otransid(root_item, trans->transid);
545 
546 	btrfs_tree_unlock(leaf);
547 	free_extent_buffer(leaf);
548 	leaf = NULL;
549 
550 	btrfs_set_root_dirid(root_item, new_dirid);
551 
552 	key.objectid = objectid;
553 	key.offset = 0;
554 	key.type = BTRFS_ROOT_ITEM_KEY;
555 	ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
556 				root_item);
557 	if (ret)
558 		goto fail;
559 
560 	key.offset = (u64)-1;
561 	new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
562 	if (IS_ERR(new_root)) {
563 		ret = PTR_ERR(new_root);
564 		btrfs_abort_transaction(trans, ret);
565 		goto fail;
566 	}
567 
568 	btrfs_record_root_in_trans(trans, new_root);
569 
570 	ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
571 	if (ret) {
572 		/* We potentially lose an unused inode item here */
573 		btrfs_abort_transaction(trans, ret);
574 		goto fail;
575 	}
576 
577 	mutex_lock(&new_root->objectid_mutex);
578 	new_root->highest_objectid = new_dirid;
579 	mutex_unlock(&new_root->objectid_mutex);
580 
581 	/*
582 	 * insert the directory item
583 	 */
584 	ret = btrfs_set_inode_index(dir, &index);
585 	if (ret) {
586 		btrfs_abort_transaction(trans, ret);
587 		goto fail;
588 	}
589 
590 	ret = btrfs_insert_dir_item(trans, root,
591 				    name, namelen, dir, &key,
592 				    BTRFS_FT_DIR, index);
593 	if (ret) {
594 		btrfs_abort_transaction(trans, ret);
595 		goto fail;
596 	}
597 
598 	btrfs_i_size_write(dir, dir->i_size + namelen * 2);
599 	ret = btrfs_update_inode(trans, root, dir);
600 	BUG_ON(ret);
601 
602 	ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
603 				 objectid, root->root_key.objectid,
604 				 btrfs_ino(dir), index, name, namelen);
605 	BUG_ON(ret);
606 
607 	ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
608 				  root_item->uuid, BTRFS_UUID_KEY_SUBVOL,
609 				  objectid);
610 	if (ret)
611 		btrfs_abort_transaction(trans, ret);
612 
613 fail:
614 	kfree(root_item);
615 	trans->block_rsv = NULL;
616 	trans->bytes_reserved = 0;
617 	btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
618 
619 	if (async_transid) {
620 		*async_transid = trans->transid;
621 		err = btrfs_commit_transaction_async(trans, root, 1);
622 		if (err)
623 			err = btrfs_commit_transaction(trans, root);
624 	} else {
625 		err = btrfs_commit_transaction(trans, root);
626 	}
627 	if (err && !ret)
628 		ret = err;
629 
630 	if (!ret) {
631 		inode = btrfs_lookup_dentry(dir, dentry);
632 		if (IS_ERR(inode))
633 			return PTR_ERR(inode);
634 		d_instantiate(dentry, inode);
635 	}
636 	return ret;
637 
638 fail_free:
639 	kfree(root_item);
640 	return ret;
641 }
642 
643 static void btrfs_wait_for_no_snapshoting_writes(struct btrfs_root *root)
644 {
645 	s64 writers;
646 	DEFINE_WAIT(wait);
647 
648 	do {
649 		prepare_to_wait(&root->subv_writers->wait, &wait,
650 				TASK_UNINTERRUPTIBLE);
651 
652 		writers = percpu_counter_sum(&root->subv_writers->counter);
653 		if (writers)
654 			schedule();
655 
656 		finish_wait(&root->subv_writers->wait, &wait);
657 	} while (writers);
658 }
659 
660 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
661 			   struct dentry *dentry, char *name, int namelen,
662 			   u64 *async_transid, bool readonly,
663 			   struct btrfs_qgroup_inherit *inherit)
664 {
665 	struct inode *inode;
666 	struct btrfs_pending_snapshot *pending_snapshot;
667 	struct btrfs_trans_handle *trans;
668 	int ret;
669 
670 	if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
671 		return -EINVAL;
672 
673 	pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
674 	if (!pending_snapshot)
675 		return -ENOMEM;
676 
677 	pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
678 			GFP_NOFS);
679 	pending_snapshot->path = btrfs_alloc_path();
680 	if (!pending_snapshot->root_item || !pending_snapshot->path) {
681 		ret = -ENOMEM;
682 		goto free_pending;
683 	}
684 
685 	atomic_inc(&root->will_be_snapshoted);
686 	smp_mb__after_atomic();
687 	btrfs_wait_for_no_snapshoting_writes(root);
688 
689 	ret = btrfs_start_delalloc_inodes(root, 0);
690 	if (ret)
691 		goto dec_and_free;
692 
693 	btrfs_wait_ordered_extents(root, -1, 0, (u64)-1);
694 
695 	btrfs_init_block_rsv(&pending_snapshot->block_rsv,
696 			     BTRFS_BLOCK_RSV_TEMP);
697 	/*
698 	 * 1 - parent dir inode
699 	 * 2 - dir entries
700 	 * 1 - root item
701 	 * 2 - root ref/backref
702 	 * 1 - root of snapshot
703 	 * 1 - UUID item
704 	 */
705 	ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
706 					&pending_snapshot->block_rsv, 8,
707 					&pending_snapshot->qgroup_reserved,
708 					false);
709 	if (ret)
710 		goto dec_and_free;
711 
712 	pending_snapshot->dentry = dentry;
713 	pending_snapshot->root = root;
714 	pending_snapshot->readonly = readonly;
715 	pending_snapshot->dir = dir;
716 	pending_snapshot->inherit = inherit;
717 
718 	trans = btrfs_start_transaction(root, 0);
719 	if (IS_ERR(trans)) {
720 		ret = PTR_ERR(trans);
721 		goto fail;
722 	}
723 
724 	spin_lock(&root->fs_info->trans_lock);
725 	list_add(&pending_snapshot->list,
726 		 &trans->transaction->pending_snapshots);
727 	spin_unlock(&root->fs_info->trans_lock);
728 	if (async_transid) {
729 		*async_transid = trans->transid;
730 		ret = btrfs_commit_transaction_async(trans,
731 				     root->fs_info->extent_root, 1);
732 		if (ret)
733 			ret = btrfs_commit_transaction(trans, root);
734 	} else {
735 		ret = btrfs_commit_transaction(trans,
736 					       root->fs_info->extent_root);
737 	}
738 	if (ret)
739 		goto fail;
740 
741 	ret = pending_snapshot->error;
742 	if (ret)
743 		goto fail;
744 
745 	ret = btrfs_orphan_cleanup(pending_snapshot->snap);
746 	if (ret)
747 		goto fail;
748 
749 	inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
750 	if (IS_ERR(inode)) {
751 		ret = PTR_ERR(inode);
752 		goto fail;
753 	}
754 
755 	d_instantiate(dentry, inode);
756 	ret = 0;
757 fail:
758 	btrfs_subvolume_release_metadata(BTRFS_I(dir)->root,
759 					 &pending_snapshot->block_rsv,
760 					 pending_snapshot->qgroup_reserved);
761 dec_and_free:
762 	if (atomic_dec_and_test(&root->will_be_snapshoted))
763 		wake_up_atomic_t(&root->will_be_snapshoted);
764 free_pending:
765 	kfree(pending_snapshot->root_item);
766 	btrfs_free_path(pending_snapshot->path);
767 	kfree(pending_snapshot);
768 
769 	return ret;
770 }
771 
772 /*  copy of may_delete in fs/namei.c()
773  *	Check whether we can remove a link victim from directory dir, check
774  *  whether the type of victim is right.
775  *  1. We can't do it if dir is read-only (done in permission())
776  *  2. We should have write and exec permissions on dir
777  *  3. We can't remove anything from append-only dir
778  *  4. We can't do anything with immutable dir (done in permission())
779  *  5. If the sticky bit on dir is set we should either
780  *	a. be owner of dir, or
781  *	b. be owner of victim, or
782  *	c. have CAP_FOWNER capability
783  *  6. If the victim is append-only or immutable we can't do anything with
784  *     links pointing to it.
785  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
786  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
787  *  9. We can't remove a root or mountpoint.
788  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
789  *     nfs_async_unlink().
790  */
791 
792 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
793 {
794 	int error;
795 
796 	if (d_really_is_negative(victim))
797 		return -ENOENT;
798 
799 	BUG_ON(d_inode(victim->d_parent) != dir);
800 	audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
801 
802 	error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
803 	if (error)
804 		return error;
805 	if (IS_APPEND(dir))
806 		return -EPERM;
807 	if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
808 	    IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
809 		return -EPERM;
810 	if (isdir) {
811 		if (!d_is_dir(victim))
812 			return -ENOTDIR;
813 		if (IS_ROOT(victim))
814 			return -EBUSY;
815 	} else if (d_is_dir(victim))
816 		return -EISDIR;
817 	if (IS_DEADDIR(dir))
818 		return -ENOENT;
819 	if (victim->d_flags & DCACHE_NFSFS_RENAMED)
820 		return -EBUSY;
821 	return 0;
822 }
823 
824 /* copy of may_create in fs/namei.c() */
825 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
826 {
827 	if (d_really_is_positive(child))
828 		return -EEXIST;
829 	if (IS_DEADDIR(dir))
830 		return -ENOENT;
831 	return inode_permission(dir, MAY_WRITE | MAY_EXEC);
832 }
833 
834 /*
835  * Create a new subvolume below @parent.  This is largely modeled after
836  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
837  * inside this filesystem so it's quite a bit simpler.
838  */
839 static noinline int btrfs_mksubvol(struct path *parent,
840 				   char *name, int namelen,
841 				   struct btrfs_root *snap_src,
842 				   u64 *async_transid, bool readonly,
843 				   struct btrfs_qgroup_inherit *inherit)
844 {
845 	struct inode *dir  = d_inode(parent->dentry);
846 	struct dentry *dentry;
847 	int error;
848 
849 	error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
850 	if (error == -EINTR)
851 		return error;
852 
853 	dentry = lookup_one_len(name, parent->dentry, namelen);
854 	error = PTR_ERR(dentry);
855 	if (IS_ERR(dentry))
856 		goto out_unlock;
857 
858 	error = btrfs_may_create(dir, dentry);
859 	if (error)
860 		goto out_dput;
861 
862 	/*
863 	 * even if this name doesn't exist, we may get hash collisions.
864 	 * check for them now when we can safely fail
865 	 */
866 	error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
867 					       dir->i_ino, name,
868 					       namelen);
869 	if (error)
870 		goto out_dput;
871 
872 	down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
873 
874 	if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
875 		goto out_up_read;
876 
877 	if (snap_src) {
878 		error = create_snapshot(snap_src, dir, dentry, name, namelen,
879 					async_transid, readonly, inherit);
880 	} else {
881 		error = create_subvol(dir, dentry, name, namelen,
882 				      async_transid, inherit);
883 	}
884 	if (!error)
885 		fsnotify_mkdir(dir, dentry);
886 out_up_read:
887 	up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
888 out_dput:
889 	dput(dentry);
890 out_unlock:
891 	inode_unlock(dir);
892 	return error;
893 }
894 
895 /*
896  * When we're defragging a range, we don't want to kick it off again
897  * if it is really just waiting for delalloc to send it down.
898  * If we find a nice big extent or delalloc range for the bytes in the
899  * file you want to defrag, we return 0 to let you know to skip this
900  * part of the file
901  */
902 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
903 {
904 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
905 	struct extent_map *em = NULL;
906 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
907 	u64 end;
908 
909 	read_lock(&em_tree->lock);
910 	em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
911 	read_unlock(&em_tree->lock);
912 
913 	if (em) {
914 		end = extent_map_end(em);
915 		free_extent_map(em);
916 		if (end - offset > thresh)
917 			return 0;
918 	}
919 	/* if we already have a nice delalloc here, just stop */
920 	thresh /= 2;
921 	end = count_range_bits(io_tree, &offset, offset + thresh,
922 			       thresh, EXTENT_DELALLOC, 1);
923 	if (end >= thresh)
924 		return 0;
925 	return 1;
926 }
927 
928 /*
929  * helper function to walk through a file and find extents
930  * newer than a specific transid, and smaller than thresh.
931  *
932  * This is used by the defragging code to find new and small
933  * extents
934  */
935 static int find_new_extents(struct btrfs_root *root,
936 			    struct inode *inode, u64 newer_than,
937 			    u64 *off, u32 thresh)
938 {
939 	struct btrfs_path *path;
940 	struct btrfs_key min_key;
941 	struct extent_buffer *leaf;
942 	struct btrfs_file_extent_item *extent;
943 	int type;
944 	int ret;
945 	u64 ino = btrfs_ino(inode);
946 
947 	path = btrfs_alloc_path();
948 	if (!path)
949 		return -ENOMEM;
950 
951 	min_key.objectid = ino;
952 	min_key.type = BTRFS_EXTENT_DATA_KEY;
953 	min_key.offset = *off;
954 
955 	while (1) {
956 		ret = btrfs_search_forward(root, &min_key, path, newer_than);
957 		if (ret != 0)
958 			goto none;
959 process_slot:
960 		if (min_key.objectid != ino)
961 			goto none;
962 		if (min_key.type != BTRFS_EXTENT_DATA_KEY)
963 			goto none;
964 
965 		leaf = path->nodes[0];
966 		extent = btrfs_item_ptr(leaf, path->slots[0],
967 					struct btrfs_file_extent_item);
968 
969 		type = btrfs_file_extent_type(leaf, extent);
970 		if (type == BTRFS_FILE_EXTENT_REG &&
971 		    btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
972 		    check_defrag_in_cache(inode, min_key.offset, thresh)) {
973 			*off = min_key.offset;
974 			btrfs_free_path(path);
975 			return 0;
976 		}
977 
978 		path->slots[0]++;
979 		if (path->slots[0] < btrfs_header_nritems(leaf)) {
980 			btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
981 			goto process_slot;
982 		}
983 
984 		if (min_key.offset == (u64)-1)
985 			goto none;
986 
987 		min_key.offset++;
988 		btrfs_release_path(path);
989 	}
990 none:
991 	btrfs_free_path(path);
992 	return -ENOENT;
993 }
994 
995 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
996 {
997 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
998 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
999 	struct extent_map *em;
1000 	u64 len = PAGE_SIZE;
1001 
1002 	/*
1003 	 * hopefully we have this extent in the tree already, try without
1004 	 * the full extent lock
1005 	 */
1006 	read_lock(&em_tree->lock);
1007 	em = lookup_extent_mapping(em_tree, start, len);
1008 	read_unlock(&em_tree->lock);
1009 
1010 	if (!em) {
1011 		struct extent_state *cached = NULL;
1012 		u64 end = start + len - 1;
1013 
1014 		/* get the big lock and read metadata off disk */
1015 		lock_extent_bits(io_tree, start, end, &cached);
1016 		em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
1017 		unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
1018 
1019 		if (IS_ERR(em))
1020 			return NULL;
1021 	}
1022 
1023 	return em;
1024 }
1025 
1026 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1027 {
1028 	struct extent_map *next;
1029 	bool ret = true;
1030 
1031 	/* this is the last extent */
1032 	if (em->start + em->len >= i_size_read(inode))
1033 		return false;
1034 
1035 	next = defrag_lookup_extent(inode, em->start + em->len);
1036 	if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1037 		ret = false;
1038 	else if ((em->block_start + em->block_len == next->block_start) &&
1039 		 (em->block_len > SZ_128K && next->block_len > SZ_128K))
1040 		ret = false;
1041 
1042 	free_extent_map(next);
1043 	return ret;
1044 }
1045 
1046 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1047 			       u64 *last_len, u64 *skip, u64 *defrag_end,
1048 			       int compress)
1049 {
1050 	struct extent_map *em;
1051 	int ret = 1;
1052 	bool next_mergeable = true;
1053 	bool prev_mergeable = true;
1054 
1055 	/*
1056 	 * make sure that once we start defragging an extent, we keep on
1057 	 * defragging it
1058 	 */
1059 	if (start < *defrag_end)
1060 		return 1;
1061 
1062 	*skip = 0;
1063 
1064 	em = defrag_lookup_extent(inode, start);
1065 	if (!em)
1066 		return 0;
1067 
1068 	/* this will cover holes, and inline extents */
1069 	if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1070 		ret = 0;
1071 		goto out;
1072 	}
1073 
1074 	if (!*defrag_end)
1075 		prev_mergeable = false;
1076 
1077 	next_mergeable = defrag_check_next_extent(inode, em);
1078 	/*
1079 	 * we hit a real extent, if it is big or the next extent is not a
1080 	 * real extent, don't bother defragging it
1081 	 */
1082 	if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1083 	    (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1084 		ret = 0;
1085 out:
1086 	/*
1087 	 * last_len ends up being a counter of how many bytes we've defragged.
1088 	 * every time we choose not to defrag an extent, we reset *last_len
1089 	 * so that the next tiny extent will force a defrag.
1090 	 *
1091 	 * The end result of this is that tiny extents before a single big
1092 	 * extent will force at least part of that big extent to be defragged.
1093 	 */
1094 	if (ret) {
1095 		*defrag_end = extent_map_end(em);
1096 	} else {
1097 		*last_len = 0;
1098 		*skip = extent_map_end(em);
1099 		*defrag_end = 0;
1100 	}
1101 
1102 	free_extent_map(em);
1103 	return ret;
1104 }
1105 
1106 /*
1107  * it doesn't do much good to defrag one or two pages
1108  * at a time.  This pulls in a nice chunk of pages
1109  * to COW and defrag.
1110  *
1111  * It also makes sure the delalloc code has enough
1112  * dirty data to avoid making new small extents as part
1113  * of the defrag
1114  *
1115  * It's a good idea to start RA on this range
1116  * before calling this.
1117  */
1118 static int cluster_pages_for_defrag(struct inode *inode,
1119 				    struct page **pages,
1120 				    unsigned long start_index,
1121 				    unsigned long num_pages)
1122 {
1123 	unsigned long file_end;
1124 	u64 isize = i_size_read(inode);
1125 	u64 page_start;
1126 	u64 page_end;
1127 	u64 page_cnt;
1128 	int ret;
1129 	int i;
1130 	int i_done;
1131 	struct btrfs_ordered_extent *ordered;
1132 	struct extent_state *cached_state = NULL;
1133 	struct extent_io_tree *tree;
1134 	gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1135 
1136 	file_end = (isize - 1) >> PAGE_SHIFT;
1137 	if (!isize || start_index > file_end)
1138 		return 0;
1139 
1140 	page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1141 
1142 	ret = btrfs_delalloc_reserve_space(inode,
1143 			start_index << PAGE_SHIFT,
1144 			page_cnt << PAGE_SHIFT);
1145 	if (ret)
1146 		return ret;
1147 	i_done = 0;
1148 	tree = &BTRFS_I(inode)->io_tree;
1149 
1150 	/* step one, lock all the pages */
1151 	for (i = 0; i < page_cnt; i++) {
1152 		struct page *page;
1153 again:
1154 		page = find_or_create_page(inode->i_mapping,
1155 					   start_index + i, mask);
1156 		if (!page)
1157 			break;
1158 
1159 		page_start = page_offset(page);
1160 		page_end = page_start + PAGE_SIZE - 1;
1161 		while (1) {
1162 			lock_extent_bits(tree, page_start, page_end,
1163 					 &cached_state);
1164 			ordered = btrfs_lookup_ordered_extent(inode,
1165 							      page_start);
1166 			unlock_extent_cached(tree, page_start, page_end,
1167 					     &cached_state, GFP_NOFS);
1168 			if (!ordered)
1169 				break;
1170 
1171 			unlock_page(page);
1172 			btrfs_start_ordered_extent(inode, ordered, 1);
1173 			btrfs_put_ordered_extent(ordered);
1174 			lock_page(page);
1175 			/*
1176 			 * we unlocked the page above, so we need check if
1177 			 * it was released or not.
1178 			 */
1179 			if (page->mapping != inode->i_mapping) {
1180 				unlock_page(page);
1181 				put_page(page);
1182 				goto again;
1183 			}
1184 		}
1185 
1186 		if (!PageUptodate(page)) {
1187 			btrfs_readpage(NULL, page);
1188 			lock_page(page);
1189 			if (!PageUptodate(page)) {
1190 				unlock_page(page);
1191 				put_page(page);
1192 				ret = -EIO;
1193 				break;
1194 			}
1195 		}
1196 
1197 		if (page->mapping != inode->i_mapping) {
1198 			unlock_page(page);
1199 			put_page(page);
1200 			goto again;
1201 		}
1202 
1203 		pages[i] = page;
1204 		i_done++;
1205 	}
1206 	if (!i_done || ret)
1207 		goto out;
1208 
1209 	if (!(inode->i_sb->s_flags & MS_ACTIVE))
1210 		goto out;
1211 
1212 	/*
1213 	 * so now we have a nice long stream of locked
1214 	 * and up to date pages, lets wait on them
1215 	 */
1216 	for (i = 0; i < i_done; i++)
1217 		wait_on_page_writeback(pages[i]);
1218 
1219 	page_start = page_offset(pages[0]);
1220 	page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1221 
1222 	lock_extent_bits(&BTRFS_I(inode)->io_tree,
1223 			 page_start, page_end - 1, &cached_state);
1224 	clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1225 			  page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1226 			  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1227 			  &cached_state, GFP_NOFS);
1228 
1229 	if (i_done != page_cnt) {
1230 		spin_lock(&BTRFS_I(inode)->lock);
1231 		BTRFS_I(inode)->outstanding_extents++;
1232 		spin_unlock(&BTRFS_I(inode)->lock);
1233 		btrfs_delalloc_release_space(inode,
1234 				start_index << PAGE_SHIFT,
1235 				(page_cnt - i_done) << PAGE_SHIFT);
1236 	}
1237 
1238 
1239 	set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1240 			  &cached_state);
1241 
1242 	unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1243 			     page_start, page_end - 1, &cached_state,
1244 			     GFP_NOFS);
1245 
1246 	for (i = 0; i < i_done; i++) {
1247 		clear_page_dirty_for_io(pages[i]);
1248 		ClearPageChecked(pages[i]);
1249 		set_page_extent_mapped(pages[i]);
1250 		set_page_dirty(pages[i]);
1251 		unlock_page(pages[i]);
1252 		put_page(pages[i]);
1253 	}
1254 	return i_done;
1255 out:
1256 	for (i = 0; i < i_done; i++) {
1257 		unlock_page(pages[i]);
1258 		put_page(pages[i]);
1259 	}
1260 	btrfs_delalloc_release_space(inode,
1261 			start_index << PAGE_SHIFT,
1262 			page_cnt << PAGE_SHIFT);
1263 	return ret;
1264 
1265 }
1266 
1267 int btrfs_defrag_file(struct inode *inode, struct file *file,
1268 		      struct btrfs_ioctl_defrag_range_args *range,
1269 		      u64 newer_than, unsigned long max_to_defrag)
1270 {
1271 	struct btrfs_root *root = BTRFS_I(inode)->root;
1272 	struct file_ra_state *ra = NULL;
1273 	unsigned long last_index;
1274 	u64 isize = i_size_read(inode);
1275 	u64 last_len = 0;
1276 	u64 skip = 0;
1277 	u64 defrag_end = 0;
1278 	u64 newer_off = range->start;
1279 	unsigned long i;
1280 	unsigned long ra_index = 0;
1281 	int ret;
1282 	int defrag_count = 0;
1283 	int compress_type = BTRFS_COMPRESS_ZLIB;
1284 	u32 extent_thresh = range->extent_thresh;
1285 	unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1286 	unsigned long cluster = max_cluster;
1287 	u64 new_align = ~((u64)SZ_128K - 1);
1288 	struct page **pages = NULL;
1289 
1290 	if (isize == 0)
1291 		return 0;
1292 
1293 	if (range->start >= isize)
1294 		return -EINVAL;
1295 
1296 	if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1297 		if (range->compress_type > BTRFS_COMPRESS_TYPES)
1298 			return -EINVAL;
1299 		if (range->compress_type)
1300 			compress_type = range->compress_type;
1301 	}
1302 
1303 	if (extent_thresh == 0)
1304 		extent_thresh = SZ_256K;
1305 
1306 	/*
1307 	 * if we were not given a file, allocate a readahead
1308 	 * context
1309 	 */
1310 	if (!file) {
1311 		ra = kzalloc(sizeof(*ra), GFP_NOFS);
1312 		if (!ra)
1313 			return -ENOMEM;
1314 		file_ra_state_init(ra, inode->i_mapping);
1315 	} else {
1316 		ra = &file->f_ra;
1317 	}
1318 
1319 	pages = kmalloc_array(max_cluster, sizeof(struct page *),
1320 			GFP_NOFS);
1321 	if (!pages) {
1322 		ret = -ENOMEM;
1323 		goto out_ra;
1324 	}
1325 
1326 	/* find the last page to defrag */
1327 	if (range->start + range->len > range->start) {
1328 		last_index = min_t(u64, isize - 1,
1329 			 range->start + range->len - 1) >> PAGE_SHIFT;
1330 	} else {
1331 		last_index = (isize - 1) >> PAGE_SHIFT;
1332 	}
1333 
1334 	if (newer_than) {
1335 		ret = find_new_extents(root, inode, newer_than,
1336 				       &newer_off, SZ_64K);
1337 		if (!ret) {
1338 			range->start = newer_off;
1339 			/*
1340 			 * we always align our defrag to help keep
1341 			 * the extents in the file evenly spaced
1342 			 */
1343 			i = (newer_off & new_align) >> PAGE_SHIFT;
1344 		} else
1345 			goto out_ra;
1346 	} else {
1347 		i = range->start >> PAGE_SHIFT;
1348 	}
1349 	if (!max_to_defrag)
1350 		max_to_defrag = last_index - i + 1;
1351 
1352 	/*
1353 	 * make writeback starts from i, so the defrag range can be
1354 	 * written sequentially.
1355 	 */
1356 	if (i < inode->i_mapping->writeback_index)
1357 		inode->i_mapping->writeback_index = i;
1358 
1359 	while (i <= last_index && defrag_count < max_to_defrag &&
1360 	       (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1361 		/*
1362 		 * make sure we stop running if someone unmounts
1363 		 * the FS
1364 		 */
1365 		if (!(inode->i_sb->s_flags & MS_ACTIVE))
1366 			break;
1367 
1368 		if (btrfs_defrag_cancelled(root->fs_info)) {
1369 			btrfs_debug(root->fs_info, "defrag_file cancelled");
1370 			ret = -EAGAIN;
1371 			break;
1372 		}
1373 
1374 		if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1375 					 extent_thresh, &last_len, &skip,
1376 					 &defrag_end, range->flags &
1377 					 BTRFS_DEFRAG_RANGE_COMPRESS)) {
1378 			unsigned long next;
1379 			/*
1380 			 * the should_defrag function tells us how much to skip
1381 			 * bump our counter by the suggested amount
1382 			 */
1383 			next = DIV_ROUND_UP(skip, PAGE_SIZE);
1384 			i = max(i + 1, next);
1385 			continue;
1386 		}
1387 
1388 		if (!newer_than) {
1389 			cluster = (PAGE_ALIGN(defrag_end) >>
1390 				   PAGE_SHIFT) - i;
1391 			cluster = min(cluster, max_cluster);
1392 		} else {
1393 			cluster = max_cluster;
1394 		}
1395 
1396 		if (i + cluster > ra_index) {
1397 			ra_index = max(i, ra_index);
1398 			btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1399 				       cluster);
1400 			ra_index += cluster;
1401 		}
1402 
1403 		inode_lock(inode);
1404 		if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1405 			BTRFS_I(inode)->force_compress = compress_type;
1406 		ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1407 		if (ret < 0) {
1408 			inode_unlock(inode);
1409 			goto out_ra;
1410 		}
1411 
1412 		defrag_count += ret;
1413 		balance_dirty_pages_ratelimited(inode->i_mapping);
1414 		inode_unlock(inode);
1415 
1416 		if (newer_than) {
1417 			if (newer_off == (u64)-1)
1418 				break;
1419 
1420 			if (ret > 0)
1421 				i += ret;
1422 
1423 			newer_off = max(newer_off + 1,
1424 					(u64)i << PAGE_SHIFT);
1425 
1426 			ret = find_new_extents(root, inode, newer_than,
1427 					       &newer_off, SZ_64K);
1428 			if (!ret) {
1429 				range->start = newer_off;
1430 				i = (newer_off & new_align) >> PAGE_SHIFT;
1431 			} else {
1432 				break;
1433 			}
1434 		} else {
1435 			if (ret > 0) {
1436 				i += ret;
1437 				last_len += ret << PAGE_SHIFT;
1438 			} else {
1439 				i++;
1440 				last_len = 0;
1441 			}
1442 		}
1443 	}
1444 
1445 	if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1446 		filemap_flush(inode->i_mapping);
1447 		if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1448 			     &BTRFS_I(inode)->runtime_flags))
1449 			filemap_flush(inode->i_mapping);
1450 	}
1451 
1452 	if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1453 		/* the filemap_flush will queue IO into the worker threads, but
1454 		 * we have to make sure the IO is actually started and that
1455 		 * ordered extents get created before we return
1456 		 */
1457 		atomic_inc(&root->fs_info->async_submit_draining);
1458 		while (atomic_read(&root->fs_info->nr_async_submits) ||
1459 		      atomic_read(&root->fs_info->async_delalloc_pages)) {
1460 			wait_event(root->fs_info->async_submit_wait,
1461 			   (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1462 			    atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1463 		}
1464 		atomic_dec(&root->fs_info->async_submit_draining);
1465 	}
1466 
1467 	if (range->compress_type == BTRFS_COMPRESS_LZO) {
1468 		btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1469 	}
1470 
1471 	ret = defrag_count;
1472 
1473 out_ra:
1474 	if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1475 		inode_lock(inode);
1476 		BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1477 		inode_unlock(inode);
1478 	}
1479 	if (!file)
1480 		kfree(ra);
1481 	kfree(pages);
1482 	return ret;
1483 }
1484 
1485 static noinline int btrfs_ioctl_resize(struct file *file,
1486 					void __user *arg)
1487 {
1488 	u64 new_size;
1489 	u64 old_size;
1490 	u64 devid = 1;
1491 	struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1492 	struct btrfs_ioctl_vol_args *vol_args;
1493 	struct btrfs_trans_handle *trans;
1494 	struct btrfs_device *device = NULL;
1495 	char *sizestr;
1496 	char *retptr;
1497 	char *devstr = NULL;
1498 	int ret = 0;
1499 	int mod = 0;
1500 
1501 	if (!capable(CAP_SYS_ADMIN))
1502 		return -EPERM;
1503 
1504 	ret = mnt_want_write_file(file);
1505 	if (ret)
1506 		return ret;
1507 
1508 	if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1509 			1)) {
1510 		mnt_drop_write_file(file);
1511 		return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1512 	}
1513 
1514 	mutex_lock(&root->fs_info->volume_mutex);
1515 	vol_args = memdup_user(arg, sizeof(*vol_args));
1516 	if (IS_ERR(vol_args)) {
1517 		ret = PTR_ERR(vol_args);
1518 		goto out;
1519 	}
1520 
1521 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1522 
1523 	sizestr = vol_args->name;
1524 	devstr = strchr(sizestr, ':');
1525 	if (devstr) {
1526 		sizestr = devstr + 1;
1527 		*devstr = '\0';
1528 		devstr = vol_args->name;
1529 		ret = kstrtoull(devstr, 10, &devid);
1530 		if (ret)
1531 			goto out_free;
1532 		if (!devid) {
1533 			ret = -EINVAL;
1534 			goto out_free;
1535 		}
1536 		btrfs_info(root->fs_info, "resizing devid %llu", devid);
1537 	}
1538 
1539 	device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1540 	if (!device) {
1541 		btrfs_info(root->fs_info, "resizer unable to find device %llu",
1542 		       devid);
1543 		ret = -ENODEV;
1544 		goto out_free;
1545 	}
1546 
1547 	if (!device->writeable) {
1548 		btrfs_info(root->fs_info,
1549 			   "resizer unable to apply on readonly device %llu",
1550 		       devid);
1551 		ret = -EPERM;
1552 		goto out_free;
1553 	}
1554 
1555 	if (!strcmp(sizestr, "max"))
1556 		new_size = device->bdev->bd_inode->i_size;
1557 	else {
1558 		if (sizestr[0] == '-') {
1559 			mod = -1;
1560 			sizestr++;
1561 		} else if (sizestr[0] == '+') {
1562 			mod = 1;
1563 			sizestr++;
1564 		}
1565 		new_size = memparse(sizestr, &retptr);
1566 		if (*retptr != '\0' || new_size == 0) {
1567 			ret = -EINVAL;
1568 			goto out_free;
1569 		}
1570 	}
1571 
1572 	if (device->is_tgtdev_for_dev_replace) {
1573 		ret = -EPERM;
1574 		goto out_free;
1575 	}
1576 
1577 	old_size = btrfs_device_get_total_bytes(device);
1578 
1579 	if (mod < 0) {
1580 		if (new_size > old_size) {
1581 			ret = -EINVAL;
1582 			goto out_free;
1583 		}
1584 		new_size = old_size - new_size;
1585 	} else if (mod > 0) {
1586 		if (new_size > ULLONG_MAX - old_size) {
1587 			ret = -ERANGE;
1588 			goto out_free;
1589 		}
1590 		new_size = old_size + new_size;
1591 	}
1592 
1593 	if (new_size < SZ_256M) {
1594 		ret = -EINVAL;
1595 		goto out_free;
1596 	}
1597 	if (new_size > device->bdev->bd_inode->i_size) {
1598 		ret = -EFBIG;
1599 		goto out_free;
1600 	}
1601 
1602 	new_size = div_u64(new_size, root->sectorsize);
1603 	new_size *= root->sectorsize;
1604 
1605 	btrfs_info_in_rcu(root->fs_info, "new size for %s is %llu",
1606 		      rcu_str_deref(device->name), new_size);
1607 
1608 	if (new_size > old_size) {
1609 		trans = btrfs_start_transaction(root, 0);
1610 		if (IS_ERR(trans)) {
1611 			ret = PTR_ERR(trans);
1612 			goto out_free;
1613 		}
1614 		ret = btrfs_grow_device(trans, device, new_size);
1615 		btrfs_commit_transaction(trans, root);
1616 	} else if (new_size < old_size) {
1617 		ret = btrfs_shrink_device(device, new_size);
1618 	} /* equal, nothing need to do */
1619 
1620 out_free:
1621 	kfree(vol_args);
1622 out:
1623 	mutex_unlock(&root->fs_info->volume_mutex);
1624 	atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1625 	mnt_drop_write_file(file);
1626 	return ret;
1627 }
1628 
1629 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1630 				char *name, unsigned long fd, int subvol,
1631 				u64 *transid, bool readonly,
1632 				struct btrfs_qgroup_inherit *inherit)
1633 {
1634 	int namelen;
1635 	int ret = 0;
1636 
1637 	if (!S_ISDIR(file_inode(file)->i_mode))
1638 		return -ENOTDIR;
1639 
1640 	ret = mnt_want_write_file(file);
1641 	if (ret)
1642 		goto out;
1643 
1644 	namelen = strlen(name);
1645 	if (strchr(name, '/')) {
1646 		ret = -EINVAL;
1647 		goto out_drop_write;
1648 	}
1649 
1650 	if (name[0] == '.' &&
1651 	   (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1652 		ret = -EEXIST;
1653 		goto out_drop_write;
1654 	}
1655 
1656 	if (subvol) {
1657 		ret = btrfs_mksubvol(&file->f_path, name, namelen,
1658 				     NULL, transid, readonly, inherit);
1659 	} else {
1660 		struct fd src = fdget(fd);
1661 		struct inode *src_inode;
1662 		if (!src.file) {
1663 			ret = -EINVAL;
1664 			goto out_drop_write;
1665 		}
1666 
1667 		src_inode = file_inode(src.file);
1668 		if (src_inode->i_sb != file_inode(file)->i_sb) {
1669 			btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1670 				   "Snapshot src from another FS");
1671 			ret = -EXDEV;
1672 		} else if (!inode_owner_or_capable(src_inode)) {
1673 			/*
1674 			 * Subvolume creation is not restricted, but snapshots
1675 			 * are limited to own subvolumes only
1676 			 */
1677 			ret = -EPERM;
1678 		} else {
1679 			ret = btrfs_mksubvol(&file->f_path, name, namelen,
1680 					     BTRFS_I(src_inode)->root,
1681 					     transid, readonly, inherit);
1682 		}
1683 		fdput(src);
1684 	}
1685 out_drop_write:
1686 	mnt_drop_write_file(file);
1687 out:
1688 	return ret;
1689 }
1690 
1691 static noinline int btrfs_ioctl_snap_create(struct file *file,
1692 					    void __user *arg, int subvol)
1693 {
1694 	struct btrfs_ioctl_vol_args *vol_args;
1695 	int ret;
1696 
1697 	if (!S_ISDIR(file_inode(file)->i_mode))
1698 		return -ENOTDIR;
1699 
1700 	vol_args = memdup_user(arg, sizeof(*vol_args));
1701 	if (IS_ERR(vol_args))
1702 		return PTR_ERR(vol_args);
1703 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1704 
1705 	ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1706 					      vol_args->fd, subvol,
1707 					      NULL, false, NULL);
1708 
1709 	kfree(vol_args);
1710 	return ret;
1711 }
1712 
1713 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1714 					       void __user *arg, int subvol)
1715 {
1716 	struct btrfs_ioctl_vol_args_v2 *vol_args;
1717 	int ret;
1718 	u64 transid = 0;
1719 	u64 *ptr = NULL;
1720 	bool readonly = false;
1721 	struct btrfs_qgroup_inherit *inherit = NULL;
1722 
1723 	if (!S_ISDIR(file_inode(file)->i_mode))
1724 		return -ENOTDIR;
1725 
1726 	vol_args = memdup_user(arg, sizeof(*vol_args));
1727 	if (IS_ERR(vol_args))
1728 		return PTR_ERR(vol_args);
1729 	vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1730 
1731 	if (vol_args->flags &
1732 	    ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1733 	      BTRFS_SUBVOL_QGROUP_INHERIT)) {
1734 		ret = -EOPNOTSUPP;
1735 		goto free_args;
1736 	}
1737 
1738 	if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1739 		ptr = &transid;
1740 	if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1741 		readonly = true;
1742 	if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1743 		if (vol_args->size > PAGE_SIZE) {
1744 			ret = -EINVAL;
1745 			goto free_args;
1746 		}
1747 		inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1748 		if (IS_ERR(inherit)) {
1749 			ret = PTR_ERR(inherit);
1750 			goto free_args;
1751 		}
1752 	}
1753 
1754 	ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1755 					      vol_args->fd, subvol, ptr,
1756 					      readonly, inherit);
1757 	if (ret)
1758 		goto free_inherit;
1759 
1760 	if (ptr && copy_to_user(arg +
1761 				offsetof(struct btrfs_ioctl_vol_args_v2,
1762 					transid),
1763 				ptr, sizeof(*ptr)))
1764 		ret = -EFAULT;
1765 
1766 free_inherit:
1767 	kfree(inherit);
1768 free_args:
1769 	kfree(vol_args);
1770 	return ret;
1771 }
1772 
1773 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1774 						void __user *arg)
1775 {
1776 	struct inode *inode = file_inode(file);
1777 	struct btrfs_root *root = BTRFS_I(inode)->root;
1778 	int ret = 0;
1779 	u64 flags = 0;
1780 
1781 	if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1782 		return -EINVAL;
1783 
1784 	down_read(&root->fs_info->subvol_sem);
1785 	if (btrfs_root_readonly(root))
1786 		flags |= BTRFS_SUBVOL_RDONLY;
1787 	up_read(&root->fs_info->subvol_sem);
1788 
1789 	if (copy_to_user(arg, &flags, sizeof(flags)))
1790 		ret = -EFAULT;
1791 
1792 	return ret;
1793 }
1794 
1795 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1796 					      void __user *arg)
1797 {
1798 	struct inode *inode = file_inode(file);
1799 	struct btrfs_root *root = BTRFS_I(inode)->root;
1800 	struct btrfs_trans_handle *trans;
1801 	u64 root_flags;
1802 	u64 flags;
1803 	int ret = 0;
1804 
1805 	if (!inode_owner_or_capable(inode))
1806 		return -EPERM;
1807 
1808 	ret = mnt_want_write_file(file);
1809 	if (ret)
1810 		goto out;
1811 
1812 	if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1813 		ret = -EINVAL;
1814 		goto out_drop_write;
1815 	}
1816 
1817 	if (copy_from_user(&flags, arg, sizeof(flags))) {
1818 		ret = -EFAULT;
1819 		goto out_drop_write;
1820 	}
1821 
1822 	if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1823 		ret = -EINVAL;
1824 		goto out_drop_write;
1825 	}
1826 
1827 	if (flags & ~BTRFS_SUBVOL_RDONLY) {
1828 		ret = -EOPNOTSUPP;
1829 		goto out_drop_write;
1830 	}
1831 
1832 	down_write(&root->fs_info->subvol_sem);
1833 
1834 	/* nothing to do */
1835 	if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1836 		goto out_drop_sem;
1837 
1838 	root_flags = btrfs_root_flags(&root->root_item);
1839 	if (flags & BTRFS_SUBVOL_RDONLY) {
1840 		btrfs_set_root_flags(&root->root_item,
1841 				     root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1842 	} else {
1843 		/*
1844 		 * Block RO -> RW transition if this subvolume is involved in
1845 		 * send
1846 		 */
1847 		spin_lock(&root->root_item_lock);
1848 		if (root->send_in_progress == 0) {
1849 			btrfs_set_root_flags(&root->root_item,
1850 				     root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1851 			spin_unlock(&root->root_item_lock);
1852 		} else {
1853 			spin_unlock(&root->root_item_lock);
1854 			btrfs_warn(root->fs_info,
1855 			"Attempt to set subvolume %llu read-write during send",
1856 					root->root_key.objectid);
1857 			ret = -EPERM;
1858 			goto out_drop_sem;
1859 		}
1860 	}
1861 
1862 	trans = btrfs_start_transaction(root, 1);
1863 	if (IS_ERR(trans)) {
1864 		ret = PTR_ERR(trans);
1865 		goto out_reset;
1866 	}
1867 
1868 	ret = btrfs_update_root(trans, root->fs_info->tree_root,
1869 				&root->root_key, &root->root_item);
1870 
1871 	btrfs_commit_transaction(trans, root);
1872 out_reset:
1873 	if (ret)
1874 		btrfs_set_root_flags(&root->root_item, root_flags);
1875 out_drop_sem:
1876 	up_write(&root->fs_info->subvol_sem);
1877 out_drop_write:
1878 	mnt_drop_write_file(file);
1879 out:
1880 	return ret;
1881 }
1882 
1883 /*
1884  * helper to check if the subvolume references other subvolumes
1885  */
1886 static noinline int may_destroy_subvol(struct btrfs_root *root)
1887 {
1888 	struct btrfs_path *path;
1889 	struct btrfs_dir_item *di;
1890 	struct btrfs_key key;
1891 	u64 dir_id;
1892 	int ret;
1893 
1894 	path = btrfs_alloc_path();
1895 	if (!path)
1896 		return -ENOMEM;
1897 
1898 	/* Make sure this root isn't set as the default subvol */
1899 	dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1900 	di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1901 				   dir_id, "default", 7, 0);
1902 	if (di && !IS_ERR(di)) {
1903 		btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1904 		if (key.objectid == root->root_key.objectid) {
1905 			ret = -EPERM;
1906 			btrfs_err(root->fs_info, "deleting default subvolume "
1907 				  "%llu is not allowed", key.objectid);
1908 			goto out;
1909 		}
1910 		btrfs_release_path(path);
1911 	}
1912 
1913 	key.objectid = root->root_key.objectid;
1914 	key.type = BTRFS_ROOT_REF_KEY;
1915 	key.offset = (u64)-1;
1916 
1917 	ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1918 				&key, path, 0, 0);
1919 	if (ret < 0)
1920 		goto out;
1921 	BUG_ON(ret == 0);
1922 
1923 	ret = 0;
1924 	if (path->slots[0] > 0) {
1925 		path->slots[0]--;
1926 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1927 		if (key.objectid == root->root_key.objectid &&
1928 		    key.type == BTRFS_ROOT_REF_KEY)
1929 			ret = -ENOTEMPTY;
1930 	}
1931 out:
1932 	btrfs_free_path(path);
1933 	return ret;
1934 }
1935 
1936 static noinline int key_in_sk(struct btrfs_key *key,
1937 			      struct btrfs_ioctl_search_key *sk)
1938 {
1939 	struct btrfs_key test;
1940 	int ret;
1941 
1942 	test.objectid = sk->min_objectid;
1943 	test.type = sk->min_type;
1944 	test.offset = sk->min_offset;
1945 
1946 	ret = btrfs_comp_cpu_keys(key, &test);
1947 	if (ret < 0)
1948 		return 0;
1949 
1950 	test.objectid = sk->max_objectid;
1951 	test.type = sk->max_type;
1952 	test.offset = sk->max_offset;
1953 
1954 	ret = btrfs_comp_cpu_keys(key, &test);
1955 	if (ret > 0)
1956 		return 0;
1957 	return 1;
1958 }
1959 
1960 static noinline int copy_to_sk(struct btrfs_path *path,
1961 			       struct btrfs_key *key,
1962 			       struct btrfs_ioctl_search_key *sk,
1963 			       size_t *buf_size,
1964 			       char __user *ubuf,
1965 			       unsigned long *sk_offset,
1966 			       int *num_found)
1967 {
1968 	u64 found_transid;
1969 	struct extent_buffer *leaf;
1970 	struct btrfs_ioctl_search_header sh;
1971 	struct btrfs_key test;
1972 	unsigned long item_off;
1973 	unsigned long item_len;
1974 	int nritems;
1975 	int i;
1976 	int slot;
1977 	int ret = 0;
1978 
1979 	leaf = path->nodes[0];
1980 	slot = path->slots[0];
1981 	nritems = btrfs_header_nritems(leaf);
1982 
1983 	if (btrfs_header_generation(leaf) > sk->max_transid) {
1984 		i = nritems;
1985 		goto advance_key;
1986 	}
1987 	found_transid = btrfs_header_generation(leaf);
1988 
1989 	for (i = slot; i < nritems; i++) {
1990 		item_off = btrfs_item_ptr_offset(leaf, i);
1991 		item_len = btrfs_item_size_nr(leaf, i);
1992 
1993 		btrfs_item_key_to_cpu(leaf, key, i);
1994 		if (!key_in_sk(key, sk))
1995 			continue;
1996 
1997 		if (sizeof(sh) + item_len > *buf_size) {
1998 			if (*num_found) {
1999 				ret = 1;
2000 				goto out;
2001 			}
2002 
2003 			/*
2004 			 * return one empty item back for v1, which does not
2005 			 * handle -EOVERFLOW
2006 			 */
2007 
2008 			*buf_size = sizeof(sh) + item_len;
2009 			item_len = 0;
2010 			ret = -EOVERFLOW;
2011 		}
2012 
2013 		if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2014 			ret = 1;
2015 			goto out;
2016 		}
2017 
2018 		sh.objectid = key->objectid;
2019 		sh.offset = key->offset;
2020 		sh.type = key->type;
2021 		sh.len = item_len;
2022 		sh.transid = found_transid;
2023 
2024 		/* copy search result header */
2025 		if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
2026 			ret = -EFAULT;
2027 			goto out;
2028 		}
2029 
2030 		*sk_offset += sizeof(sh);
2031 
2032 		if (item_len) {
2033 			char __user *up = ubuf + *sk_offset;
2034 			/* copy the item */
2035 			if (read_extent_buffer_to_user(leaf, up,
2036 						       item_off, item_len)) {
2037 				ret = -EFAULT;
2038 				goto out;
2039 			}
2040 
2041 			*sk_offset += item_len;
2042 		}
2043 		(*num_found)++;
2044 
2045 		if (ret) /* -EOVERFLOW from above */
2046 			goto out;
2047 
2048 		if (*num_found >= sk->nr_items) {
2049 			ret = 1;
2050 			goto out;
2051 		}
2052 	}
2053 advance_key:
2054 	ret = 0;
2055 	test.objectid = sk->max_objectid;
2056 	test.type = sk->max_type;
2057 	test.offset = sk->max_offset;
2058 	if (btrfs_comp_cpu_keys(key, &test) >= 0)
2059 		ret = 1;
2060 	else if (key->offset < (u64)-1)
2061 		key->offset++;
2062 	else if (key->type < (u8)-1) {
2063 		key->offset = 0;
2064 		key->type++;
2065 	} else if (key->objectid < (u64)-1) {
2066 		key->offset = 0;
2067 		key->type = 0;
2068 		key->objectid++;
2069 	} else
2070 		ret = 1;
2071 out:
2072 	/*
2073 	 *  0: all items from this leaf copied, continue with next
2074 	 *  1: * more items can be copied, but unused buffer is too small
2075 	 *     * all items were found
2076 	 *     Either way, it will stops the loop which iterates to the next
2077 	 *     leaf
2078 	 *  -EOVERFLOW: item was to large for buffer
2079 	 *  -EFAULT: could not copy extent buffer back to userspace
2080 	 */
2081 	return ret;
2082 }
2083 
2084 static noinline int search_ioctl(struct inode *inode,
2085 				 struct btrfs_ioctl_search_key *sk,
2086 				 size_t *buf_size,
2087 				 char __user *ubuf)
2088 {
2089 	struct btrfs_root *root;
2090 	struct btrfs_key key;
2091 	struct btrfs_path *path;
2092 	struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
2093 	int ret;
2094 	int num_found = 0;
2095 	unsigned long sk_offset = 0;
2096 
2097 	if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2098 		*buf_size = sizeof(struct btrfs_ioctl_search_header);
2099 		return -EOVERFLOW;
2100 	}
2101 
2102 	path = btrfs_alloc_path();
2103 	if (!path)
2104 		return -ENOMEM;
2105 
2106 	if (sk->tree_id == 0) {
2107 		/* search the root of the inode that was passed */
2108 		root = BTRFS_I(inode)->root;
2109 	} else {
2110 		key.objectid = sk->tree_id;
2111 		key.type = BTRFS_ROOT_ITEM_KEY;
2112 		key.offset = (u64)-1;
2113 		root = btrfs_read_fs_root_no_name(info, &key);
2114 		if (IS_ERR(root)) {
2115 			btrfs_free_path(path);
2116 			return -ENOENT;
2117 		}
2118 	}
2119 
2120 	key.objectid = sk->min_objectid;
2121 	key.type = sk->min_type;
2122 	key.offset = sk->min_offset;
2123 
2124 	while (1) {
2125 		ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2126 		if (ret != 0) {
2127 			if (ret > 0)
2128 				ret = 0;
2129 			goto err;
2130 		}
2131 		ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2132 				 &sk_offset, &num_found);
2133 		btrfs_release_path(path);
2134 		if (ret)
2135 			break;
2136 
2137 	}
2138 	if (ret > 0)
2139 		ret = 0;
2140 err:
2141 	sk->nr_items = num_found;
2142 	btrfs_free_path(path);
2143 	return ret;
2144 }
2145 
2146 static noinline int btrfs_ioctl_tree_search(struct file *file,
2147 					   void __user *argp)
2148 {
2149 	struct btrfs_ioctl_search_args __user *uargs;
2150 	struct btrfs_ioctl_search_key sk;
2151 	struct inode *inode;
2152 	int ret;
2153 	size_t buf_size;
2154 
2155 	if (!capable(CAP_SYS_ADMIN))
2156 		return -EPERM;
2157 
2158 	uargs = (struct btrfs_ioctl_search_args __user *)argp;
2159 
2160 	if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2161 		return -EFAULT;
2162 
2163 	buf_size = sizeof(uargs->buf);
2164 
2165 	inode = file_inode(file);
2166 	ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2167 
2168 	/*
2169 	 * In the origin implementation an overflow is handled by returning a
2170 	 * search header with a len of zero, so reset ret.
2171 	 */
2172 	if (ret == -EOVERFLOW)
2173 		ret = 0;
2174 
2175 	if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2176 		ret = -EFAULT;
2177 	return ret;
2178 }
2179 
2180 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2181 					       void __user *argp)
2182 {
2183 	struct btrfs_ioctl_search_args_v2 __user *uarg;
2184 	struct btrfs_ioctl_search_args_v2 args;
2185 	struct inode *inode;
2186 	int ret;
2187 	size_t buf_size;
2188 	const size_t buf_limit = SZ_16M;
2189 
2190 	if (!capable(CAP_SYS_ADMIN))
2191 		return -EPERM;
2192 
2193 	/* copy search header and buffer size */
2194 	uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2195 	if (copy_from_user(&args, uarg, sizeof(args)))
2196 		return -EFAULT;
2197 
2198 	buf_size = args.buf_size;
2199 
2200 	if (buf_size < sizeof(struct btrfs_ioctl_search_header))
2201 		return -EOVERFLOW;
2202 
2203 	/* limit result size to 16MB */
2204 	if (buf_size > buf_limit)
2205 		buf_size = buf_limit;
2206 
2207 	inode = file_inode(file);
2208 	ret = search_ioctl(inode, &args.key, &buf_size,
2209 			   (char *)(&uarg->buf[0]));
2210 	if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2211 		ret = -EFAULT;
2212 	else if (ret == -EOVERFLOW &&
2213 		copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2214 		ret = -EFAULT;
2215 
2216 	return ret;
2217 }
2218 
2219 /*
2220  * Search INODE_REFs to identify path name of 'dirid' directory
2221  * in a 'tree_id' tree. and sets path name to 'name'.
2222  */
2223 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2224 				u64 tree_id, u64 dirid, char *name)
2225 {
2226 	struct btrfs_root *root;
2227 	struct btrfs_key key;
2228 	char *ptr;
2229 	int ret = -1;
2230 	int slot;
2231 	int len;
2232 	int total_len = 0;
2233 	struct btrfs_inode_ref *iref;
2234 	struct extent_buffer *l;
2235 	struct btrfs_path *path;
2236 
2237 	if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2238 		name[0]='\0';
2239 		return 0;
2240 	}
2241 
2242 	path = btrfs_alloc_path();
2243 	if (!path)
2244 		return -ENOMEM;
2245 
2246 	ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2247 
2248 	key.objectid = tree_id;
2249 	key.type = BTRFS_ROOT_ITEM_KEY;
2250 	key.offset = (u64)-1;
2251 	root = btrfs_read_fs_root_no_name(info, &key);
2252 	if (IS_ERR(root)) {
2253 		btrfs_err(info, "could not find root %llu", tree_id);
2254 		ret = -ENOENT;
2255 		goto out;
2256 	}
2257 
2258 	key.objectid = dirid;
2259 	key.type = BTRFS_INODE_REF_KEY;
2260 	key.offset = (u64)-1;
2261 
2262 	while (1) {
2263 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2264 		if (ret < 0)
2265 			goto out;
2266 		else if (ret > 0) {
2267 			ret = btrfs_previous_item(root, path, dirid,
2268 						  BTRFS_INODE_REF_KEY);
2269 			if (ret < 0)
2270 				goto out;
2271 			else if (ret > 0) {
2272 				ret = -ENOENT;
2273 				goto out;
2274 			}
2275 		}
2276 
2277 		l = path->nodes[0];
2278 		slot = path->slots[0];
2279 		btrfs_item_key_to_cpu(l, &key, slot);
2280 
2281 		iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2282 		len = btrfs_inode_ref_name_len(l, iref);
2283 		ptr -= len + 1;
2284 		total_len += len + 1;
2285 		if (ptr < name) {
2286 			ret = -ENAMETOOLONG;
2287 			goto out;
2288 		}
2289 
2290 		*(ptr + len) = '/';
2291 		read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2292 
2293 		if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2294 			break;
2295 
2296 		btrfs_release_path(path);
2297 		key.objectid = key.offset;
2298 		key.offset = (u64)-1;
2299 		dirid = key.objectid;
2300 	}
2301 	memmove(name, ptr, total_len);
2302 	name[total_len] = '\0';
2303 	ret = 0;
2304 out:
2305 	btrfs_free_path(path);
2306 	return ret;
2307 }
2308 
2309 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2310 					   void __user *argp)
2311 {
2312 	 struct btrfs_ioctl_ino_lookup_args *args;
2313 	 struct inode *inode;
2314 	int ret = 0;
2315 
2316 	args = memdup_user(argp, sizeof(*args));
2317 	if (IS_ERR(args))
2318 		return PTR_ERR(args);
2319 
2320 	inode = file_inode(file);
2321 
2322 	/*
2323 	 * Unprivileged query to obtain the containing subvolume root id. The
2324 	 * path is reset so it's consistent with btrfs_search_path_in_tree.
2325 	 */
2326 	if (args->treeid == 0)
2327 		args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2328 
2329 	if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2330 		args->name[0] = 0;
2331 		goto out;
2332 	}
2333 
2334 	if (!capable(CAP_SYS_ADMIN)) {
2335 		ret = -EPERM;
2336 		goto out;
2337 	}
2338 
2339 	ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2340 					args->treeid, args->objectid,
2341 					args->name);
2342 
2343 out:
2344 	if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2345 		ret = -EFAULT;
2346 
2347 	kfree(args);
2348 	return ret;
2349 }
2350 
2351 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2352 					     void __user *arg)
2353 {
2354 	struct dentry *parent = file->f_path.dentry;
2355 	struct dentry *dentry;
2356 	struct inode *dir = d_inode(parent);
2357 	struct inode *inode;
2358 	struct btrfs_root *root = BTRFS_I(dir)->root;
2359 	struct btrfs_root *dest = NULL;
2360 	struct btrfs_ioctl_vol_args *vol_args;
2361 	struct btrfs_trans_handle *trans;
2362 	struct btrfs_block_rsv block_rsv;
2363 	u64 root_flags;
2364 	u64 qgroup_reserved;
2365 	int namelen;
2366 	int ret;
2367 	int err = 0;
2368 
2369 	if (!S_ISDIR(dir->i_mode))
2370 		return -ENOTDIR;
2371 
2372 	vol_args = memdup_user(arg, sizeof(*vol_args));
2373 	if (IS_ERR(vol_args))
2374 		return PTR_ERR(vol_args);
2375 
2376 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2377 	namelen = strlen(vol_args->name);
2378 	if (strchr(vol_args->name, '/') ||
2379 	    strncmp(vol_args->name, "..", namelen) == 0) {
2380 		err = -EINVAL;
2381 		goto out;
2382 	}
2383 
2384 	err = mnt_want_write_file(file);
2385 	if (err)
2386 		goto out;
2387 
2388 
2389 	err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2390 	if (err == -EINTR)
2391 		goto out_drop_write;
2392 	dentry = lookup_one_len(vol_args->name, parent, namelen);
2393 	if (IS_ERR(dentry)) {
2394 		err = PTR_ERR(dentry);
2395 		goto out_unlock_dir;
2396 	}
2397 
2398 	if (d_really_is_negative(dentry)) {
2399 		err = -ENOENT;
2400 		goto out_dput;
2401 	}
2402 
2403 	inode = d_inode(dentry);
2404 	dest = BTRFS_I(inode)->root;
2405 	if (!capable(CAP_SYS_ADMIN)) {
2406 		/*
2407 		 * Regular user.  Only allow this with a special mount
2408 		 * option, when the user has write+exec access to the
2409 		 * subvol root, and when rmdir(2) would have been
2410 		 * allowed.
2411 		 *
2412 		 * Note that this is _not_ check that the subvol is
2413 		 * empty or doesn't contain data that we wouldn't
2414 		 * otherwise be able to delete.
2415 		 *
2416 		 * Users who want to delete empty subvols should try
2417 		 * rmdir(2).
2418 		 */
2419 		err = -EPERM;
2420 		if (!btrfs_test_opt(root->fs_info, USER_SUBVOL_RM_ALLOWED))
2421 			goto out_dput;
2422 
2423 		/*
2424 		 * Do not allow deletion if the parent dir is the same
2425 		 * as the dir to be deleted.  That means the ioctl
2426 		 * must be called on the dentry referencing the root
2427 		 * of the subvol, not a random directory contained
2428 		 * within it.
2429 		 */
2430 		err = -EINVAL;
2431 		if (root == dest)
2432 			goto out_dput;
2433 
2434 		err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2435 		if (err)
2436 			goto out_dput;
2437 	}
2438 
2439 	/* check if subvolume may be deleted by a user */
2440 	err = btrfs_may_delete(dir, dentry, 1);
2441 	if (err)
2442 		goto out_dput;
2443 
2444 	if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2445 		err = -EINVAL;
2446 		goto out_dput;
2447 	}
2448 
2449 	inode_lock(inode);
2450 
2451 	/*
2452 	 * Don't allow to delete a subvolume with send in progress. This is
2453 	 * inside the i_mutex so the error handling that has to drop the bit
2454 	 * again is not run concurrently.
2455 	 */
2456 	spin_lock(&dest->root_item_lock);
2457 	root_flags = btrfs_root_flags(&dest->root_item);
2458 	if (dest->send_in_progress == 0) {
2459 		btrfs_set_root_flags(&dest->root_item,
2460 				root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2461 		spin_unlock(&dest->root_item_lock);
2462 	} else {
2463 		spin_unlock(&dest->root_item_lock);
2464 		btrfs_warn(root->fs_info,
2465 			"Attempt to delete subvolume %llu during send",
2466 			dest->root_key.objectid);
2467 		err = -EPERM;
2468 		goto out_unlock_inode;
2469 	}
2470 
2471 	down_write(&root->fs_info->subvol_sem);
2472 
2473 	err = may_destroy_subvol(dest);
2474 	if (err)
2475 		goto out_up_write;
2476 
2477 	btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2478 	/*
2479 	 * One for dir inode, two for dir entries, two for root
2480 	 * ref/backref.
2481 	 */
2482 	err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2483 					       5, &qgroup_reserved, true);
2484 	if (err)
2485 		goto out_up_write;
2486 
2487 	trans = btrfs_start_transaction(root, 0);
2488 	if (IS_ERR(trans)) {
2489 		err = PTR_ERR(trans);
2490 		goto out_release;
2491 	}
2492 	trans->block_rsv = &block_rsv;
2493 	trans->bytes_reserved = block_rsv.size;
2494 
2495 	btrfs_record_snapshot_destroy(trans, dir);
2496 
2497 	ret = btrfs_unlink_subvol(trans, root, dir,
2498 				dest->root_key.objectid,
2499 				dentry->d_name.name,
2500 				dentry->d_name.len);
2501 	if (ret) {
2502 		err = ret;
2503 		btrfs_abort_transaction(trans, ret);
2504 		goto out_end_trans;
2505 	}
2506 
2507 	btrfs_record_root_in_trans(trans, dest);
2508 
2509 	memset(&dest->root_item.drop_progress, 0,
2510 		sizeof(dest->root_item.drop_progress));
2511 	dest->root_item.drop_level = 0;
2512 	btrfs_set_root_refs(&dest->root_item, 0);
2513 
2514 	if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2515 		ret = btrfs_insert_orphan_item(trans,
2516 					root->fs_info->tree_root,
2517 					dest->root_key.objectid);
2518 		if (ret) {
2519 			btrfs_abort_transaction(trans, ret);
2520 			err = ret;
2521 			goto out_end_trans;
2522 		}
2523 	}
2524 
2525 	ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2526 				  dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2527 				  dest->root_key.objectid);
2528 	if (ret && ret != -ENOENT) {
2529 		btrfs_abort_transaction(trans, ret);
2530 		err = ret;
2531 		goto out_end_trans;
2532 	}
2533 	if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2534 		ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2535 					  dest->root_item.received_uuid,
2536 					  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2537 					  dest->root_key.objectid);
2538 		if (ret && ret != -ENOENT) {
2539 			btrfs_abort_transaction(trans, ret);
2540 			err = ret;
2541 			goto out_end_trans;
2542 		}
2543 	}
2544 
2545 out_end_trans:
2546 	trans->block_rsv = NULL;
2547 	trans->bytes_reserved = 0;
2548 	ret = btrfs_end_transaction(trans, root);
2549 	if (ret && !err)
2550 		err = ret;
2551 	inode->i_flags |= S_DEAD;
2552 out_release:
2553 	btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2554 out_up_write:
2555 	up_write(&root->fs_info->subvol_sem);
2556 	if (err) {
2557 		spin_lock(&dest->root_item_lock);
2558 		root_flags = btrfs_root_flags(&dest->root_item);
2559 		btrfs_set_root_flags(&dest->root_item,
2560 				root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2561 		spin_unlock(&dest->root_item_lock);
2562 	}
2563 out_unlock_inode:
2564 	inode_unlock(inode);
2565 	if (!err) {
2566 		d_invalidate(dentry);
2567 		btrfs_invalidate_inodes(dest);
2568 		d_delete(dentry);
2569 		ASSERT(dest->send_in_progress == 0);
2570 
2571 		/* the last ref */
2572 		if (dest->ino_cache_inode) {
2573 			iput(dest->ino_cache_inode);
2574 			dest->ino_cache_inode = NULL;
2575 		}
2576 	}
2577 out_dput:
2578 	dput(dentry);
2579 out_unlock_dir:
2580 	inode_unlock(dir);
2581 out_drop_write:
2582 	mnt_drop_write_file(file);
2583 out:
2584 	kfree(vol_args);
2585 	return err;
2586 }
2587 
2588 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2589 {
2590 	struct inode *inode = file_inode(file);
2591 	struct btrfs_root *root = BTRFS_I(inode)->root;
2592 	struct btrfs_ioctl_defrag_range_args *range;
2593 	int ret;
2594 
2595 	ret = mnt_want_write_file(file);
2596 	if (ret)
2597 		return ret;
2598 
2599 	if (btrfs_root_readonly(root)) {
2600 		ret = -EROFS;
2601 		goto out;
2602 	}
2603 
2604 	switch (inode->i_mode & S_IFMT) {
2605 	case S_IFDIR:
2606 		if (!capable(CAP_SYS_ADMIN)) {
2607 			ret = -EPERM;
2608 			goto out;
2609 		}
2610 		ret = btrfs_defrag_root(root);
2611 		if (ret)
2612 			goto out;
2613 		ret = btrfs_defrag_root(root->fs_info->extent_root);
2614 		break;
2615 	case S_IFREG:
2616 		if (!(file->f_mode & FMODE_WRITE)) {
2617 			ret = -EINVAL;
2618 			goto out;
2619 		}
2620 
2621 		range = kzalloc(sizeof(*range), GFP_KERNEL);
2622 		if (!range) {
2623 			ret = -ENOMEM;
2624 			goto out;
2625 		}
2626 
2627 		if (argp) {
2628 			if (copy_from_user(range, argp,
2629 					   sizeof(*range))) {
2630 				ret = -EFAULT;
2631 				kfree(range);
2632 				goto out;
2633 			}
2634 			/* compression requires us to start the IO */
2635 			if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2636 				range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2637 				range->extent_thresh = (u32)-1;
2638 			}
2639 		} else {
2640 			/* the rest are all set to zero by kzalloc */
2641 			range->len = (u64)-1;
2642 		}
2643 		ret = btrfs_defrag_file(file_inode(file), file,
2644 					range, 0, 0);
2645 		if (ret > 0)
2646 			ret = 0;
2647 		kfree(range);
2648 		break;
2649 	default:
2650 		ret = -EINVAL;
2651 	}
2652 out:
2653 	mnt_drop_write_file(file);
2654 	return ret;
2655 }
2656 
2657 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2658 {
2659 	struct btrfs_ioctl_vol_args *vol_args;
2660 	int ret;
2661 
2662 	if (!capable(CAP_SYS_ADMIN))
2663 		return -EPERM;
2664 
2665 	if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2666 			1)) {
2667 		return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2668 	}
2669 
2670 	mutex_lock(&root->fs_info->volume_mutex);
2671 	vol_args = memdup_user(arg, sizeof(*vol_args));
2672 	if (IS_ERR(vol_args)) {
2673 		ret = PTR_ERR(vol_args);
2674 		goto out;
2675 	}
2676 
2677 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2678 	ret = btrfs_init_new_device(root, vol_args->name);
2679 
2680 	if (!ret)
2681 		btrfs_info(root->fs_info, "disk added %s",vol_args->name);
2682 
2683 	kfree(vol_args);
2684 out:
2685 	mutex_unlock(&root->fs_info->volume_mutex);
2686 	atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2687 	return ret;
2688 }
2689 
2690 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
2691 {
2692 	struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2693 	struct btrfs_ioctl_vol_args_v2 *vol_args;
2694 	int ret;
2695 
2696 	if (!capable(CAP_SYS_ADMIN))
2697 		return -EPERM;
2698 
2699 	ret = mnt_want_write_file(file);
2700 	if (ret)
2701 		return ret;
2702 
2703 	vol_args = memdup_user(arg, sizeof(*vol_args));
2704 	if (IS_ERR(vol_args)) {
2705 		ret = PTR_ERR(vol_args);
2706 		goto err_drop;
2707 	}
2708 
2709 	/* Check for compatibility reject unknown flags */
2710 	if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED)
2711 		return -EOPNOTSUPP;
2712 
2713 	if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2714 			1)) {
2715 		ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2716 		goto out;
2717 	}
2718 
2719 	mutex_lock(&root->fs_info->volume_mutex);
2720 	if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
2721 		ret = btrfs_rm_device(root, NULL, vol_args->devid);
2722 	} else {
2723 		vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2724 		ret = btrfs_rm_device(root, vol_args->name, 0);
2725 	}
2726 	mutex_unlock(&root->fs_info->volume_mutex);
2727 	atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2728 
2729 	if (!ret) {
2730 		if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
2731 			btrfs_info(root->fs_info, "device deleted: id %llu",
2732 					vol_args->devid);
2733 		else
2734 			btrfs_info(root->fs_info, "device deleted: %s",
2735 					vol_args->name);
2736 	}
2737 out:
2738 	kfree(vol_args);
2739 err_drop:
2740 	mnt_drop_write_file(file);
2741 	return ret;
2742 }
2743 
2744 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2745 {
2746 	struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2747 	struct btrfs_ioctl_vol_args *vol_args;
2748 	int ret;
2749 
2750 	if (!capable(CAP_SYS_ADMIN))
2751 		return -EPERM;
2752 
2753 	ret = mnt_want_write_file(file);
2754 	if (ret)
2755 		return ret;
2756 
2757 	if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2758 			1)) {
2759 		ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2760 		goto out_drop_write;
2761 	}
2762 
2763 	vol_args = memdup_user(arg, sizeof(*vol_args));
2764 	if (IS_ERR(vol_args)) {
2765 		ret = PTR_ERR(vol_args);
2766 		goto out;
2767 	}
2768 
2769 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2770 	mutex_lock(&root->fs_info->volume_mutex);
2771 	ret = btrfs_rm_device(root, vol_args->name, 0);
2772 	mutex_unlock(&root->fs_info->volume_mutex);
2773 
2774 	if (!ret)
2775 		btrfs_info(root->fs_info, "disk deleted %s",vol_args->name);
2776 	kfree(vol_args);
2777 out:
2778 	atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2779 out_drop_write:
2780 	mnt_drop_write_file(file);
2781 
2782 	return ret;
2783 }
2784 
2785 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2786 {
2787 	struct btrfs_ioctl_fs_info_args *fi_args;
2788 	struct btrfs_device *device;
2789 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2790 	int ret = 0;
2791 
2792 	fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2793 	if (!fi_args)
2794 		return -ENOMEM;
2795 
2796 	mutex_lock(&fs_devices->device_list_mutex);
2797 	fi_args->num_devices = fs_devices->num_devices;
2798 	memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2799 
2800 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
2801 		if (device->devid > fi_args->max_id)
2802 			fi_args->max_id = device->devid;
2803 	}
2804 	mutex_unlock(&fs_devices->device_list_mutex);
2805 
2806 	fi_args->nodesize = root->fs_info->super_copy->nodesize;
2807 	fi_args->sectorsize = root->fs_info->super_copy->sectorsize;
2808 	fi_args->clone_alignment = root->fs_info->super_copy->sectorsize;
2809 
2810 	if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2811 		ret = -EFAULT;
2812 
2813 	kfree(fi_args);
2814 	return ret;
2815 }
2816 
2817 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2818 {
2819 	struct btrfs_ioctl_dev_info_args *di_args;
2820 	struct btrfs_device *dev;
2821 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2822 	int ret = 0;
2823 	char *s_uuid = NULL;
2824 
2825 	di_args = memdup_user(arg, sizeof(*di_args));
2826 	if (IS_ERR(di_args))
2827 		return PTR_ERR(di_args);
2828 
2829 	if (!btrfs_is_empty_uuid(di_args->uuid))
2830 		s_uuid = di_args->uuid;
2831 
2832 	mutex_lock(&fs_devices->device_list_mutex);
2833 	dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2834 
2835 	if (!dev) {
2836 		ret = -ENODEV;
2837 		goto out;
2838 	}
2839 
2840 	di_args->devid = dev->devid;
2841 	di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2842 	di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2843 	memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2844 	if (dev->name) {
2845 		struct rcu_string *name;
2846 
2847 		rcu_read_lock();
2848 		name = rcu_dereference(dev->name);
2849 		strncpy(di_args->path, name->str, sizeof(di_args->path));
2850 		rcu_read_unlock();
2851 		di_args->path[sizeof(di_args->path) - 1] = 0;
2852 	} else {
2853 		di_args->path[0] = '\0';
2854 	}
2855 
2856 out:
2857 	mutex_unlock(&fs_devices->device_list_mutex);
2858 	if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2859 		ret = -EFAULT;
2860 
2861 	kfree(di_args);
2862 	return ret;
2863 }
2864 
2865 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
2866 {
2867 	struct page *page;
2868 
2869 	page = grab_cache_page(inode->i_mapping, index);
2870 	if (!page)
2871 		return ERR_PTR(-ENOMEM);
2872 
2873 	if (!PageUptodate(page)) {
2874 		int ret;
2875 
2876 		ret = btrfs_readpage(NULL, page);
2877 		if (ret)
2878 			return ERR_PTR(ret);
2879 		lock_page(page);
2880 		if (!PageUptodate(page)) {
2881 			unlock_page(page);
2882 			put_page(page);
2883 			return ERR_PTR(-EIO);
2884 		}
2885 		if (page->mapping != inode->i_mapping) {
2886 			unlock_page(page);
2887 			put_page(page);
2888 			return ERR_PTR(-EAGAIN);
2889 		}
2890 	}
2891 
2892 	return page;
2893 }
2894 
2895 static int gather_extent_pages(struct inode *inode, struct page **pages,
2896 			       int num_pages, u64 off)
2897 {
2898 	int i;
2899 	pgoff_t index = off >> PAGE_SHIFT;
2900 
2901 	for (i = 0; i < num_pages; i++) {
2902 again:
2903 		pages[i] = extent_same_get_page(inode, index + i);
2904 		if (IS_ERR(pages[i])) {
2905 			int err = PTR_ERR(pages[i]);
2906 
2907 			if (err == -EAGAIN)
2908 				goto again;
2909 			pages[i] = NULL;
2910 			return err;
2911 		}
2912 	}
2913 	return 0;
2914 }
2915 
2916 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
2917 			     bool retry_range_locking)
2918 {
2919 	/*
2920 	 * Do any pending delalloc/csum calculations on inode, one way or
2921 	 * another, and lock file content.
2922 	 * The locking order is:
2923 	 *
2924 	 *   1) pages
2925 	 *   2) range in the inode's io tree
2926 	 */
2927 	while (1) {
2928 		struct btrfs_ordered_extent *ordered;
2929 		lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2930 		ordered = btrfs_lookup_first_ordered_extent(inode,
2931 							    off + len - 1);
2932 		if ((!ordered ||
2933 		     ordered->file_offset + ordered->len <= off ||
2934 		     ordered->file_offset >= off + len) &&
2935 		    !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2936 				    off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2937 			if (ordered)
2938 				btrfs_put_ordered_extent(ordered);
2939 			break;
2940 		}
2941 		unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2942 		if (ordered)
2943 			btrfs_put_ordered_extent(ordered);
2944 		if (!retry_range_locking)
2945 			return -EAGAIN;
2946 		btrfs_wait_ordered_range(inode, off, len);
2947 	}
2948 	return 0;
2949 }
2950 
2951 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
2952 {
2953 	inode_unlock(inode1);
2954 	inode_unlock(inode2);
2955 }
2956 
2957 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
2958 {
2959 	if (inode1 < inode2)
2960 		swap(inode1, inode2);
2961 
2962 	inode_lock_nested(inode1, I_MUTEX_PARENT);
2963 	inode_lock_nested(inode2, I_MUTEX_CHILD);
2964 }
2965 
2966 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
2967 				      struct inode *inode2, u64 loff2, u64 len)
2968 {
2969 	unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2970 	unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2971 }
2972 
2973 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
2974 				    struct inode *inode2, u64 loff2, u64 len,
2975 				    bool retry_range_locking)
2976 {
2977 	int ret;
2978 
2979 	if (inode1 < inode2) {
2980 		swap(inode1, inode2);
2981 		swap(loff1, loff2);
2982 	}
2983 	ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
2984 	if (ret)
2985 		return ret;
2986 	ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
2987 	if (ret)
2988 		unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
2989 			      loff1 + len - 1);
2990 	return ret;
2991 }
2992 
2993 struct cmp_pages {
2994 	int		num_pages;
2995 	struct page	**src_pages;
2996 	struct page	**dst_pages;
2997 };
2998 
2999 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
3000 {
3001 	int i;
3002 	struct page *pg;
3003 
3004 	for (i = 0; i < cmp->num_pages; i++) {
3005 		pg = cmp->src_pages[i];
3006 		if (pg) {
3007 			unlock_page(pg);
3008 			put_page(pg);
3009 		}
3010 		pg = cmp->dst_pages[i];
3011 		if (pg) {
3012 			unlock_page(pg);
3013 			put_page(pg);
3014 		}
3015 	}
3016 	kfree(cmp->src_pages);
3017 	kfree(cmp->dst_pages);
3018 }
3019 
3020 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
3021 				  struct inode *dst, u64 dst_loff,
3022 				  u64 len, struct cmp_pages *cmp)
3023 {
3024 	int ret;
3025 	int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT;
3026 	struct page **src_pgarr, **dst_pgarr;
3027 
3028 	/*
3029 	 * We must gather up all the pages before we initiate our
3030 	 * extent locking. We use an array for the page pointers. Size
3031 	 * of the array is bounded by len, which is in turn bounded by
3032 	 * BTRFS_MAX_DEDUPE_LEN.
3033 	 */
3034 	src_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
3035 	dst_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
3036 	if (!src_pgarr || !dst_pgarr) {
3037 		kfree(src_pgarr);
3038 		kfree(dst_pgarr);
3039 		return -ENOMEM;
3040 	}
3041 	cmp->num_pages = num_pages;
3042 	cmp->src_pages = src_pgarr;
3043 	cmp->dst_pages = dst_pgarr;
3044 
3045 	ret = gather_extent_pages(src, cmp->src_pages, cmp->num_pages, loff);
3046 	if (ret)
3047 		goto out;
3048 
3049 	ret = gather_extent_pages(dst, cmp->dst_pages, cmp->num_pages, dst_loff);
3050 
3051 out:
3052 	if (ret)
3053 		btrfs_cmp_data_free(cmp);
3054 	return 0;
3055 }
3056 
3057 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
3058 			  u64 dst_loff, u64 len, struct cmp_pages *cmp)
3059 {
3060 	int ret = 0;
3061 	int i;
3062 	struct page *src_page, *dst_page;
3063 	unsigned int cmp_len = PAGE_SIZE;
3064 	void *addr, *dst_addr;
3065 
3066 	i = 0;
3067 	while (len) {
3068 		if (len < PAGE_SIZE)
3069 			cmp_len = len;
3070 
3071 		BUG_ON(i >= cmp->num_pages);
3072 
3073 		src_page = cmp->src_pages[i];
3074 		dst_page = cmp->dst_pages[i];
3075 		ASSERT(PageLocked(src_page));
3076 		ASSERT(PageLocked(dst_page));
3077 
3078 		addr = kmap_atomic(src_page);
3079 		dst_addr = kmap_atomic(dst_page);
3080 
3081 		flush_dcache_page(src_page);
3082 		flush_dcache_page(dst_page);
3083 
3084 		if (memcmp(addr, dst_addr, cmp_len))
3085 			ret = -EBADE;
3086 
3087 		kunmap_atomic(addr);
3088 		kunmap_atomic(dst_addr);
3089 
3090 		if (ret)
3091 			break;
3092 
3093 		len -= cmp_len;
3094 		i++;
3095 	}
3096 
3097 	return ret;
3098 }
3099 
3100 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3101 				     u64 olen)
3102 {
3103 	u64 len = *plen;
3104 	u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3105 
3106 	if (off + olen > inode->i_size || off + olen < off)
3107 		return -EINVAL;
3108 
3109 	/* if we extend to eof, continue to block boundary */
3110 	if (off + len == inode->i_size)
3111 		*plen = len = ALIGN(inode->i_size, bs) - off;
3112 
3113 	/* Check that we are block aligned - btrfs_clone() requires this */
3114 	if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3115 		return -EINVAL;
3116 
3117 	return 0;
3118 }
3119 
3120 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3121 			     struct inode *dst, u64 dst_loff)
3122 {
3123 	int ret;
3124 	u64 len = olen;
3125 	struct cmp_pages cmp;
3126 	int same_inode = 0;
3127 	u64 same_lock_start = 0;
3128 	u64 same_lock_len = 0;
3129 
3130 	if (src == dst)
3131 		same_inode = 1;
3132 
3133 	if (len == 0)
3134 		return 0;
3135 
3136 	if (same_inode) {
3137 		inode_lock(src);
3138 
3139 		ret = extent_same_check_offsets(src, loff, &len, olen);
3140 		if (ret)
3141 			goto out_unlock;
3142 		ret = extent_same_check_offsets(src, dst_loff, &len, olen);
3143 		if (ret)
3144 			goto out_unlock;
3145 
3146 		/*
3147 		 * Single inode case wants the same checks, except we
3148 		 * don't want our length pushed out past i_size as
3149 		 * comparing that data range makes no sense.
3150 		 *
3151 		 * extent_same_check_offsets() will do this for an
3152 		 * unaligned length at i_size, so catch it here and
3153 		 * reject the request.
3154 		 *
3155 		 * This effectively means we require aligned extents
3156 		 * for the single-inode case, whereas the other cases
3157 		 * allow an unaligned length so long as it ends at
3158 		 * i_size.
3159 		 */
3160 		if (len != olen) {
3161 			ret = -EINVAL;
3162 			goto out_unlock;
3163 		}
3164 
3165 		/* Check for overlapping ranges */
3166 		if (dst_loff + len > loff && dst_loff < loff + len) {
3167 			ret = -EINVAL;
3168 			goto out_unlock;
3169 		}
3170 
3171 		same_lock_start = min_t(u64, loff, dst_loff);
3172 		same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3173 	} else {
3174 		btrfs_double_inode_lock(src, dst);
3175 
3176 		ret = extent_same_check_offsets(src, loff, &len, olen);
3177 		if (ret)
3178 			goto out_unlock;
3179 
3180 		ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3181 		if (ret)
3182 			goto out_unlock;
3183 	}
3184 
3185 	/* don't make the dst file partly checksummed */
3186 	if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3187 	    (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3188 		ret = -EINVAL;
3189 		goto out_unlock;
3190 	}
3191 
3192 again:
3193 	ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp);
3194 	if (ret)
3195 		goto out_unlock;
3196 
3197 	if (same_inode)
3198 		ret = lock_extent_range(src, same_lock_start, same_lock_len,
3199 					false);
3200 	else
3201 		ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3202 					       false);
3203 	/*
3204 	 * If one of the inodes has dirty pages in the respective range or
3205 	 * ordered extents, we need to flush dellaloc and wait for all ordered
3206 	 * extents in the range. We must unlock the pages and the ranges in the
3207 	 * io trees to avoid deadlocks when flushing delalloc (requires locking
3208 	 * pages) and when waiting for ordered extents to complete (they require
3209 	 * range locking).
3210 	 */
3211 	if (ret == -EAGAIN) {
3212 		/*
3213 		 * Ranges in the io trees already unlocked. Now unlock all
3214 		 * pages before waiting for all IO to complete.
3215 		 */
3216 		btrfs_cmp_data_free(&cmp);
3217 		if (same_inode) {
3218 			btrfs_wait_ordered_range(src, same_lock_start,
3219 						 same_lock_len);
3220 		} else {
3221 			btrfs_wait_ordered_range(src, loff, len);
3222 			btrfs_wait_ordered_range(dst, dst_loff, len);
3223 		}
3224 		goto again;
3225 	}
3226 	ASSERT(ret == 0);
3227 	if (WARN_ON(ret)) {
3228 		/* ranges in the io trees already unlocked */
3229 		btrfs_cmp_data_free(&cmp);
3230 		return ret;
3231 	}
3232 
3233 	/* pass original length for comparison so we stay within i_size */
3234 	ret = btrfs_cmp_data(src, loff, dst, dst_loff, olen, &cmp);
3235 	if (ret == 0)
3236 		ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3237 
3238 	if (same_inode)
3239 		unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3240 			      same_lock_start + same_lock_len - 1);
3241 	else
3242 		btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3243 
3244 	btrfs_cmp_data_free(&cmp);
3245 out_unlock:
3246 	if (same_inode)
3247 		inode_unlock(src);
3248 	else
3249 		btrfs_double_inode_unlock(src, dst);
3250 
3251 	return ret;
3252 }
3253 
3254 #define BTRFS_MAX_DEDUPE_LEN	SZ_16M
3255 
3256 ssize_t btrfs_dedupe_file_range(struct file *src_file, u64 loff, u64 olen,
3257 				struct file *dst_file, u64 dst_loff)
3258 {
3259 	struct inode *src = file_inode(src_file);
3260 	struct inode *dst = file_inode(dst_file);
3261 	u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3262 	ssize_t res;
3263 
3264 	if (olen > BTRFS_MAX_DEDUPE_LEN)
3265 		olen = BTRFS_MAX_DEDUPE_LEN;
3266 
3267 	if (WARN_ON_ONCE(bs < PAGE_SIZE)) {
3268 		/*
3269 		 * Btrfs does not support blocksize < page_size. As a
3270 		 * result, btrfs_cmp_data() won't correctly handle
3271 		 * this situation without an update.
3272 		 */
3273 		return -EINVAL;
3274 	}
3275 
3276 	res = btrfs_extent_same(src, loff, olen, dst, dst_loff);
3277 	if (res)
3278 		return res;
3279 	return olen;
3280 }
3281 
3282 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3283 				     struct inode *inode,
3284 				     u64 endoff,
3285 				     const u64 destoff,
3286 				     const u64 olen,
3287 				     int no_time_update)
3288 {
3289 	struct btrfs_root *root = BTRFS_I(inode)->root;
3290 	int ret;
3291 
3292 	inode_inc_iversion(inode);
3293 	if (!no_time_update)
3294 		inode->i_mtime = inode->i_ctime = current_fs_time(inode->i_sb);
3295 	/*
3296 	 * We round up to the block size at eof when determining which
3297 	 * extents to clone above, but shouldn't round up the file size.
3298 	 */
3299 	if (endoff > destoff + olen)
3300 		endoff = destoff + olen;
3301 	if (endoff > inode->i_size)
3302 		btrfs_i_size_write(inode, endoff);
3303 
3304 	ret = btrfs_update_inode(trans, root, inode);
3305 	if (ret) {
3306 		btrfs_abort_transaction(trans, ret);
3307 		btrfs_end_transaction(trans, root);
3308 		goto out;
3309 	}
3310 	ret = btrfs_end_transaction(trans, root);
3311 out:
3312 	return ret;
3313 }
3314 
3315 static void clone_update_extent_map(struct inode *inode,
3316 				    const struct btrfs_trans_handle *trans,
3317 				    const struct btrfs_path *path,
3318 				    const u64 hole_offset,
3319 				    const u64 hole_len)
3320 {
3321 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3322 	struct extent_map *em;
3323 	int ret;
3324 
3325 	em = alloc_extent_map();
3326 	if (!em) {
3327 		set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3328 			&BTRFS_I(inode)->runtime_flags);
3329 		return;
3330 	}
3331 
3332 	if (path) {
3333 		struct btrfs_file_extent_item *fi;
3334 
3335 		fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3336 				    struct btrfs_file_extent_item);
3337 		btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3338 		em->generation = -1;
3339 		if (btrfs_file_extent_type(path->nodes[0], fi) ==
3340 		    BTRFS_FILE_EXTENT_INLINE)
3341 			set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3342 				&BTRFS_I(inode)->runtime_flags);
3343 	} else {
3344 		em->start = hole_offset;
3345 		em->len = hole_len;
3346 		em->ram_bytes = em->len;
3347 		em->orig_start = hole_offset;
3348 		em->block_start = EXTENT_MAP_HOLE;
3349 		em->block_len = 0;
3350 		em->orig_block_len = 0;
3351 		em->compress_type = BTRFS_COMPRESS_NONE;
3352 		em->generation = trans->transid;
3353 	}
3354 
3355 	while (1) {
3356 		write_lock(&em_tree->lock);
3357 		ret = add_extent_mapping(em_tree, em, 1);
3358 		write_unlock(&em_tree->lock);
3359 		if (ret != -EEXIST) {
3360 			free_extent_map(em);
3361 			break;
3362 		}
3363 		btrfs_drop_extent_cache(inode, em->start,
3364 					em->start + em->len - 1, 0);
3365 	}
3366 
3367 	if (ret)
3368 		set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3369 			&BTRFS_I(inode)->runtime_flags);
3370 }
3371 
3372 /*
3373  * Make sure we do not end up inserting an inline extent into a file that has
3374  * already other (non-inline) extents. If a file has an inline extent it can
3375  * not have any other extents and the (single) inline extent must start at the
3376  * file offset 0. Failing to respect these rules will lead to file corruption,
3377  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3378  *
3379  * We can have extents that have been already written to disk or we can have
3380  * dirty ranges still in delalloc, in which case the extent maps and items are
3381  * created only when we run delalloc, and the delalloc ranges might fall outside
3382  * the range we are currently locking in the inode's io tree. So we check the
3383  * inode's i_size because of that (i_size updates are done while holding the
3384  * i_mutex, which we are holding here).
3385  * We also check to see if the inode has a size not greater than "datal" but has
3386  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3387  * protected against such concurrent fallocate calls by the i_mutex).
3388  *
3389  * If the file has no extents but a size greater than datal, do not allow the
3390  * copy because we would need turn the inline extent into a non-inline one (even
3391  * with NO_HOLES enabled). If we find our destination inode only has one inline
3392  * extent, just overwrite it with the source inline extent if its size is less
3393  * than the source extent's size, or we could copy the source inline extent's
3394  * data into the destination inode's inline extent if the later is greater then
3395  * the former.
3396  */
3397 static int clone_copy_inline_extent(struct inode *src,
3398 				    struct inode *dst,
3399 				    struct btrfs_trans_handle *trans,
3400 				    struct btrfs_path *path,
3401 				    struct btrfs_key *new_key,
3402 				    const u64 drop_start,
3403 				    const u64 datal,
3404 				    const u64 skip,
3405 				    const u64 size,
3406 				    char *inline_data)
3407 {
3408 	struct btrfs_root *root = BTRFS_I(dst)->root;
3409 	const u64 aligned_end = ALIGN(new_key->offset + datal,
3410 				      root->sectorsize);
3411 	int ret;
3412 	struct btrfs_key key;
3413 
3414 	if (new_key->offset > 0)
3415 		return -EOPNOTSUPP;
3416 
3417 	key.objectid = btrfs_ino(dst);
3418 	key.type = BTRFS_EXTENT_DATA_KEY;
3419 	key.offset = 0;
3420 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3421 	if (ret < 0) {
3422 		return ret;
3423 	} else if (ret > 0) {
3424 		if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3425 			ret = btrfs_next_leaf(root, path);
3426 			if (ret < 0)
3427 				return ret;
3428 			else if (ret > 0)
3429 				goto copy_inline_extent;
3430 		}
3431 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3432 		if (key.objectid == btrfs_ino(dst) &&
3433 		    key.type == BTRFS_EXTENT_DATA_KEY) {
3434 			ASSERT(key.offset > 0);
3435 			return -EOPNOTSUPP;
3436 		}
3437 	} else if (i_size_read(dst) <= datal) {
3438 		struct btrfs_file_extent_item *ei;
3439 		u64 ext_len;
3440 
3441 		/*
3442 		 * If the file size is <= datal, make sure there are no other
3443 		 * extents following (can happen do to an fallocate call with
3444 		 * the flag FALLOC_FL_KEEP_SIZE).
3445 		 */
3446 		ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3447 				    struct btrfs_file_extent_item);
3448 		/*
3449 		 * If it's an inline extent, it can not have other extents
3450 		 * following it.
3451 		 */
3452 		if (btrfs_file_extent_type(path->nodes[0], ei) ==
3453 		    BTRFS_FILE_EXTENT_INLINE)
3454 			goto copy_inline_extent;
3455 
3456 		ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3457 		if (ext_len > aligned_end)
3458 			return -EOPNOTSUPP;
3459 
3460 		ret = btrfs_next_item(root, path);
3461 		if (ret < 0) {
3462 			return ret;
3463 		} else if (ret == 0) {
3464 			btrfs_item_key_to_cpu(path->nodes[0], &key,
3465 					      path->slots[0]);
3466 			if (key.objectid == btrfs_ino(dst) &&
3467 			    key.type == BTRFS_EXTENT_DATA_KEY)
3468 				return -EOPNOTSUPP;
3469 		}
3470 	}
3471 
3472 copy_inline_extent:
3473 	/*
3474 	 * We have no extent items, or we have an extent at offset 0 which may
3475 	 * or may not be inlined. All these cases are dealt the same way.
3476 	 */
3477 	if (i_size_read(dst) > datal) {
3478 		/*
3479 		 * If the destination inode has an inline extent...
3480 		 * This would require copying the data from the source inline
3481 		 * extent into the beginning of the destination's inline extent.
3482 		 * But this is really complex, both extents can be compressed
3483 		 * or just one of them, which would require decompressing and
3484 		 * re-compressing data (which could increase the new compressed
3485 		 * size, not allowing the compressed data to fit anymore in an
3486 		 * inline extent).
3487 		 * So just don't support this case for now (it should be rare,
3488 		 * we are not really saving space when cloning inline extents).
3489 		 */
3490 		return -EOPNOTSUPP;
3491 	}
3492 
3493 	btrfs_release_path(path);
3494 	ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3495 	if (ret)
3496 		return ret;
3497 	ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3498 	if (ret)
3499 		return ret;
3500 
3501 	if (skip) {
3502 		const u32 start = btrfs_file_extent_calc_inline_size(0);
3503 
3504 		memmove(inline_data + start, inline_data + start + skip, datal);
3505 	}
3506 
3507 	write_extent_buffer(path->nodes[0], inline_data,
3508 			    btrfs_item_ptr_offset(path->nodes[0],
3509 						  path->slots[0]),
3510 			    size);
3511 	inode_add_bytes(dst, datal);
3512 
3513 	return 0;
3514 }
3515 
3516 /**
3517  * btrfs_clone() - clone a range from inode file to another
3518  *
3519  * @src: Inode to clone from
3520  * @inode: Inode to clone to
3521  * @off: Offset within source to start clone from
3522  * @olen: Original length, passed by user, of range to clone
3523  * @olen_aligned: Block-aligned value of olen
3524  * @destoff: Offset within @inode to start clone
3525  * @no_time_update: Whether to update mtime/ctime on the target inode
3526  */
3527 static int btrfs_clone(struct inode *src, struct inode *inode,
3528 		       const u64 off, const u64 olen, const u64 olen_aligned,
3529 		       const u64 destoff, int no_time_update)
3530 {
3531 	struct btrfs_root *root = BTRFS_I(inode)->root;
3532 	struct btrfs_path *path = NULL;
3533 	struct extent_buffer *leaf;
3534 	struct btrfs_trans_handle *trans;
3535 	char *buf = NULL;
3536 	struct btrfs_key key;
3537 	u32 nritems;
3538 	int slot;
3539 	int ret;
3540 	const u64 len = olen_aligned;
3541 	u64 last_dest_end = destoff;
3542 
3543 	ret = -ENOMEM;
3544 	buf = kmalloc(root->nodesize, GFP_KERNEL | __GFP_NOWARN);
3545 	if (!buf) {
3546 		buf = vmalloc(root->nodesize);
3547 		if (!buf)
3548 			return ret;
3549 	}
3550 
3551 	path = btrfs_alloc_path();
3552 	if (!path) {
3553 		kvfree(buf);
3554 		return ret;
3555 	}
3556 
3557 	path->reada = READA_FORWARD;
3558 	/* clone data */
3559 	key.objectid = btrfs_ino(src);
3560 	key.type = BTRFS_EXTENT_DATA_KEY;
3561 	key.offset = off;
3562 
3563 	while (1) {
3564 		u64 next_key_min_offset = key.offset + 1;
3565 
3566 		/*
3567 		 * note the key will change type as we walk through the
3568 		 * tree.
3569 		 */
3570 		path->leave_spinning = 1;
3571 		ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3572 				0, 0);
3573 		if (ret < 0)
3574 			goto out;
3575 		/*
3576 		 * First search, if no extent item that starts at offset off was
3577 		 * found but the previous item is an extent item, it's possible
3578 		 * it might overlap our target range, therefore process it.
3579 		 */
3580 		if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3581 			btrfs_item_key_to_cpu(path->nodes[0], &key,
3582 					      path->slots[0] - 1);
3583 			if (key.type == BTRFS_EXTENT_DATA_KEY)
3584 				path->slots[0]--;
3585 		}
3586 
3587 		nritems = btrfs_header_nritems(path->nodes[0]);
3588 process_slot:
3589 		if (path->slots[0] >= nritems) {
3590 			ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3591 			if (ret < 0)
3592 				goto out;
3593 			if (ret > 0)
3594 				break;
3595 			nritems = btrfs_header_nritems(path->nodes[0]);
3596 		}
3597 		leaf = path->nodes[0];
3598 		slot = path->slots[0];
3599 
3600 		btrfs_item_key_to_cpu(leaf, &key, slot);
3601 		if (key.type > BTRFS_EXTENT_DATA_KEY ||
3602 		    key.objectid != btrfs_ino(src))
3603 			break;
3604 
3605 		if (key.type == BTRFS_EXTENT_DATA_KEY) {
3606 			struct btrfs_file_extent_item *extent;
3607 			int type;
3608 			u32 size;
3609 			struct btrfs_key new_key;
3610 			u64 disko = 0, diskl = 0;
3611 			u64 datao = 0, datal = 0;
3612 			u8 comp;
3613 			u64 drop_start;
3614 
3615 			extent = btrfs_item_ptr(leaf, slot,
3616 						struct btrfs_file_extent_item);
3617 			comp = btrfs_file_extent_compression(leaf, extent);
3618 			type = btrfs_file_extent_type(leaf, extent);
3619 			if (type == BTRFS_FILE_EXTENT_REG ||
3620 			    type == BTRFS_FILE_EXTENT_PREALLOC) {
3621 				disko = btrfs_file_extent_disk_bytenr(leaf,
3622 								      extent);
3623 				diskl = btrfs_file_extent_disk_num_bytes(leaf,
3624 								 extent);
3625 				datao = btrfs_file_extent_offset(leaf, extent);
3626 				datal = btrfs_file_extent_num_bytes(leaf,
3627 								    extent);
3628 			} else if (type == BTRFS_FILE_EXTENT_INLINE) {
3629 				/* take upper bound, may be compressed */
3630 				datal = btrfs_file_extent_ram_bytes(leaf,
3631 								    extent);
3632 			}
3633 
3634 			/*
3635 			 * The first search might have left us at an extent
3636 			 * item that ends before our target range's start, can
3637 			 * happen if we have holes and NO_HOLES feature enabled.
3638 			 */
3639 			if (key.offset + datal <= off) {
3640 				path->slots[0]++;
3641 				goto process_slot;
3642 			} else if (key.offset >= off + len) {
3643 				break;
3644 			}
3645 			next_key_min_offset = key.offset + datal;
3646 			size = btrfs_item_size_nr(leaf, slot);
3647 			read_extent_buffer(leaf, buf,
3648 					   btrfs_item_ptr_offset(leaf, slot),
3649 					   size);
3650 
3651 			btrfs_release_path(path);
3652 			path->leave_spinning = 0;
3653 
3654 			memcpy(&new_key, &key, sizeof(new_key));
3655 			new_key.objectid = btrfs_ino(inode);
3656 			if (off <= key.offset)
3657 				new_key.offset = key.offset + destoff - off;
3658 			else
3659 				new_key.offset = destoff;
3660 
3661 			/*
3662 			 * Deal with a hole that doesn't have an extent item
3663 			 * that represents it (NO_HOLES feature enabled).
3664 			 * This hole is either in the middle of the cloning
3665 			 * range or at the beginning (fully overlaps it or
3666 			 * partially overlaps it).
3667 			 */
3668 			if (new_key.offset != last_dest_end)
3669 				drop_start = last_dest_end;
3670 			else
3671 				drop_start = new_key.offset;
3672 
3673 			/*
3674 			 * 1 - adjusting old extent (we may have to split it)
3675 			 * 1 - add new extent
3676 			 * 1 - inode update
3677 			 */
3678 			trans = btrfs_start_transaction(root, 3);
3679 			if (IS_ERR(trans)) {
3680 				ret = PTR_ERR(trans);
3681 				goto out;
3682 			}
3683 
3684 			if (type == BTRFS_FILE_EXTENT_REG ||
3685 			    type == BTRFS_FILE_EXTENT_PREALLOC) {
3686 				/*
3687 				 *    a  | --- range to clone ---|  b
3688 				 * | ------------- extent ------------- |
3689 				 */
3690 
3691 				/* subtract range b */
3692 				if (key.offset + datal > off + len)
3693 					datal = off + len - key.offset;
3694 
3695 				/* subtract range a */
3696 				if (off > key.offset) {
3697 					datao += off - key.offset;
3698 					datal -= off - key.offset;
3699 				}
3700 
3701 				ret = btrfs_drop_extents(trans, root, inode,
3702 							 drop_start,
3703 							 new_key.offset + datal,
3704 							 1);
3705 				if (ret) {
3706 					if (ret != -EOPNOTSUPP)
3707 						btrfs_abort_transaction(trans,
3708 									ret);
3709 					btrfs_end_transaction(trans, root);
3710 					goto out;
3711 				}
3712 
3713 				ret = btrfs_insert_empty_item(trans, root, path,
3714 							      &new_key, size);
3715 				if (ret) {
3716 					btrfs_abort_transaction(trans, ret);
3717 					btrfs_end_transaction(trans, root);
3718 					goto out;
3719 				}
3720 
3721 				leaf = path->nodes[0];
3722 				slot = path->slots[0];
3723 				write_extent_buffer(leaf, buf,
3724 					    btrfs_item_ptr_offset(leaf, slot),
3725 					    size);
3726 
3727 				extent = btrfs_item_ptr(leaf, slot,
3728 						struct btrfs_file_extent_item);
3729 
3730 				/* disko == 0 means it's a hole */
3731 				if (!disko)
3732 					datao = 0;
3733 
3734 				btrfs_set_file_extent_offset(leaf, extent,
3735 							     datao);
3736 				btrfs_set_file_extent_num_bytes(leaf, extent,
3737 								datal);
3738 
3739 				if (disko) {
3740 					inode_add_bytes(inode, datal);
3741 					ret = btrfs_inc_extent_ref(trans, root,
3742 							disko, diskl, 0,
3743 							root->root_key.objectid,
3744 							btrfs_ino(inode),
3745 							new_key.offset - datao);
3746 					if (ret) {
3747 						btrfs_abort_transaction(trans,
3748 									ret);
3749 						btrfs_end_transaction(trans,
3750 								      root);
3751 						goto out;
3752 
3753 					}
3754 				}
3755 			} else if (type == BTRFS_FILE_EXTENT_INLINE) {
3756 				u64 skip = 0;
3757 				u64 trim = 0;
3758 
3759 				if (off > key.offset) {
3760 					skip = off - key.offset;
3761 					new_key.offset += skip;
3762 				}
3763 
3764 				if (key.offset + datal > off + len)
3765 					trim = key.offset + datal - (off + len);
3766 
3767 				if (comp && (skip || trim)) {
3768 					ret = -EINVAL;
3769 					btrfs_end_transaction(trans, root);
3770 					goto out;
3771 				}
3772 				size -= skip + trim;
3773 				datal -= skip + trim;
3774 
3775 				ret = clone_copy_inline_extent(src, inode,
3776 							       trans, path,
3777 							       &new_key,
3778 							       drop_start,
3779 							       datal,
3780 							       skip, size, buf);
3781 				if (ret) {
3782 					if (ret != -EOPNOTSUPP)
3783 						btrfs_abort_transaction(trans,
3784 									ret);
3785 					btrfs_end_transaction(trans, root);
3786 					goto out;
3787 				}
3788 				leaf = path->nodes[0];
3789 				slot = path->slots[0];
3790 			}
3791 
3792 			/* If we have an implicit hole (NO_HOLES feature). */
3793 			if (drop_start < new_key.offset)
3794 				clone_update_extent_map(inode, trans,
3795 						NULL, drop_start,
3796 						new_key.offset - drop_start);
3797 
3798 			clone_update_extent_map(inode, trans, path, 0, 0);
3799 
3800 			btrfs_mark_buffer_dirty(leaf);
3801 			btrfs_release_path(path);
3802 
3803 			last_dest_end = ALIGN(new_key.offset + datal,
3804 					      root->sectorsize);
3805 			ret = clone_finish_inode_update(trans, inode,
3806 							last_dest_end,
3807 							destoff, olen,
3808 							no_time_update);
3809 			if (ret)
3810 				goto out;
3811 			if (new_key.offset + datal >= destoff + len)
3812 				break;
3813 		}
3814 		btrfs_release_path(path);
3815 		key.offset = next_key_min_offset;
3816 	}
3817 	ret = 0;
3818 
3819 	if (last_dest_end < destoff + len) {
3820 		/*
3821 		 * We have an implicit hole (NO_HOLES feature is enabled) that
3822 		 * fully or partially overlaps our cloning range at its end.
3823 		 */
3824 		btrfs_release_path(path);
3825 
3826 		/*
3827 		 * 1 - remove extent(s)
3828 		 * 1 - inode update
3829 		 */
3830 		trans = btrfs_start_transaction(root, 2);
3831 		if (IS_ERR(trans)) {
3832 			ret = PTR_ERR(trans);
3833 			goto out;
3834 		}
3835 		ret = btrfs_drop_extents(trans, root, inode,
3836 					 last_dest_end, destoff + len, 1);
3837 		if (ret) {
3838 			if (ret != -EOPNOTSUPP)
3839 				btrfs_abort_transaction(trans, ret);
3840 			btrfs_end_transaction(trans, root);
3841 			goto out;
3842 		}
3843 		clone_update_extent_map(inode, trans, NULL, last_dest_end,
3844 					destoff + len - last_dest_end);
3845 		ret = clone_finish_inode_update(trans, inode, destoff + len,
3846 						destoff, olen, no_time_update);
3847 	}
3848 
3849 out:
3850 	btrfs_free_path(path);
3851 	kvfree(buf);
3852 	return ret;
3853 }
3854 
3855 static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
3856 					u64 off, u64 olen, u64 destoff)
3857 {
3858 	struct inode *inode = file_inode(file);
3859 	struct inode *src = file_inode(file_src);
3860 	struct btrfs_root *root = BTRFS_I(inode)->root;
3861 	int ret;
3862 	u64 len = olen;
3863 	u64 bs = root->fs_info->sb->s_blocksize;
3864 	int same_inode = src == inode;
3865 
3866 	/*
3867 	 * TODO:
3868 	 * - split compressed inline extents.  annoying: we need to
3869 	 *   decompress into destination's address_space (the file offset
3870 	 *   may change, so source mapping won't do), then recompress (or
3871 	 *   otherwise reinsert) a subrange.
3872 	 *
3873 	 * - split destination inode's inline extents.  The inline extents can
3874 	 *   be either compressed or non-compressed.
3875 	 */
3876 
3877 	if (btrfs_root_readonly(root))
3878 		return -EROFS;
3879 
3880 	if (file_src->f_path.mnt != file->f_path.mnt ||
3881 	    src->i_sb != inode->i_sb)
3882 		return -EXDEV;
3883 
3884 	/* don't make the dst file partly checksummed */
3885 	if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3886 	    (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3887 		return -EINVAL;
3888 
3889 	if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3890 		return -EISDIR;
3891 
3892 	if (!same_inode) {
3893 		btrfs_double_inode_lock(src, inode);
3894 	} else {
3895 		inode_lock(src);
3896 	}
3897 
3898 	/* determine range to clone */
3899 	ret = -EINVAL;
3900 	if (off + len > src->i_size || off + len < off)
3901 		goto out_unlock;
3902 	if (len == 0)
3903 		olen = len = src->i_size - off;
3904 	/* if we extend to eof, continue to block boundary */
3905 	if (off + len == src->i_size)
3906 		len = ALIGN(src->i_size, bs) - off;
3907 
3908 	if (len == 0) {
3909 		ret = 0;
3910 		goto out_unlock;
3911 	}
3912 
3913 	/* verify the end result is block aligned */
3914 	if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3915 	    !IS_ALIGNED(destoff, bs))
3916 		goto out_unlock;
3917 
3918 	/* verify if ranges are overlapped within the same file */
3919 	if (same_inode) {
3920 		if (destoff + len > off && destoff < off + len)
3921 			goto out_unlock;
3922 	}
3923 
3924 	if (destoff > inode->i_size) {
3925 		ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3926 		if (ret)
3927 			goto out_unlock;
3928 	}
3929 
3930 	/*
3931 	 * Lock the target range too. Right after we replace the file extent
3932 	 * items in the fs tree (which now point to the cloned data), we might
3933 	 * have a worker replace them with extent items relative to a write
3934 	 * operation that was issued before this clone operation (i.e. confront
3935 	 * with inode.c:btrfs_finish_ordered_io).
3936 	 */
3937 	if (same_inode) {
3938 		u64 lock_start = min_t(u64, off, destoff);
3939 		u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
3940 
3941 		ret = lock_extent_range(src, lock_start, lock_len, true);
3942 	} else {
3943 		ret = btrfs_double_extent_lock(src, off, inode, destoff, len,
3944 					       true);
3945 	}
3946 	ASSERT(ret == 0);
3947 	if (WARN_ON(ret)) {
3948 		/* ranges in the io trees already unlocked */
3949 		goto out_unlock;
3950 	}
3951 
3952 	ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
3953 
3954 	if (same_inode) {
3955 		u64 lock_start = min_t(u64, off, destoff);
3956 		u64 lock_end = max_t(u64, off, destoff) + len - 1;
3957 
3958 		unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
3959 	} else {
3960 		btrfs_double_extent_unlock(src, off, inode, destoff, len);
3961 	}
3962 	/*
3963 	 * Truncate page cache pages so that future reads will see the cloned
3964 	 * data immediately and not the previous data.
3965 	 */
3966 	truncate_inode_pages_range(&inode->i_data,
3967 				round_down(destoff, PAGE_SIZE),
3968 				round_up(destoff + len, PAGE_SIZE) - 1);
3969 out_unlock:
3970 	if (!same_inode)
3971 		btrfs_double_inode_unlock(src, inode);
3972 	else
3973 		inode_unlock(src);
3974 	return ret;
3975 }
3976 
3977 ssize_t btrfs_copy_file_range(struct file *file_in, loff_t pos_in,
3978 			      struct file *file_out, loff_t pos_out,
3979 			      size_t len, unsigned int flags)
3980 {
3981 	ssize_t ret;
3982 
3983 	ret = btrfs_clone_files(file_out, file_in, pos_in, len, pos_out);
3984 	if (ret == 0)
3985 		ret = len;
3986 	return ret;
3987 }
3988 
3989 int btrfs_clone_file_range(struct file *src_file, loff_t off,
3990 		struct file *dst_file, loff_t destoff, u64 len)
3991 {
3992 	return btrfs_clone_files(dst_file, src_file, off, len, destoff);
3993 }
3994 
3995 /*
3996  * there are many ways the trans_start and trans_end ioctls can lead
3997  * to deadlocks.  They should only be used by applications that
3998  * basically own the machine, and have a very in depth understanding
3999  * of all the possible deadlocks and enospc problems.
4000  */
4001 static long btrfs_ioctl_trans_start(struct file *file)
4002 {
4003 	struct inode *inode = file_inode(file);
4004 	struct btrfs_root *root = BTRFS_I(inode)->root;
4005 	struct btrfs_trans_handle *trans;
4006 	int ret;
4007 
4008 	ret = -EPERM;
4009 	if (!capable(CAP_SYS_ADMIN))
4010 		goto out;
4011 
4012 	ret = -EINPROGRESS;
4013 	if (file->private_data)
4014 		goto out;
4015 
4016 	ret = -EROFS;
4017 	if (btrfs_root_readonly(root))
4018 		goto out;
4019 
4020 	ret = mnt_want_write_file(file);
4021 	if (ret)
4022 		goto out;
4023 
4024 	atomic_inc(&root->fs_info->open_ioctl_trans);
4025 
4026 	ret = -ENOMEM;
4027 	trans = btrfs_start_ioctl_transaction(root);
4028 	if (IS_ERR(trans))
4029 		goto out_drop;
4030 
4031 	file->private_data = trans;
4032 	return 0;
4033 
4034 out_drop:
4035 	atomic_dec(&root->fs_info->open_ioctl_trans);
4036 	mnt_drop_write_file(file);
4037 out:
4038 	return ret;
4039 }
4040 
4041 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
4042 {
4043 	struct inode *inode = file_inode(file);
4044 	struct btrfs_root *root = BTRFS_I(inode)->root;
4045 	struct btrfs_root *new_root;
4046 	struct btrfs_dir_item *di;
4047 	struct btrfs_trans_handle *trans;
4048 	struct btrfs_path *path;
4049 	struct btrfs_key location;
4050 	struct btrfs_disk_key disk_key;
4051 	u64 objectid = 0;
4052 	u64 dir_id;
4053 	int ret;
4054 
4055 	if (!capable(CAP_SYS_ADMIN))
4056 		return -EPERM;
4057 
4058 	ret = mnt_want_write_file(file);
4059 	if (ret)
4060 		return ret;
4061 
4062 	if (copy_from_user(&objectid, argp, sizeof(objectid))) {
4063 		ret = -EFAULT;
4064 		goto out;
4065 	}
4066 
4067 	if (!objectid)
4068 		objectid = BTRFS_FS_TREE_OBJECTID;
4069 
4070 	location.objectid = objectid;
4071 	location.type = BTRFS_ROOT_ITEM_KEY;
4072 	location.offset = (u64)-1;
4073 
4074 	new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
4075 	if (IS_ERR(new_root)) {
4076 		ret = PTR_ERR(new_root);
4077 		goto out;
4078 	}
4079 
4080 	path = btrfs_alloc_path();
4081 	if (!path) {
4082 		ret = -ENOMEM;
4083 		goto out;
4084 	}
4085 	path->leave_spinning = 1;
4086 
4087 	trans = btrfs_start_transaction(root, 1);
4088 	if (IS_ERR(trans)) {
4089 		btrfs_free_path(path);
4090 		ret = PTR_ERR(trans);
4091 		goto out;
4092 	}
4093 
4094 	dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
4095 	di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
4096 				   dir_id, "default", 7, 1);
4097 	if (IS_ERR_OR_NULL(di)) {
4098 		btrfs_free_path(path);
4099 		btrfs_end_transaction(trans, root);
4100 		btrfs_err(new_root->fs_info, "Umm, you don't have the default dir"
4101 			   "item, this isn't going to work");
4102 		ret = -ENOENT;
4103 		goto out;
4104 	}
4105 
4106 	btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
4107 	btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
4108 	btrfs_mark_buffer_dirty(path->nodes[0]);
4109 	btrfs_free_path(path);
4110 
4111 	btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
4112 	btrfs_end_transaction(trans, root);
4113 out:
4114 	mnt_drop_write_file(file);
4115 	return ret;
4116 }
4117 
4118 void btrfs_get_block_group_info(struct list_head *groups_list,
4119 				struct btrfs_ioctl_space_info *space)
4120 {
4121 	struct btrfs_block_group_cache *block_group;
4122 
4123 	space->total_bytes = 0;
4124 	space->used_bytes = 0;
4125 	space->flags = 0;
4126 	list_for_each_entry(block_group, groups_list, list) {
4127 		space->flags = block_group->flags;
4128 		space->total_bytes += block_group->key.offset;
4129 		space->used_bytes +=
4130 			btrfs_block_group_used(&block_group->item);
4131 	}
4132 }
4133 
4134 static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
4135 {
4136 	struct btrfs_ioctl_space_args space_args;
4137 	struct btrfs_ioctl_space_info space;
4138 	struct btrfs_ioctl_space_info *dest;
4139 	struct btrfs_ioctl_space_info *dest_orig;
4140 	struct btrfs_ioctl_space_info __user *user_dest;
4141 	struct btrfs_space_info *info;
4142 	u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
4143 		       BTRFS_BLOCK_GROUP_SYSTEM,
4144 		       BTRFS_BLOCK_GROUP_METADATA,
4145 		       BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
4146 	int num_types = 4;
4147 	int alloc_size;
4148 	int ret = 0;
4149 	u64 slot_count = 0;
4150 	int i, c;
4151 
4152 	if (copy_from_user(&space_args,
4153 			   (struct btrfs_ioctl_space_args __user *)arg,
4154 			   sizeof(space_args)))
4155 		return -EFAULT;
4156 
4157 	for (i = 0; i < num_types; i++) {
4158 		struct btrfs_space_info *tmp;
4159 
4160 		info = NULL;
4161 		rcu_read_lock();
4162 		list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
4163 					list) {
4164 			if (tmp->flags == types[i]) {
4165 				info = tmp;
4166 				break;
4167 			}
4168 		}
4169 		rcu_read_unlock();
4170 
4171 		if (!info)
4172 			continue;
4173 
4174 		down_read(&info->groups_sem);
4175 		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4176 			if (!list_empty(&info->block_groups[c]))
4177 				slot_count++;
4178 		}
4179 		up_read(&info->groups_sem);
4180 	}
4181 
4182 	/*
4183 	 * Global block reserve, exported as a space_info
4184 	 */
4185 	slot_count++;
4186 
4187 	/* space_slots == 0 means they are asking for a count */
4188 	if (space_args.space_slots == 0) {
4189 		space_args.total_spaces = slot_count;
4190 		goto out;
4191 	}
4192 
4193 	slot_count = min_t(u64, space_args.space_slots, slot_count);
4194 
4195 	alloc_size = sizeof(*dest) * slot_count;
4196 
4197 	/* we generally have at most 6 or so space infos, one for each raid
4198 	 * level.  So, a whole page should be more than enough for everyone
4199 	 */
4200 	if (alloc_size > PAGE_SIZE)
4201 		return -ENOMEM;
4202 
4203 	space_args.total_spaces = 0;
4204 	dest = kmalloc(alloc_size, GFP_KERNEL);
4205 	if (!dest)
4206 		return -ENOMEM;
4207 	dest_orig = dest;
4208 
4209 	/* now we have a buffer to copy into */
4210 	for (i = 0; i < num_types; i++) {
4211 		struct btrfs_space_info *tmp;
4212 
4213 		if (!slot_count)
4214 			break;
4215 
4216 		info = NULL;
4217 		rcu_read_lock();
4218 		list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
4219 					list) {
4220 			if (tmp->flags == types[i]) {
4221 				info = tmp;
4222 				break;
4223 			}
4224 		}
4225 		rcu_read_unlock();
4226 
4227 		if (!info)
4228 			continue;
4229 		down_read(&info->groups_sem);
4230 		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4231 			if (!list_empty(&info->block_groups[c])) {
4232 				btrfs_get_block_group_info(
4233 					&info->block_groups[c], &space);
4234 				memcpy(dest, &space, sizeof(space));
4235 				dest++;
4236 				space_args.total_spaces++;
4237 				slot_count--;
4238 			}
4239 			if (!slot_count)
4240 				break;
4241 		}
4242 		up_read(&info->groups_sem);
4243 	}
4244 
4245 	/*
4246 	 * Add global block reserve
4247 	 */
4248 	if (slot_count) {
4249 		struct btrfs_block_rsv *block_rsv = &root->fs_info->global_block_rsv;
4250 
4251 		spin_lock(&block_rsv->lock);
4252 		space.total_bytes = block_rsv->size;
4253 		space.used_bytes = block_rsv->size - block_rsv->reserved;
4254 		spin_unlock(&block_rsv->lock);
4255 		space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4256 		memcpy(dest, &space, sizeof(space));
4257 		space_args.total_spaces++;
4258 	}
4259 
4260 	user_dest = (struct btrfs_ioctl_space_info __user *)
4261 		(arg + sizeof(struct btrfs_ioctl_space_args));
4262 
4263 	if (copy_to_user(user_dest, dest_orig, alloc_size))
4264 		ret = -EFAULT;
4265 
4266 	kfree(dest_orig);
4267 out:
4268 	if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4269 		ret = -EFAULT;
4270 
4271 	return ret;
4272 }
4273 
4274 /*
4275  * there are many ways the trans_start and trans_end ioctls can lead
4276  * to deadlocks.  They should only be used by applications that
4277  * basically own the machine, and have a very in depth understanding
4278  * of all the possible deadlocks and enospc problems.
4279  */
4280 long btrfs_ioctl_trans_end(struct file *file)
4281 {
4282 	struct inode *inode = file_inode(file);
4283 	struct btrfs_root *root = BTRFS_I(inode)->root;
4284 	struct btrfs_trans_handle *trans;
4285 
4286 	trans = file->private_data;
4287 	if (!trans)
4288 		return -EINVAL;
4289 	file->private_data = NULL;
4290 
4291 	btrfs_end_transaction(trans, root);
4292 
4293 	atomic_dec(&root->fs_info->open_ioctl_trans);
4294 
4295 	mnt_drop_write_file(file);
4296 	return 0;
4297 }
4298 
4299 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4300 					    void __user *argp)
4301 {
4302 	struct btrfs_trans_handle *trans;
4303 	u64 transid;
4304 	int ret;
4305 
4306 	trans = btrfs_attach_transaction_barrier(root);
4307 	if (IS_ERR(trans)) {
4308 		if (PTR_ERR(trans) != -ENOENT)
4309 			return PTR_ERR(trans);
4310 
4311 		/* No running transaction, don't bother */
4312 		transid = root->fs_info->last_trans_committed;
4313 		goto out;
4314 	}
4315 	transid = trans->transid;
4316 	ret = btrfs_commit_transaction_async(trans, root, 0);
4317 	if (ret) {
4318 		btrfs_end_transaction(trans, root);
4319 		return ret;
4320 	}
4321 out:
4322 	if (argp)
4323 		if (copy_to_user(argp, &transid, sizeof(transid)))
4324 			return -EFAULT;
4325 	return 0;
4326 }
4327 
4328 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
4329 					   void __user *argp)
4330 {
4331 	u64 transid;
4332 
4333 	if (argp) {
4334 		if (copy_from_user(&transid, argp, sizeof(transid)))
4335 			return -EFAULT;
4336 	} else {
4337 		transid = 0;  /* current trans */
4338 	}
4339 	return btrfs_wait_for_commit(root, transid);
4340 }
4341 
4342 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4343 {
4344 	struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4345 	struct btrfs_ioctl_scrub_args *sa;
4346 	int ret;
4347 
4348 	if (!capable(CAP_SYS_ADMIN))
4349 		return -EPERM;
4350 
4351 	sa = memdup_user(arg, sizeof(*sa));
4352 	if (IS_ERR(sa))
4353 		return PTR_ERR(sa);
4354 
4355 	if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4356 		ret = mnt_want_write_file(file);
4357 		if (ret)
4358 			goto out;
4359 	}
4360 
4361 	ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
4362 			      &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4363 			      0);
4364 
4365 	if (copy_to_user(arg, sa, sizeof(*sa)))
4366 		ret = -EFAULT;
4367 
4368 	if (!(sa->flags & BTRFS_SCRUB_READONLY))
4369 		mnt_drop_write_file(file);
4370 out:
4371 	kfree(sa);
4372 	return ret;
4373 }
4374 
4375 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
4376 {
4377 	if (!capable(CAP_SYS_ADMIN))
4378 		return -EPERM;
4379 
4380 	return btrfs_scrub_cancel(root->fs_info);
4381 }
4382 
4383 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
4384 				       void __user *arg)
4385 {
4386 	struct btrfs_ioctl_scrub_args *sa;
4387 	int ret;
4388 
4389 	if (!capable(CAP_SYS_ADMIN))
4390 		return -EPERM;
4391 
4392 	sa = memdup_user(arg, sizeof(*sa));
4393 	if (IS_ERR(sa))
4394 		return PTR_ERR(sa);
4395 
4396 	ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
4397 
4398 	if (copy_to_user(arg, sa, sizeof(*sa)))
4399 		ret = -EFAULT;
4400 
4401 	kfree(sa);
4402 	return ret;
4403 }
4404 
4405 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
4406 				      void __user *arg)
4407 {
4408 	struct btrfs_ioctl_get_dev_stats *sa;
4409 	int ret;
4410 
4411 	sa = memdup_user(arg, sizeof(*sa));
4412 	if (IS_ERR(sa))
4413 		return PTR_ERR(sa);
4414 
4415 	if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4416 		kfree(sa);
4417 		return -EPERM;
4418 	}
4419 
4420 	ret = btrfs_get_dev_stats(root, sa);
4421 
4422 	if (copy_to_user(arg, sa, sizeof(*sa)))
4423 		ret = -EFAULT;
4424 
4425 	kfree(sa);
4426 	return ret;
4427 }
4428 
4429 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
4430 {
4431 	struct btrfs_ioctl_dev_replace_args *p;
4432 	int ret;
4433 
4434 	if (!capable(CAP_SYS_ADMIN))
4435 		return -EPERM;
4436 
4437 	p = memdup_user(arg, sizeof(*p));
4438 	if (IS_ERR(p))
4439 		return PTR_ERR(p);
4440 
4441 	switch (p->cmd) {
4442 	case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4443 		if (root->fs_info->sb->s_flags & MS_RDONLY) {
4444 			ret = -EROFS;
4445 			goto out;
4446 		}
4447 		if (atomic_xchg(
4448 			&root->fs_info->mutually_exclusive_operation_running,
4449 			1)) {
4450 			ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4451 		} else {
4452 			ret = btrfs_dev_replace_by_ioctl(root, p);
4453 			atomic_set(
4454 			 &root->fs_info->mutually_exclusive_operation_running,
4455 			 0);
4456 		}
4457 		break;
4458 	case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4459 		btrfs_dev_replace_status(root->fs_info, p);
4460 		ret = 0;
4461 		break;
4462 	case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4463 		ret = btrfs_dev_replace_cancel(root->fs_info, p);
4464 		break;
4465 	default:
4466 		ret = -EINVAL;
4467 		break;
4468 	}
4469 
4470 	if (copy_to_user(arg, p, sizeof(*p)))
4471 		ret = -EFAULT;
4472 out:
4473 	kfree(p);
4474 	return ret;
4475 }
4476 
4477 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4478 {
4479 	int ret = 0;
4480 	int i;
4481 	u64 rel_ptr;
4482 	int size;
4483 	struct btrfs_ioctl_ino_path_args *ipa = NULL;
4484 	struct inode_fs_paths *ipath = NULL;
4485 	struct btrfs_path *path;
4486 
4487 	if (!capable(CAP_DAC_READ_SEARCH))
4488 		return -EPERM;
4489 
4490 	path = btrfs_alloc_path();
4491 	if (!path) {
4492 		ret = -ENOMEM;
4493 		goto out;
4494 	}
4495 
4496 	ipa = memdup_user(arg, sizeof(*ipa));
4497 	if (IS_ERR(ipa)) {
4498 		ret = PTR_ERR(ipa);
4499 		ipa = NULL;
4500 		goto out;
4501 	}
4502 
4503 	size = min_t(u32, ipa->size, 4096);
4504 	ipath = init_ipath(size, root, path);
4505 	if (IS_ERR(ipath)) {
4506 		ret = PTR_ERR(ipath);
4507 		ipath = NULL;
4508 		goto out;
4509 	}
4510 
4511 	ret = paths_from_inode(ipa->inum, ipath);
4512 	if (ret < 0)
4513 		goto out;
4514 
4515 	for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4516 		rel_ptr = ipath->fspath->val[i] -
4517 			  (u64)(unsigned long)ipath->fspath->val;
4518 		ipath->fspath->val[i] = rel_ptr;
4519 	}
4520 
4521 	ret = copy_to_user((void *)(unsigned long)ipa->fspath,
4522 			   (void *)(unsigned long)ipath->fspath, size);
4523 	if (ret) {
4524 		ret = -EFAULT;
4525 		goto out;
4526 	}
4527 
4528 out:
4529 	btrfs_free_path(path);
4530 	free_ipath(ipath);
4531 	kfree(ipa);
4532 
4533 	return ret;
4534 }
4535 
4536 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4537 {
4538 	struct btrfs_data_container *inodes = ctx;
4539 	const size_t c = 3 * sizeof(u64);
4540 
4541 	if (inodes->bytes_left >= c) {
4542 		inodes->bytes_left -= c;
4543 		inodes->val[inodes->elem_cnt] = inum;
4544 		inodes->val[inodes->elem_cnt + 1] = offset;
4545 		inodes->val[inodes->elem_cnt + 2] = root;
4546 		inodes->elem_cnt += 3;
4547 	} else {
4548 		inodes->bytes_missing += c - inodes->bytes_left;
4549 		inodes->bytes_left = 0;
4550 		inodes->elem_missed += 3;
4551 	}
4552 
4553 	return 0;
4554 }
4555 
4556 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
4557 					void __user *arg)
4558 {
4559 	int ret = 0;
4560 	int size;
4561 	struct btrfs_ioctl_logical_ino_args *loi;
4562 	struct btrfs_data_container *inodes = NULL;
4563 	struct btrfs_path *path = NULL;
4564 
4565 	if (!capable(CAP_SYS_ADMIN))
4566 		return -EPERM;
4567 
4568 	loi = memdup_user(arg, sizeof(*loi));
4569 	if (IS_ERR(loi)) {
4570 		ret = PTR_ERR(loi);
4571 		loi = NULL;
4572 		goto out;
4573 	}
4574 
4575 	path = btrfs_alloc_path();
4576 	if (!path) {
4577 		ret = -ENOMEM;
4578 		goto out;
4579 	}
4580 
4581 	size = min_t(u32, loi->size, SZ_64K);
4582 	inodes = init_data_container(size);
4583 	if (IS_ERR(inodes)) {
4584 		ret = PTR_ERR(inodes);
4585 		inodes = NULL;
4586 		goto out;
4587 	}
4588 
4589 	ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
4590 					  build_ino_list, inodes);
4591 	if (ret == -EINVAL)
4592 		ret = -ENOENT;
4593 	if (ret < 0)
4594 		goto out;
4595 
4596 	ret = copy_to_user((void *)(unsigned long)loi->inodes,
4597 			   (void *)(unsigned long)inodes, size);
4598 	if (ret)
4599 		ret = -EFAULT;
4600 
4601 out:
4602 	btrfs_free_path(path);
4603 	vfree(inodes);
4604 	kfree(loi);
4605 
4606 	return ret;
4607 }
4608 
4609 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4610 			       struct btrfs_ioctl_balance_args *bargs)
4611 {
4612 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4613 
4614 	bargs->flags = bctl->flags;
4615 
4616 	if (atomic_read(&fs_info->balance_running))
4617 		bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4618 	if (atomic_read(&fs_info->balance_pause_req))
4619 		bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4620 	if (atomic_read(&fs_info->balance_cancel_req))
4621 		bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4622 
4623 	memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4624 	memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4625 	memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4626 
4627 	if (lock) {
4628 		spin_lock(&fs_info->balance_lock);
4629 		memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4630 		spin_unlock(&fs_info->balance_lock);
4631 	} else {
4632 		memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4633 	}
4634 }
4635 
4636 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4637 {
4638 	struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4639 	struct btrfs_fs_info *fs_info = root->fs_info;
4640 	struct btrfs_ioctl_balance_args *bargs;
4641 	struct btrfs_balance_control *bctl;
4642 	bool need_unlock; /* for mut. excl. ops lock */
4643 	int ret;
4644 
4645 	if (!capable(CAP_SYS_ADMIN))
4646 		return -EPERM;
4647 
4648 	ret = mnt_want_write_file(file);
4649 	if (ret)
4650 		return ret;
4651 
4652 again:
4653 	if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
4654 		mutex_lock(&fs_info->volume_mutex);
4655 		mutex_lock(&fs_info->balance_mutex);
4656 		need_unlock = true;
4657 		goto locked;
4658 	}
4659 
4660 	/*
4661 	 * mut. excl. ops lock is locked.  Three possibilities:
4662 	 *   (1) some other op is running
4663 	 *   (2) balance is running
4664 	 *   (3) balance is paused -- special case (think resume)
4665 	 */
4666 	mutex_lock(&fs_info->balance_mutex);
4667 	if (fs_info->balance_ctl) {
4668 		/* this is either (2) or (3) */
4669 		if (!atomic_read(&fs_info->balance_running)) {
4670 			mutex_unlock(&fs_info->balance_mutex);
4671 			if (!mutex_trylock(&fs_info->volume_mutex))
4672 				goto again;
4673 			mutex_lock(&fs_info->balance_mutex);
4674 
4675 			if (fs_info->balance_ctl &&
4676 			    !atomic_read(&fs_info->balance_running)) {
4677 				/* this is (3) */
4678 				need_unlock = false;
4679 				goto locked;
4680 			}
4681 
4682 			mutex_unlock(&fs_info->balance_mutex);
4683 			mutex_unlock(&fs_info->volume_mutex);
4684 			goto again;
4685 		} else {
4686 			/* this is (2) */
4687 			mutex_unlock(&fs_info->balance_mutex);
4688 			ret = -EINPROGRESS;
4689 			goto out;
4690 		}
4691 	} else {
4692 		/* this is (1) */
4693 		mutex_unlock(&fs_info->balance_mutex);
4694 		ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4695 		goto out;
4696 	}
4697 
4698 locked:
4699 	BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
4700 
4701 	if (arg) {
4702 		bargs = memdup_user(arg, sizeof(*bargs));
4703 		if (IS_ERR(bargs)) {
4704 			ret = PTR_ERR(bargs);
4705 			goto out_unlock;
4706 		}
4707 
4708 		if (bargs->flags & BTRFS_BALANCE_RESUME) {
4709 			if (!fs_info->balance_ctl) {
4710 				ret = -ENOTCONN;
4711 				goto out_bargs;
4712 			}
4713 
4714 			bctl = fs_info->balance_ctl;
4715 			spin_lock(&fs_info->balance_lock);
4716 			bctl->flags |= BTRFS_BALANCE_RESUME;
4717 			spin_unlock(&fs_info->balance_lock);
4718 
4719 			goto do_balance;
4720 		}
4721 	} else {
4722 		bargs = NULL;
4723 	}
4724 
4725 	if (fs_info->balance_ctl) {
4726 		ret = -EINPROGRESS;
4727 		goto out_bargs;
4728 	}
4729 
4730 	bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4731 	if (!bctl) {
4732 		ret = -ENOMEM;
4733 		goto out_bargs;
4734 	}
4735 
4736 	bctl->fs_info = fs_info;
4737 	if (arg) {
4738 		memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4739 		memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4740 		memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4741 
4742 		bctl->flags = bargs->flags;
4743 	} else {
4744 		/* balance everything - no filters */
4745 		bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4746 	}
4747 
4748 	if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4749 		ret = -EINVAL;
4750 		goto out_bctl;
4751 	}
4752 
4753 do_balance:
4754 	/*
4755 	 * Ownership of bctl and mutually_exclusive_operation_running
4756 	 * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4757 	 * or, if restriper was paused all the way until unmount, in
4758 	 * free_fs_info.  mutually_exclusive_operation_running is
4759 	 * cleared in __cancel_balance.
4760 	 */
4761 	need_unlock = false;
4762 
4763 	ret = btrfs_balance(bctl, bargs);
4764 	bctl = NULL;
4765 
4766 	if (arg) {
4767 		if (copy_to_user(arg, bargs, sizeof(*bargs)))
4768 			ret = -EFAULT;
4769 	}
4770 
4771 out_bctl:
4772 	kfree(bctl);
4773 out_bargs:
4774 	kfree(bargs);
4775 out_unlock:
4776 	mutex_unlock(&fs_info->balance_mutex);
4777 	mutex_unlock(&fs_info->volume_mutex);
4778 	if (need_unlock)
4779 		atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
4780 out:
4781 	mnt_drop_write_file(file);
4782 	return ret;
4783 }
4784 
4785 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
4786 {
4787 	if (!capable(CAP_SYS_ADMIN))
4788 		return -EPERM;
4789 
4790 	switch (cmd) {
4791 	case BTRFS_BALANCE_CTL_PAUSE:
4792 		return btrfs_pause_balance(root->fs_info);
4793 	case BTRFS_BALANCE_CTL_CANCEL:
4794 		return btrfs_cancel_balance(root->fs_info);
4795 	}
4796 
4797 	return -EINVAL;
4798 }
4799 
4800 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
4801 					 void __user *arg)
4802 {
4803 	struct btrfs_fs_info *fs_info = root->fs_info;
4804 	struct btrfs_ioctl_balance_args *bargs;
4805 	int ret = 0;
4806 
4807 	if (!capable(CAP_SYS_ADMIN))
4808 		return -EPERM;
4809 
4810 	mutex_lock(&fs_info->balance_mutex);
4811 	if (!fs_info->balance_ctl) {
4812 		ret = -ENOTCONN;
4813 		goto out;
4814 	}
4815 
4816 	bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4817 	if (!bargs) {
4818 		ret = -ENOMEM;
4819 		goto out;
4820 	}
4821 
4822 	update_ioctl_balance_args(fs_info, 1, bargs);
4823 
4824 	if (copy_to_user(arg, bargs, sizeof(*bargs)))
4825 		ret = -EFAULT;
4826 
4827 	kfree(bargs);
4828 out:
4829 	mutex_unlock(&fs_info->balance_mutex);
4830 	return ret;
4831 }
4832 
4833 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4834 {
4835 	struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4836 	struct btrfs_ioctl_quota_ctl_args *sa;
4837 	struct btrfs_trans_handle *trans = NULL;
4838 	int ret;
4839 	int err;
4840 
4841 	if (!capable(CAP_SYS_ADMIN))
4842 		return -EPERM;
4843 
4844 	ret = mnt_want_write_file(file);
4845 	if (ret)
4846 		return ret;
4847 
4848 	sa = memdup_user(arg, sizeof(*sa));
4849 	if (IS_ERR(sa)) {
4850 		ret = PTR_ERR(sa);
4851 		goto drop_write;
4852 	}
4853 
4854 	down_write(&root->fs_info->subvol_sem);
4855 	trans = btrfs_start_transaction(root->fs_info->tree_root, 2);
4856 	if (IS_ERR(trans)) {
4857 		ret = PTR_ERR(trans);
4858 		goto out;
4859 	}
4860 
4861 	switch (sa->cmd) {
4862 	case BTRFS_QUOTA_CTL_ENABLE:
4863 		ret = btrfs_quota_enable(trans, root->fs_info);
4864 		break;
4865 	case BTRFS_QUOTA_CTL_DISABLE:
4866 		ret = btrfs_quota_disable(trans, root->fs_info);
4867 		break;
4868 	default:
4869 		ret = -EINVAL;
4870 		break;
4871 	}
4872 
4873 	err = btrfs_commit_transaction(trans, root->fs_info->tree_root);
4874 	if (err && !ret)
4875 		ret = err;
4876 out:
4877 	kfree(sa);
4878 	up_write(&root->fs_info->subvol_sem);
4879 drop_write:
4880 	mnt_drop_write_file(file);
4881 	return ret;
4882 }
4883 
4884 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4885 {
4886 	struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4887 	struct btrfs_ioctl_qgroup_assign_args *sa;
4888 	struct btrfs_trans_handle *trans;
4889 	int ret;
4890 	int err;
4891 
4892 	if (!capable(CAP_SYS_ADMIN))
4893 		return -EPERM;
4894 
4895 	ret = mnt_want_write_file(file);
4896 	if (ret)
4897 		return ret;
4898 
4899 	sa = memdup_user(arg, sizeof(*sa));
4900 	if (IS_ERR(sa)) {
4901 		ret = PTR_ERR(sa);
4902 		goto drop_write;
4903 	}
4904 
4905 	trans = btrfs_join_transaction(root);
4906 	if (IS_ERR(trans)) {
4907 		ret = PTR_ERR(trans);
4908 		goto out;
4909 	}
4910 
4911 	/* FIXME: check if the IDs really exist */
4912 	if (sa->assign) {
4913 		ret = btrfs_add_qgroup_relation(trans, root->fs_info,
4914 						sa->src, sa->dst);
4915 	} else {
4916 		ret = btrfs_del_qgroup_relation(trans, root->fs_info,
4917 						sa->src, sa->dst);
4918 	}
4919 
4920 	/* update qgroup status and info */
4921 	err = btrfs_run_qgroups(trans, root->fs_info);
4922 	if (err < 0)
4923 		btrfs_handle_fs_error(root->fs_info, err,
4924 			    "failed to update qgroup status and info");
4925 	err = btrfs_end_transaction(trans, root);
4926 	if (err && !ret)
4927 		ret = err;
4928 
4929 out:
4930 	kfree(sa);
4931 drop_write:
4932 	mnt_drop_write_file(file);
4933 	return ret;
4934 }
4935 
4936 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4937 {
4938 	struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4939 	struct btrfs_ioctl_qgroup_create_args *sa;
4940 	struct btrfs_trans_handle *trans;
4941 	int ret;
4942 	int err;
4943 
4944 	if (!capable(CAP_SYS_ADMIN))
4945 		return -EPERM;
4946 
4947 	ret = mnt_want_write_file(file);
4948 	if (ret)
4949 		return ret;
4950 
4951 	sa = memdup_user(arg, sizeof(*sa));
4952 	if (IS_ERR(sa)) {
4953 		ret = PTR_ERR(sa);
4954 		goto drop_write;
4955 	}
4956 
4957 	if (!sa->qgroupid) {
4958 		ret = -EINVAL;
4959 		goto out;
4960 	}
4961 
4962 	trans = btrfs_join_transaction(root);
4963 	if (IS_ERR(trans)) {
4964 		ret = PTR_ERR(trans);
4965 		goto out;
4966 	}
4967 
4968 	/* FIXME: check if the IDs really exist */
4969 	if (sa->create) {
4970 		ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid);
4971 	} else {
4972 		ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
4973 	}
4974 
4975 	err = btrfs_end_transaction(trans, root);
4976 	if (err && !ret)
4977 		ret = err;
4978 
4979 out:
4980 	kfree(sa);
4981 drop_write:
4982 	mnt_drop_write_file(file);
4983 	return ret;
4984 }
4985 
4986 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4987 {
4988 	struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4989 	struct btrfs_ioctl_qgroup_limit_args *sa;
4990 	struct btrfs_trans_handle *trans;
4991 	int ret;
4992 	int err;
4993 	u64 qgroupid;
4994 
4995 	if (!capable(CAP_SYS_ADMIN))
4996 		return -EPERM;
4997 
4998 	ret = mnt_want_write_file(file);
4999 	if (ret)
5000 		return ret;
5001 
5002 	sa = memdup_user(arg, sizeof(*sa));
5003 	if (IS_ERR(sa)) {
5004 		ret = PTR_ERR(sa);
5005 		goto drop_write;
5006 	}
5007 
5008 	trans = btrfs_join_transaction(root);
5009 	if (IS_ERR(trans)) {
5010 		ret = PTR_ERR(trans);
5011 		goto out;
5012 	}
5013 
5014 	qgroupid = sa->qgroupid;
5015 	if (!qgroupid) {
5016 		/* take the current subvol as qgroup */
5017 		qgroupid = root->root_key.objectid;
5018 	}
5019 
5020 	/* FIXME: check if the IDs really exist */
5021 	ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
5022 
5023 	err = btrfs_end_transaction(trans, root);
5024 	if (err && !ret)
5025 		ret = err;
5026 
5027 out:
5028 	kfree(sa);
5029 drop_write:
5030 	mnt_drop_write_file(file);
5031 	return ret;
5032 }
5033 
5034 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
5035 {
5036 	struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5037 	struct btrfs_ioctl_quota_rescan_args *qsa;
5038 	int ret;
5039 
5040 	if (!capable(CAP_SYS_ADMIN))
5041 		return -EPERM;
5042 
5043 	ret = mnt_want_write_file(file);
5044 	if (ret)
5045 		return ret;
5046 
5047 	qsa = memdup_user(arg, sizeof(*qsa));
5048 	if (IS_ERR(qsa)) {
5049 		ret = PTR_ERR(qsa);
5050 		goto drop_write;
5051 	}
5052 
5053 	if (qsa->flags) {
5054 		ret = -EINVAL;
5055 		goto out;
5056 	}
5057 
5058 	ret = btrfs_qgroup_rescan(root->fs_info);
5059 
5060 out:
5061 	kfree(qsa);
5062 drop_write:
5063 	mnt_drop_write_file(file);
5064 	return ret;
5065 }
5066 
5067 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
5068 {
5069 	struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5070 	struct btrfs_ioctl_quota_rescan_args *qsa;
5071 	int ret = 0;
5072 
5073 	if (!capable(CAP_SYS_ADMIN))
5074 		return -EPERM;
5075 
5076 	qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
5077 	if (!qsa)
5078 		return -ENOMEM;
5079 
5080 	if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
5081 		qsa->flags = 1;
5082 		qsa->progress = root->fs_info->qgroup_rescan_progress.objectid;
5083 	}
5084 
5085 	if (copy_to_user(arg, qsa, sizeof(*qsa)))
5086 		ret = -EFAULT;
5087 
5088 	kfree(qsa);
5089 	return ret;
5090 }
5091 
5092 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
5093 {
5094 	struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5095 
5096 	if (!capable(CAP_SYS_ADMIN))
5097 		return -EPERM;
5098 
5099 	return btrfs_qgroup_wait_for_completion(root->fs_info, true);
5100 }
5101 
5102 static long _btrfs_ioctl_set_received_subvol(struct file *file,
5103 					    struct btrfs_ioctl_received_subvol_args *sa)
5104 {
5105 	struct inode *inode = file_inode(file);
5106 	struct btrfs_root *root = BTRFS_I(inode)->root;
5107 	struct btrfs_root_item *root_item = &root->root_item;
5108 	struct btrfs_trans_handle *trans;
5109 	struct timespec ct = current_fs_time(inode->i_sb);
5110 	int ret = 0;
5111 	int received_uuid_changed;
5112 
5113 	if (!inode_owner_or_capable(inode))
5114 		return -EPERM;
5115 
5116 	ret = mnt_want_write_file(file);
5117 	if (ret < 0)
5118 		return ret;
5119 
5120 	down_write(&root->fs_info->subvol_sem);
5121 
5122 	if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
5123 		ret = -EINVAL;
5124 		goto out;
5125 	}
5126 
5127 	if (btrfs_root_readonly(root)) {
5128 		ret = -EROFS;
5129 		goto out;
5130 	}
5131 
5132 	/*
5133 	 * 1 - root item
5134 	 * 2 - uuid items (received uuid + subvol uuid)
5135 	 */
5136 	trans = btrfs_start_transaction(root, 3);
5137 	if (IS_ERR(trans)) {
5138 		ret = PTR_ERR(trans);
5139 		trans = NULL;
5140 		goto out;
5141 	}
5142 
5143 	sa->rtransid = trans->transid;
5144 	sa->rtime.sec = ct.tv_sec;
5145 	sa->rtime.nsec = ct.tv_nsec;
5146 
5147 	received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5148 				       BTRFS_UUID_SIZE);
5149 	if (received_uuid_changed &&
5150 	    !btrfs_is_empty_uuid(root_item->received_uuid))
5151 		btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
5152 				    root_item->received_uuid,
5153 				    BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5154 				    root->root_key.objectid);
5155 	memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5156 	btrfs_set_root_stransid(root_item, sa->stransid);
5157 	btrfs_set_root_rtransid(root_item, sa->rtransid);
5158 	btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5159 	btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5160 	btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5161 	btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5162 
5163 	ret = btrfs_update_root(trans, root->fs_info->tree_root,
5164 				&root->root_key, &root->root_item);
5165 	if (ret < 0) {
5166 		btrfs_end_transaction(trans, root);
5167 		goto out;
5168 	}
5169 	if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5170 		ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
5171 					  sa->uuid,
5172 					  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5173 					  root->root_key.objectid);
5174 		if (ret < 0 && ret != -EEXIST) {
5175 			btrfs_abort_transaction(trans, ret);
5176 			goto out;
5177 		}
5178 	}
5179 	ret = btrfs_commit_transaction(trans, root);
5180 	if (ret < 0) {
5181 		btrfs_abort_transaction(trans, ret);
5182 		goto out;
5183 	}
5184 
5185 out:
5186 	up_write(&root->fs_info->subvol_sem);
5187 	mnt_drop_write_file(file);
5188 	return ret;
5189 }
5190 
5191 #ifdef CONFIG_64BIT
5192 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5193 						void __user *arg)
5194 {
5195 	struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5196 	struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5197 	int ret = 0;
5198 
5199 	args32 = memdup_user(arg, sizeof(*args32));
5200 	if (IS_ERR(args32)) {
5201 		ret = PTR_ERR(args32);
5202 		args32 = NULL;
5203 		goto out;
5204 	}
5205 
5206 	args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
5207 	if (!args64) {
5208 		ret = -ENOMEM;
5209 		goto out;
5210 	}
5211 
5212 	memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5213 	args64->stransid = args32->stransid;
5214 	args64->rtransid = args32->rtransid;
5215 	args64->stime.sec = args32->stime.sec;
5216 	args64->stime.nsec = args32->stime.nsec;
5217 	args64->rtime.sec = args32->rtime.sec;
5218 	args64->rtime.nsec = args32->rtime.nsec;
5219 	args64->flags = args32->flags;
5220 
5221 	ret = _btrfs_ioctl_set_received_subvol(file, args64);
5222 	if (ret)
5223 		goto out;
5224 
5225 	memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5226 	args32->stransid = args64->stransid;
5227 	args32->rtransid = args64->rtransid;
5228 	args32->stime.sec = args64->stime.sec;
5229 	args32->stime.nsec = args64->stime.nsec;
5230 	args32->rtime.sec = args64->rtime.sec;
5231 	args32->rtime.nsec = args64->rtime.nsec;
5232 	args32->flags = args64->flags;
5233 
5234 	ret = copy_to_user(arg, args32, sizeof(*args32));
5235 	if (ret)
5236 		ret = -EFAULT;
5237 
5238 out:
5239 	kfree(args32);
5240 	kfree(args64);
5241 	return ret;
5242 }
5243 #endif
5244 
5245 static long btrfs_ioctl_set_received_subvol(struct file *file,
5246 					    void __user *arg)
5247 {
5248 	struct btrfs_ioctl_received_subvol_args *sa = NULL;
5249 	int ret = 0;
5250 
5251 	sa = memdup_user(arg, sizeof(*sa));
5252 	if (IS_ERR(sa)) {
5253 		ret = PTR_ERR(sa);
5254 		sa = NULL;
5255 		goto out;
5256 	}
5257 
5258 	ret = _btrfs_ioctl_set_received_subvol(file, sa);
5259 
5260 	if (ret)
5261 		goto out;
5262 
5263 	ret = copy_to_user(arg, sa, sizeof(*sa));
5264 	if (ret)
5265 		ret = -EFAULT;
5266 
5267 out:
5268 	kfree(sa);
5269 	return ret;
5270 }
5271 
5272 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5273 {
5274 	struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5275 	size_t len;
5276 	int ret;
5277 	char label[BTRFS_LABEL_SIZE];
5278 
5279 	spin_lock(&root->fs_info->super_lock);
5280 	memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5281 	spin_unlock(&root->fs_info->super_lock);
5282 
5283 	len = strnlen(label, BTRFS_LABEL_SIZE);
5284 
5285 	if (len == BTRFS_LABEL_SIZE) {
5286 		btrfs_warn(root->fs_info,
5287 			"label is too long, return the first %zu bytes", --len);
5288 	}
5289 
5290 	ret = copy_to_user(arg, label, len);
5291 
5292 	return ret ? -EFAULT : 0;
5293 }
5294 
5295 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5296 {
5297 	struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5298 	struct btrfs_super_block *super_block = root->fs_info->super_copy;
5299 	struct btrfs_trans_handle *trans;
5300 	char label[BTRFS_LABEL_SIZE];
5301 	int ret;
5302 
5303 	if (!capable(CAP_SYS_ADMIN))
5304 		return -EPERM;
5305 
5306 	if (copy_from_user(label, arg, sizeof(label)))
5307 		return -EFAULT;
5308 
5309 	if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5310 		btrfs_err(root->fs_info, "unable to set label with more than %d bytes",
5311 		       BTRFS_LABEL_SIZE - 1);
5312 		return -EINVAL;
5313 	}
5314 
5315 	ret = mnt_want_write_file(file);
5316 	if (ret)
5317 		return ret;
5318 
5319 	trans = btrfs_start_transaction(root, 0);
5320 	if (IS_ERR(trans)) {
5321 		ret = PTR_ERR(trans);
5322 		goto out_unlock;
5323 	}
5324 
5325 	spin_lock(&root->fs_info->super_lock);
5326 	strcpy(super_block->label, label);
5327 	spin_unlock(&root->fs_info->super_lock);
5328 	ret = btrfs_commit_transaction(trans, root);
5329 
5330 out_unlock:
5331 	mnt_drop_write_file(file);
5332 	return ret;
5333 }
5334 
5335 #define INIT_FEATURE_FLAGS(suffix) \
5336 	{ .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5337 	  .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5338 	  .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5339 
5340 int btrfs_ioctl_get_supported_features(void __user *arg)
5341 {
5342 	static const struct btrfs_ioctl_feature_flags features[3] = {
5343 		INIT_FEATURE_FLAGS(SUPP),
5344 		INIT_FEATURE_FLAGS(SAFE_SET),
5345 		INIT_FEATURE_FLAGS(SAFE_CLEAR)
5346 	};
5347 
5348 	if (copy_to_user(arg, &features, sizeof(features)))
5349 		return -EFAULT;
5350 
5351 	return 0;
5352 }
5353 
5354 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5355 {
5356 	struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5357 	struct btrfs_super_block *super_block = root->fs_info->super_copy;
5358 	struct btrfs_ioctl_feature_flags features;
5359 
5360 	features.compat_flags = btrfs_super_compat_flags(super_block);
5361 	features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5362 	features.incompat_flags = btrfs_super_incompat_flags(super_block);
5363 
5364 	if (copy_to_user(arg, &features, sizeof(features)))
5365 		return -EFAULT;
5366 
5367 	return 0;
5368 }
5369 
5370 static int check_feature_bits(struct btrfs_root *root,
5371 			      enum btrfs_feature_set set,
5372 			      u64 change_mask, u64 flags, u64 supported_flags,
5373 			      u64 safe_set, u64 safe_clear)
5374 {
5375 	const char *type = btrfs_feature_set_names[set];
5376 	char *names;
5377 	u64 disallowed, unsupported;
5378 	u64 set_mask = flags & change_mask;
5379 	u64 clear_mask = ~flags & change_mask;
5380 
5381 	unsupported = set_mask & ~supported_flags;
5382 	if (unsupported) {
5383 		names = btrfs_printable_features(set, unsupported);
5384 		if (names) {
5385 			btrfs_warn(root->fs_info,
5386 			   "this kernel does not support the %s feature bit%s",
5387 			   names, strchr(names, ',') ? "s" : "");
5388 			kfree(names);
5389 		} else
5390 			btrfs_warn(root->fs_info,
5391 			   "this kernel does not support %s bits 0x%llx",
5392 			   type, unsupported);
5393 		return -EOPNOTSUPP;
5394 	}
5395 
5396 	disallowed = set_mask & ~safe_set;
5397 	if (disallowed) {
5398 		names = btrfs_printable_features(set, disallowed);
5399 		if (names) {
5400 			btrfs_warn(root->fs_info,
5401 			   "can't set the %s feature bit%s while mounted",
5402 			   names, strchr(names, ',') ? "s" : "");
5403 			kfree(names);
5404 		} else
5405 			btrfs_warn(root->fs_info,
5406 			   "can't set %s bits 0x%llx while mounted",
5407 			   type, disallowed);
5408 		return -EPERM;
5409 	}
5410 
5411 	disallowed = clear_mask & ~safe_clear;
5412 	if (disallowed) {
5413 		names = btrfs_printable_features(set, disallowed);
5414 		if (names) {
5415 			btrfs_warn(root->fs_info,
5416 			   "can't clear the %s feature bit%s while mounted",
5417 			   names, strchr(names, ',') ? "s" : "");
5418 			kfree(names);
5419 		} else
5420 			btrfs_warn(root->fs_info,
5421 			   "can't clear %s bits 0x%llx while mounted",
5422 			   type, disallowed);
5423 		return -EPERM;
5424 	}
5425 
5426 	return 0;
5427 }
5428 
5429 #define check_feature(root, change_mask, flags, mask_base)	\
5430 check_feature_bits(root, FEAT_##mask_base, change_mask, flags,	\
5431 		   BTRFS_FEATURE_ ## mask_base ## _SUPP,	\
5432 		   BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,	\
5433 		   BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5434 
5435 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5436 {
5437 	struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5438 	struct btrfs_super_block *super_block = root->fs_info->super_copy;
5439 	struct btrfs_ioctl_feature_flags flags[2];
5440 	struct btrfs_trans_handle *trans;
5441 	u64 newflags;
5442 	int ret;
5443 
5444 	if (!capable(CAP_SYS_ADMIN))
5445 		return -EPERM;
5446 
5447 	if (copy_from_user(flags, arg, sizeof(flags)))
5448 		return -EFAULT;
5449 
5450 	/* Nothing to do */
5451 	if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5452 	    !flags[0].incompat_flags)
5453 		return 0;
5454 
5455 	ret = check_feature(root, flags[0].compat_flags,
5456 			    flags[1].compat_flags, COMPAT);
5457 	if (ret)
5458 		return ret;
5459 
5460 	ret = check_feature(root, flags[0].compat_ro_flags,
5461 			    flags[1].compat_ro_flags, COMPAT_RO);
5462 	if (ret)
5463 		return ret;
5464 
5465 	ret = check_feature(root, flags[0].incompat_flags,
5466 			    flags[1].incompat_flags, INCOMPAT);
5467 	if (ret)
5468 		return ret;
5469 
5470 	ret = mnt_want_write_file(file);
5471 	if (ret)
5472 		return ret;
5473 
5474 	trans = btrfs_start_transaction(root, 0);
5475 	if (IS_ERR(trans)) {
5476 		ret = PTR_ERR(trans);
5477 		goto out_drop_write;
5478 	}
5479 
5480 	spin_lock(&root->fs_info->super_lock);
5481 	newflags = btrfs_super_compat_flags(super_block);
5482 	newflags |= flags[0].compat_flags & flags[1].compat_flags;
5483 	newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5484 	btrfs_set_super_compat_flags(super_block, newflags);
5485 
5486 	newflags = btrfs_super_compat_ro_flags(super_block);
5487 	newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5488 	newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5489 	btrfs_set_super_compat_ro_flags(super_block, newflags);
5490 
5491 	newflags = btrfs_super_incompat_flags(super_block);
5492 	newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5493 	newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5494 	btrfs_set_super_incompat_flags(super_block, newflags);
5495 	spin_unlock(&root->fs_info->super_lock);
5496 
5497 	ret = btrfs_commit_transaction(trans, root);
5498 out_drop_write:
5499 	mnt_drop_write_file(file);
5500 
5501 	return ret;
5502 }
5503 
5504 long btrfs_ioctl(struct file *file, unsigned int
5505 		cmd, unsigned long arg)
5506 {
5507 	struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5508 	void __user *argp = (void __user *)arg;
5509 
5510 	switch (cmd) {
5511 	case FS_IOC_GETFLAGS:
5512 		return btrfs_ioctl_getflags(file, argp);
5513 	case FS_IOC_SETFLAGS:
5514 		return btrfs_ioctl_setflags(file, argp);
5515 	case FS_IOC_GETVERSION:
5516 		return btrfs_ioctl_getversion(file, argp);
5517 	case FITRIM:
5518 		return btrfs_ioctl_fitrim(file, argp);
5519 	case BTRFS_IOC_SNAP_CREATE:
5520 		return btrfs_ioctl_snap_create(file, argp, 0);
5521 	case BTRFS_IOC_SNAP_CREATE_V2:
5522 		return btrfs_ioctl_snap_create_v2(file, argp, 0);
5523 	case BTRFS_IOC_SUBVOL_CREATE:
5524 		return btrfs_ioctl_snap_create(file, argp, 1);
5525 	case BTRFS_IOC_SUBVOL_CREATE_V2:
5526 		return btrfs_ioctl_snap_create_v2(file, argp, 1);
5527 	case BTRFS_IOC_SNAP_DESTROY:
5528 		return btrfs_ioctl_snap_destroy(file, argp);
5529 	case BTRFS_IOC_SUBVOL_GETFLAGS:
5530 		return btrfs_ioctl_subvol_getflags(file, argp);
5531 	case BTRFS_IOC_SUBVOL_SETFLAGS:
5532 		return btrfs_ioctl_subvol_setflags(file, argp);
5533 	case BTRFS_IOC_DEFAULT_SUBVOL:
5534 		return btrfs_ioctl_default_subvol(file, argp);
5535 	case BTRFS_IOC_DEFRAG:
5536 		return btrfs_ioctl_defrag(file, NULL);
5537 	case BTRFS_IOC_DEFRAG_RANGE:
5538 		return btrfs_ioctl_defrag(file, argp);
5539 	case BTRFS_IOC_RESIZE:
5540 		return btrfs_ioctl_resize(file, argp);
5541 	case BTRFS_IOC_ADD_DEV:
5542 		return btrfs_ioctl_add_dev(root, argp);
5543 	case BTRFS_IOC_RM_DEV:
5544 		return btrfs_ioctl_rm_dev(file, argp);
5545 	case BTRFS_IOC_RM_DEV_V2:
5546 		return btrfs_ioctl_rm_dev_v2(file, argp);
5547 	case BTRFS_IOC_FS_INFO:
5548 		return btrfs_ioctl_fs_info(root, argp);
5549 	case BTRFS_IOC_DEV_INFO:
5550 		return btrfs_ioctl_dev_info(root, argp);
5551 	case BTRFS_IOC_BALANCE:
5552 		return btrfs_ioctl_balance(file, NULL);
5553 	case BTRFS_IOC_TRANS_START:
5554 		return btrfs_ioctl_trans_start(file);
5555 	case BTRFS_IOC_TRANS_END:
5556 		return btrfs_ioctl_trans_end(file);
5557 	case BTRFS_IOC_TREE_SEARCH:
5558 		return btrfs_ioctl_tree_search(file, argp);
5559 	case BTRFS_IOC_TREE_SEARCH_V2:
5560 		return btrfs_ioctl_tree_search_v2(file, argp);
5561 	case BTRFS_IOC_INO_LOOKUP:
5562 		return btrfs_ioctl_ino_lookup(file, argp);
5563 	case BTRFS_IOC_INO_PATHS:
5564 		return btrfs_ioctl_ino_to_path(root, argp);
5565 	case BTRFS_IOC_LOGICAL_INO:
5566 		return btrfs_ioctl_logical_to_ino(root, argp);
5567 	case BTRFS_IOC_SPACE_INFO:
5568 		return btrfs_ioctl_space_info(root, argp);
5569 	case BTRFS_IOC_SYNC: {
5570 		int ret;
5571 
5572 		ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
5573 		if (ret)
5574 			return ret;
5575 		ret = btrfs_sync_fs(file_inode(file)->i_sb, 1);
5576 		/*
5577 		 * The transaction thread may want to do more work,
5578 		 * namely it pokes the cleaner kthread that will start
5579 		 * processing uncleaned subvols.
5580 		 */
5581 		wake_up_process(root->fs_info->transaction_kthread);
5582 		return ret;
5583 	}
5584 	case BTRFS_IOC_START_SYNC:
5585 		return btrfs_ioctl_start_sync(root, argp);
5586 	case BTRFS_IOC_WAIT_SYNC:
5587 		return btrfs_ioctl_wait_sync(root, argp);
5588 	case BTRFS_IOC_SCRUB:
5589 		return btrfs_ioctl_scrub(file, argp);
5590 	case BTRFS_IOC_SCRUB_CANCEL:
5591 		return btrfs_ioctl_scrub_cancel(root, argp);
5592 	case BTRFS_IOC_SCRUB_PROGRESS:
5593 		return btrfs_ioctl_scrub_progress(root, argp);
5594 	case BTRFS_IOC_BALANCE_V2:
5595 		return btrfs_ioctl_balance(file, argp);
5596 	case BTRFS_IOC_BALANCE_CTL:
5597 		return btrfs_ioctl_balance_ctl(root, arg);
5598 	case BTRFS_IOC_BALANCE_PROGRESS:
5599 		return btrfs_ioctl_balance_progress(root, argp);
5600 	case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5601 		return btrfs_ioctl_set_received_subvol(file, argp);
5602 #ifdef CONFIG_64BIT
5603 	case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5604 		return btrfs_ioctl_set_received_subvol_32(file, argp);
5605 #endif
5606 	case BTRFS_IOC_SEND:
5607 		return btrfs_ioctl_send(file, argp);
5608 	case BTRFS_IOC_GET_DEV_STATS:
5609 		return btrfs_ioctl_get_dev_stats(root, argp);
5610 	case BTRFS_IOC_QUOTA_CTL:
5611 		return btrfs_ioctl_quota_ctl(file, argp);
5612 	case BTRFS_IOC_QGROUP_ASSIGN:
5613 		return btrfs_ioctl_qgroup_assign(file, argp);
5614 	case BTRFS_IOC_QGROUP_CREATE:
5615 		return btrfs_ioctl_qgroup_create(file, argp);
5616 	case BTRFS_IOC_QGROUP_LIMIT:
5617 		return btrfs_ioctl_qgroup_limit(file, argp);
5618 	case BTRFS_IOC_QUOTA_RESCAN:
5619 		return btrfs_ioctl_quota_rescan(file, argp);
5620 	case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5621 		return btrfs_ioctl_quota_rescan_status(file, argp);
5622 	case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5623 		return btrfs_ioctl_quota_rescan_wait(file, argp);
5624 	case BTRFS_IOC_DEV_REPLACE:
5625 		return btrfs_ioctl_dev_replace(root, argp);
5626 	case BTRFS_IOC_GET_FSLABEL:
5627 		return btrfs_ioctl_get_fslabel(file, argp);
5628 	case BTRFS_IOC_SET_FSLABEL:
5629 		return btrfs_ioctl_set_fslabel(file, argp);
5630 	case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5631 		return btrfs_ioctl_get_supported_features(argp);
5632 	case BTRFS_IOC_GET_FEATURES:
5633 		return btrfs_ioctl_get_features(file, argp);
5634 	case BTRFS_IOC_SET_FEATURES:
5635 		return btrfs_ioctl_set_features(file, argp);
5636 	}
5637 
5638 	return -ENOTTY;
5639 }
5640 
5641 #ifdef CONFIG_COMPAT
5642 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5643 {
5644 	switch (cmd) {
5645 	case FS_IOC32_GETFLAGS:
5646 		cmd = FS_IOC_GETFLAGS;
5647 		break;
5648 	case FS_IOC32_SETFLAGS:
5649 		cmd = FS_IOC_SETFLAGS;
5650 		break;
5651 	case FS_IOC32_GETVERSION:
5652 		cmd = FS_IOC_GETVERSION;
5653 		break;
5654 	default:
5655 		return -ENOIOCTLCMD;
5656 	}
5657 
5658 	return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5659 }
5660 #endif
5661