xref: /openbmc/linux/fs/btrfs/ioctl.c (revision 93f5715e)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2007 Oracle.  All rights reserved.
4  */
5 
6 #include <linux/kernel.h>
7 #include <linux/bio.h>
8 #include <linux/file.h>
9 #include <linux/fs.h>
10 #include <linux/fsnotify.h>
11 #include <linux/pagemap.h>
12 #include <linux/highmem.h>
13 #include <linux/time.h>
14 #include <linux/string.h>
15 #include <linux/backing-dev.h>
16 #include <linux/mount.h>
17 #include <linux/namei.h>
18 #include <linux/writeback.h>
19 #include <linux/compat.h>
20 #include <linux/security.h>
21 #include <linux/xattr.h>
22 #include <linux/mm.h>
23 #include <linux/slab.h>
24 #include <linux/blkdev.h>
25 #include <linux/uuid.h>
26 #include <linux/btrfs.h>
27 #include <linux/uaccess.h>
28 #include <linux/iversion.h>
29 #include "ctree.h"
30 #include "disk-io.h"
31 #include "transaction.h"
32 #include "btrfs_inode.h"
33 #include "print-tree.h"
34 #include "volumes.h"
35 #include "locking.h"
36 #include "inode-map.h"
37 #include "backref.h"
38 #include "rcu-string.h"
39 #include "send.h"
40 #include "dev-replace.h"
41 #include "props.h"
42 #include "sysfs.h"
43 #include "qgroup.h"
44 #include "tree-log.h"
45 #include "compression.h"
46 
47 #ifdef CONFIG_64BIT
48 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
49  * structures are incorrect, as the timespec structure from userspace
50  * is 4 bytes too small. We define these alternatives here to teach
51  * the kernel about the 32-bit struct packing.
52  */
53 struct btrfs_ioctl_timespec_32 {
54 	__u64 sec;
55 	__u32 nsec;
56 } __attribute__ ((__packed__));
57 
58 struct btrfs_ioctl_received_subvol_args_32 {
59 	char	uuid[BTRFS_UUID_SIZE];	/* in */
60 	__u64	stransid;		/* in */
61 	__u64	rtransid;		/* out */
62 	struct btrfs_ioctl_timespec_32 stime; /* in */
63 	struct btrfs_ioctl_timespec_32 rtime; /* out */
64 	__u64	flags;			/* in */
65 	__u64	reserved[16];		/* in */
66 } __attribute__ ((__packed__));
67 
68 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
69 				struct btrfs_ioctl_received_subvol_args_32)
70 #endif
71 
72 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
73 struct btrfs_ioctl_send_args_32 {
74 	__s64 send_fd;			/* in */
75 	__u64 clone_sources_count;	/* in */
76 	compat_uptr_t clone_sources;	/* in */
77 	__u64 parent_root;		/* in */
78 	__u64 flags;			/* in */
79 	__u64 reserved[4];		/* in */
80 } __attribute__ ((__packed__));
81 
82 #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \
83 			       struct btrfs_ioctl_send_args_32)
84 #endif
85 
86 static int btrfs_clone(struct inode *src, struct inode *inode,
87 		       u64 off, u64 olen, u64 olen_aligned, u64 destoff,
88 		       int no_time_update);
89 
90 /* Mask out flags that are inappropriate for the given type of inode. */
91 static unsigned int btrfs_mask_fsflags_for_type(struct inode *inode,
92 		unsigned int flags)
93 {
94 	if (S_ISDIR(inode->i_mode))
95 		return flags;
96 	else if (S_ISREG(inode->i_mode))
97 		return flags & ~FS_DIRSYNC_FL;
98 	else
99 		return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
100 }
101 
102 /*
103  * Export internal inode flags to the format expected by the FS_IOC_GETFLAGS
104  * ioctl.
105  */
106 static unsigned int btrfs_inode_flags_to_fsflags(unsigned int flags)
107 {
108 	unsigned int iflags = 0;
109 
110 	if (flags & BTRFS_INODE_SYNC)
111 		iflags |= FS_SYNC_FL;
112 	if (flags & BTRFS_INODE_IMMUTABLE)
113 		iflags |= FS_IMMUTABLE_FL;
114 	if (flags & BTRFS_INODE_APPEND)
115 		iflags |= FS_APPEND_FL;
116 	if (flags & BTRFS_INODE_NODUMP)
117 		iflags |= FS_NODUMP_FL;
118 	if (flags & BTRFS_INODE_NOATIME)
119 		iflags |= FS_NOATIME_FL;
120 	if (flags & BTRFS_INODE_DIRSYNC)
121 		iflags |= FS_DIRSYNC_FL;
122 	if (flags & BTRFS_INODE_NODATACOW)
123 		iflags |= FS_NOCOW_FL;
124 
125 	if (flags & BTRFS_INODE_NOCOMPRESS)
126 		iflags |= FS_NOCOMP_FL;
127 	else if (flags & BTRFS_INODE_COMPRESS)
128 		iflags |= FS_COMPR_FL;
129 
130 	return iflags;
131 }
132 
133 /*
134  * Update inode->i_flags based on the btrfs internal flags.
135  */
136 void btrfs_sync_inode_flags_to_i_flags(struct inode *inode)
137 {
138 	struct btrfs_inode *binode = BTRFS_I(inode);
139 	unsigned int new_fl = 0;
140 
141 	if (binode->flags & BTRFS_INODE_SYNC)
142 		new_fl |= S_SYNC;
143 	if (binode->flags & BTRFS_INODE_IMMUTABLE)
144 		new_fl |= S_IMMUTABLE;
145 	if (binode->flags & BTRFS_INODE_APPEND)
146 		new_fl |= S_APPEND;
147 	if (binode->flags & BTRFS_INODE_NOATIME)
148 		new_fl |= S_NOATIME;
149 	if (binode->flags & BTRFS_INODE_DIRSYNC)
150 		new_fl |= S_DIRSYNC;
151 
152 	set_mask_bits(&inode->i_flags,
153 		      S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
154 		      new_fl);
155 }
156 
157 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
158 {
159 	struct btrfs_inode *binode = BTRFS_I(file_inode(file));
160 	unsigned int flags = btrfs_inode_flags_to_fsflags(binode->flags);
161 
162 	if (copy_to_user(arg, &flags, sizeof(flags)))
163 		return -EFAULT;
164 	return 0;
165 }
166 
167 /* Check if @flags are a supported and valid set of FS_*_FL flags */
168 static int check_fsflags(unsigned int flags)
169 {
170 	if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
171 		      FS_NOATIME_FL | FS_NODUMP_FL | \
172 		      FS_SYNC_FL | FS_DIRSYNC_FL | \
173 		      FS_NOCOMP_FL | FS_COMPR_FL |
174 		      FS_NOCOW_FL))
175 		return -EOPNOTSUPP;
176 
177 	if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
178 		return -EINVAL;
179 
180 	return 0;
181 }
182 
183 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
184 {
185 	struct inode *inode = file_inode(file);
186 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
187 	struct btrfs_inode *binode = BTRFS_I(inode);
188 	struct btrfs_root *root = binode->root;
189 	struct btrfs_trans_handle *trans;
190 	unsigned int fsflags, old_fsflags;
191 	int ret;
192 	u64 old_flags;
193 	unsigned int old_i_flags;
194 	umode_t mode;
195 
196 	if (!inode_owner_or_capable(inode))
197 		return -EPERM;
198 
199 	if (btrfs_root_readonly(root))
200 		return -EROFS;
201 
202 	if (copy_from_user(&fsflags, arg, sizeof(fsflags)))
203 		return -EFAULT;
204 
205 	ret = check_fsflags(fsflags);
206 	if (ret)
207 		return ret;
208 
209 	ret = mnt_want_write_file(file);
210 	if (ret)
211 		return ret;
212 
213 	inode_lock(inode);
214 
215 	old_flags = binode->flags;
216 	old_i_flags = inode->i_flags;
217 	mode = inode->i_mode;
218 
219 	fsflags = btrfs_mask_fsflags_for_type(inode, fsflags);
220 	old_fsflags = btrfs_inode_flags_to_fsflags(binode->flags);
221 	if ((fsflags ^ old_fsflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
222 		if (!capable(CAP_LINUX_IMMUTABLE)) {
223 			ret = -EPERM;
224 			goto out_unlock;
225 		}
226 	}
227 
228 	if (fsflags & FS_SYNC_FL)
229 		binode->flags |= BTRFS_INODE_SYNC;
230 	else
231 		binode->flags &= ~BTRFS_INODE_SYNC;
232 	if (fsflags & FS_IMMUTABLE_FL)
233 		binode->flags |= BTRFS_INODE_IMMUTABLE;
234 	else
235 		binode->flags &= ~BTRFS_INODE_IMMUTABLE;
236 	if (fsflags & FS_APPEND_FL)
237 		binode->flags |= BTRFS_INODE_APPEND;
238 	else
239 		binode->flags &= ~BTRFS_INODE_APPEND;
240 	if (fsflags & FS_NODUMP_FL)
241 		binode->flags |= BTRFS_INODE_NODUMP;
242 	else
243 		binode->flags &= ~BTRFS_INODE_NODUMP;
244 	if (fsflags & FS_NOATIME_FL)
245 		binode->flags |= BTRFS_INODE_NOATIME;
246 	else
247 		binode->flags &= ~BTRFS_INODE_NOATIME;
248 	if (fsflags & FS_DIRSYNC_FL)
249 		binode->flags |= BTRFS_INODE_DIRSYNC;
250 	else
251 		binode->flags &= ~BTRFS_INODE_DIRSYNC;
252 	if (fsflags & FS_NOCOW_FL) {
253 		if (S_ISREG(mode)) {
254 			/*
255 			 * It's safe to turn csums off here, no extents exist.
256 			 * Otherwise we want the flag to reflect the real COW
257 			 * status of the file and will not set it.
258 			 */
259 			if (inode->i_size == 0)
260 				binode->flags |= BTRFS_INODE_NODATACOW
261 					      | BTRFS_INODE_NODATASUM;
262 		} else {
263 			binode->flags |= BTRFS_INODE_NODATACOW;
264 		}
265 	} else {
266 		/*
267 		 * Revert back under same assumptions as above
268 		 */
269 		if (S_ISREG(mode)) {
270 			if (inode->i_size == 0)
271 				binode->flags &= ~(BTRFS_INODE_NODATACOW
272 				             | BTRFS_INODE_NODATASUM);
273 		} else {
274 			binode->flags &= ~BTRFS_INODE_NODATACOW;
275 		}
276 	}
277 
278 	/*
279 	 * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
280 	 * flag may be changed automatically if compression code won't make
281 	 * things smaller.
282 	 */
283 	if (fsflags & FS_NOCOMP_FL) {
284 		binode->flags &= ~BTRFS_INODE_COMPRESS;
285 		binode->flags |= BTRFS_INODE_NOCOMPRESS;
286 
287 		ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
288 		if (ret && ret != -ENODATA)
289 			goto out_drop;
290 	} else if (fsflags & FS_COMPR_FL) {
291 		const char *comp;
292 
293 		binode->flags |= BTRFS_INODE_COMPRESS;
294 		binode->flags &= ~BTRFS_INODE_NOCOMPRESS;
295 
296 		comp = btrfs_compress_type2str(fs_info->compress_type);
297 		if (!comp || comp[0] == 0)
298 			comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB);
299 
300 		ret = btrfs_set_prop(inode, "btrfs.compression",
301 				     comp, strlen(comp), 0);
302 		if (ret)
303 			goto out_drop;
304 
305 	} else {
306 		ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
307 		if (ret && ret != -ENODATA)
308 			goto out_drop;
309 		binode->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
310 	}
311 
312 	trans = btrfs_start_transaction(root, 1);
313 	if (IS_ERR(trans)) {
314 		ret = PTR_ERR(trans);
315 		goto out_drop;
316 	}
317 
318 	btrfs_sync_inode_flags_to_i_flags(inode);
319 	inode_inc_iversion(inode);
320 	inode->i_ctime = current_time(inode);
321 	ret = btrfs_update_inode(trans, root, inode);
322 
323 	btrfs_end_transaction(trans);
324  out_drop:
325 	if (ret) {
326 		binode->flags = old_flags;
327 		inode->i_flags = old_i_flags;
328 	}
329 
330  out_unlock:
331 	inode_unlock(inode);
332 	mnt_drop_write_file(file);
333 	return ret;
334 }
335 
336 /*
337  * Translate btrfs internal inode flags to xflags as expected by the
338  * FS_IOC_FSGETXATT ioctl. Filter only the supported ones, unknown flags are
339  * silently dropped.
340  */
341 static unsigned int btrfs_inode_flags_to_xflags(unsigned int flags)
342 {
343 	unsigned int xflags = 0;
344 
345 	if (flags & BTRFS_INODE_APPEND)
346 		xflags |= FS_XFLAG_APPEND;
347 	if (flags & BTRFS_INODE_IMMUTABLE)
348 		xflags |= FS_XFLAG_IMMUTABLE;
349 	if (flags & BTRFS_INODE_NOATIME)
350 		xflags |= FS_XFLAG_NOATIME;
351 	if (flags & BTRFS_INODE_NODUMP)
352 		xflags |= FS_XFLAG_NODUMP;
353 	if (flags & BTRFS_INODE_SYNC)
354 		xflags |= FS_XFLAG_SYNC;
355 
356 	return xflags;
357 }
358 
359 /* Check if @flags are a supported and valid set of FS_XFLAGS_* flags */
360 static int check_xflags(unsigned int flags)
361 {
362 	if (flags & ~(FS_XFLAG_APPEND | FS_XFLAG_IMMUTABLE | FS_XFLAG_NOATIME |
363 		      FS_XFLAG_NODUMP | FS_XFLAG_SYNC))
364 		return -EOPNOTSUPP;
365 	return 0;
366 }
367 
368 /*
369  * Set the xflags from the internal inode flags. The remaining items of fsxattr
370  * are zeroed.
371  */
372 static int btrfs_ioctl_fsgetxattr(struct file *file, void __user *arg)
373 {
374 	struct btrfs_inode *binode = BTRFS_I(file_inode(file));
375 	struct fsxattr fa;
376 
377 	memset(&fa, 0, sizeof(fa));
378 	fa.fsx_xflags = btrfs_inode_flags_to_xflags(binode->flags);
379 
380 	if (copy_to_user(arg, &fa, sizeof(fa)))
381 		return -EFAULT;
382 
383 	return 0;
384 }
385 
386 static int btrfs_ioctl_fssetxattr(struct file *file, void __user *arg)
387 {
388 	struct inode *inode = file_inode(file);
389 	struct btrfs_inode *binode = BTRFS_I(inode);
390 	struct btrfs_root *root = binode->root;
391 	struct btrfs_trans_handle *trans;
392 	struct fsxattr fa;
393 	unsigned old_flags;
394 	unsigned old_i_flags;
395 	int ret = 0;
396 
397 	if (!inode_owner_or_capable(inode))
398 		return -EPERM;
399 
400 	if (btrfs_root_readonly(root))
401 		return -EROFS;
402 
403 	memset(&fa, 0, sizeof(fa));
404 	if (copy_from_user(&fa, arg, sizeof(fa)))
405 		return -EFAULT;
406 
407 	ret = check_xflags(fa.fsx_xflags);
408 	if (ret)
409 		return ret;
410 
411 	if (fa.fsx_extsize != 0 || fa.fsx_projid != 0 || fa.fsx_cowextsize != 0)
412 		return -EOPNOTSUPP;
413 
414 	ret = mnt_want_write_file(file);
415 	if (ret)
416 		return ret;
417 
418 	inode_lock(inode);
419 
420 	old_flags = binode->flags;
421 	old_i_flags = inode->i_flags;
422 
423 	/* We need the capabilities to change append-only or immutable inode */
424 	if (((old_flags & (BTRFS_INODE_APPEND | BTRFS_INODE_IMMUTABLE)) ||
425 	     (fa.fsx_xflags & (FS_XFLAG_APPEND | FS_XFLAG_IMMUTABLE))) &&
426 	    !capable(CAP_LINUX_IMMUTABLE)) {
427 		ret = -EPERM;
428 		goto out_unlock;
429 	}
430 
431 	if (fa.fsx_xflags & FS_XFLAG_SYNC)
432 		binode->flags |= BTRFS_INODE_SYNC;
433 	else
434 		binode->flags &= ~BTRFS_INODE_SYNC;
435 	if (fa.fsx_xflags & FS_XFLAG_IMMUTABLE)
436 		binode->flags |= BTRFS_INODE_IMMUTABLE;
437 	else
438 		binode->flags &= ~BTRFS_INODE_IMMUTABLE;
439 	if (fa.fsx_xflags & FS_XFLAG_APPEND)
440 		binode->flags |= BTRFS_INODE_APPEND;
441 	else
442 		binode->flags &= ~BTRFS_INODE_APPEND;
443 	if (fa.fsx_xflags & FS_XFLAG_NODUMP)
444 		binode->flags |= BTRFS_INODE_NODUMP;
445 	else
446 		binode->flags &= ~BTRFS_INODE_NODUMP;
447 	if (fa.fsx_xflags & FS_XFLAG_NOATIME)
448 		binode->flags |= BTRFS_INODE_NOATIME;
449 	else
450 		binode->flags &= ~BTRFS_INODE_NOATIME;
451 
452 	/* 1 item for the inode */
453 	trans = btrfs_start_transaction(root, 1);
454 	if (IS_ERR(trans)) {
455 		ret = PTR_ERR(trans);
456 		goto out_unlock;
457 	}
458 
459 	btrfs_sync_inode_flags_to_i_flags(inode);
460 	inode_inc_iversion(inode);
461 	inode->i_ctime = current_time(inode);
462 	ret = btrfs_update_inode(trans, root, inode);
463 
464 	btrfs_end_transaction(trans);
465 
466 out_unlock:
467 	if (ret) {
468 		binode->flags = old_flags;
469 		inode->i_flags = old_i_flags;
470 	}
471 
472 	inode_unlock(inode);
473 	mnt_drop_write_file(file);
474 
475 	return ret;
476 }
477 
478 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
479 {
480 	struct inode *inode = file_inode(file);
481 
482 	return put_user(inode->i_generation, arg);
483 }
484 
485 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
486 {
487 	struct inode *inode = file_inode(file);
488 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
489 	struct btrfs_device *device;
490 	struct request_queue *q;
491 	struct fstrim_range range;
492 	u64 minlen = ULLONG_MAX;
493 	u64 num_devices = 0;
494 	u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
495 	int ret;
496 
497 	if (!capable(CAP_SYS_ADMIN))
498 		return -EPERM;
499 
500 	rcu_read_lock();
501 	list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
502 				dev_list) {
503 		if (!device->bdev)
504 			continue;
505 		q = bdev_get_queue(device->bdev);
506 		if (blk_queue_discard(q)) {
507 			num_devices++;
508 			minlen = min_t(u64, q->limits.discard_granularity,
509 				     minlen);
510 		}
511 	}
512 	rcu_read_unlock();
513 
514 	if (!num_devices)
515 		return -EOPNOTSUPP;
516 	if (copy_from_user(&range, arg, sizeof(range)))
517 		return -EFAULT;
518 	if (range.start > total_bytes ||
519 	    range.len < fs_info->sb->s_blocksize)
520 		return -EINVAL;
521 
522 	range.len = min(range.len, total_bytes - range.start);
523 	range.minlen = max(range.minlen, minlen);
524 	ret = btrfs_trim_fs(fs_info, &range);
525 	if (ret < 0)
526 		return ret;
527 
528 	if (copy_to_user(arg, &range, sizeof(range)))
529 		return -EFAULT;
530 
531 	return 0;
532 }
533 
534 int btrfs_is_empty_uuid(u8 *uuid)
535 {
536 	int i;
537 
538 	for (i = 0; i < BTRFS_UUID_SIZE; i++) {
539 		if (uuid[i])
540 			return 0;
541 	}
542 	return 1;
543 }
544 
545 static noinline int create_subvol(struct inode *dir,
546 				  struct dentry *dentry,
547 				  const char *name, int namelen,
548 				  u64 *async_transid,
549 				  struct btrfs_qgroup_inherit *inherit)
550 {
551 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
552 	struct btrfs_trans_handle *trans;
553 	struct btrfs_key key;
554 	struct btrfs_root_item *root_item;
555 	struct btrfs_inode_item *inode_item;
556 	struct extent_buffer *leaf;
557 	struct btrfs_root *root = BTRFS_I(dir)->root;
558 	struct btrfs_root *new_root;
559 	struct btrfs_block_rsv block_rsv;
560 	struct timespec64 cur_time = current_time(dir);
561 	struct inode *inode;
562 	int ret;
563 	int err;
564 	u64 objectid;
565 	u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
566 	u64 index = 0;
567 	uuid_le new_uuid;
568 
569 	root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
570 	if (!root_item)
571 		return -ENOMEM;
572 
573 	ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid);
574 	if (ret)
575 		goto fail_free;
576 
577 	/*
578 	 * Don't create subvolume whose level is not zero. Or qgroup will be
579 	 * screwed up since it assumes subvolume qgroup's level to be 0.
580 	 */
581 	if (btrfs_qgroup_level(objectid)) {
582 		ret = -ENOSPC;
583 		goto fail_free;
584 	}
585 
586 	btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
587 	/*
588 	 * The same as the snapshot creation, please see the comment
589 	 * of create_snapshot().
590 	 */
591 	ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 8, false);
592 	if (ret)
593 		goto fail_free;
594 
595 	trans = btrfs_start_transaction(root, 0);
596 	if (IS_ERR(trans)) {
597 		ret = PTR_ERR(trans);
598 		btrfs_subvolume_release_metadata(fs_info, &block_rsv);
599 		goto fail_free;
600 	}
601 	trans->block_rsv = &block_rsv;
602 	trans->bytes_reserved = block_rsv.size;
603 
604 	ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
605 	if (ret)
606 		goto fail;
607 
608 	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
609 	if (IS_ERR(leaf)) {
610 		ret = PTR_ERR(leaf);
611 		goto fail;
612 	}
613 
614 	btrfs_mark_buffer_dirty(leaf);
615 
616 	inode_item = &root_item->inode;
617 	btrfs_set_stack_inode_generation(inode_item, 1);
618 	btrfs_set_stack_inode_size(inode_item, 3);
619 	btrfs_set_stack_inode_nlink(inode_item, 1);
620 	btrfs_set_stack_inode_nbytes(inode_item,
621 				     fs_info->nodesize);
622 	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
623 
624 	btrfs_set_root_flags(root_item, 0);
625 	btrfs_set_root_limit(root_item, 0);
626 	btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
627 
628 	btrfs_set_root_bytenr(root_item, leaf->start);
629 	btrfs_set_root_generation(root_item, trans->transid);
630 	btrfs_set_root_level(root_item, 0);
631 	btrfs_set_root_refs(root_item, 1);
632 	btrfs_set_root_used(root_item, leaf->len);
633 	btrfs_set_root_last_snapshot(root_item, 0);
634 
635 	btrfs_set_root_generation_v2(root_item,
636 			btrfs_root_generation(root_item));
637 	uuid_le_gen(&new_uuid);
638 	memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
639 	btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
640 	btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
641 	root_item->ctime = root_item->otime;
642 	btrfs_set_root_ctransid(root_item, trans->transid);
643 	btrfs_set_root_otransid(root_item, trans->transid);
644 
645 	btrfs_tree_unlock(leaf);
646 	free_extent_buffer(leaf);
647 	leaf = NULL;
648 
649 	btrfs_set_root_dirid(root_item, new_dirid);
650 
651 	key.objectid = objectid;
652 	key.offset = 0;
653 	key.type = BTRFS_ROOT_ITEM_KEY;
654 	ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
655 				root_item);
656 	if (ret)
657 		goto fail;
658 
659 	key.offset = (u64)-1;
660 	new_root = btrfs_read_fs_root_no_name(fs_info, &key);
661 	if (IS_ERR(new_root)) {
662 		ret = PTR_ERR(new_root);
663 		btrfs_abort_transaction(trans, ret);
664 		goto fail;
665 	}
666 
667 	btrfs_record_root_in_trans(trans, new_root);
668 
669 	ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
670 	if (ret) {
671 		/* We potentially lose an unused inode item here */
672 		btrfs_abort_transaction(trans, ret);
673 		goto fail;
674 	}
675 
676 	mutex_lock(&new_root->objectid_mutex);
677 	new_root->highest_objectid = new_dirid;
678 	mutex_unlock(&new_root->objectid_mutex);
679 
680 	/*
681 	 * insert the directory item
682 	 */
683 	ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
684 	if (ret) {
685 		btrfs_abort_transaction(trans, ret);
686 		goto fail;
687 	}
688 
689 	ret = btrfs_insert_dir_item(trans, root,
690 				    name, namelen, BTRFS_I(dir), &key,
691 				    BTRFS_FT_DIR, index);
692 	if (ret) {
693 		btrfs_abort_transaction(trans, ret);
694 		goto fail;
695 	}
696 
697 	btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
698 	ret = btrfs_update_inode(trans, root, dir);
699 	BUG_ON(ret);
700 
701 	ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid,
702 				 btrfs_ino(BTRFS_I(dir)), index, name, namelen);
703 	BUG_ON(ret);
704 
705 	ret = btrfs_uuid_tree_add(trans, root_item->uuid,
706 				  BTRFS_UUID_KEY_SUBVOL, objectid);
707 	if (ret)
708 		btrfs_abort_transaction(trans, ret);
709 
710 fail:
711 	kfree(root_item);
712 	trans->block_rsv = NULL;
713 	trans->bytes_reserved = 0;
714 	btrfs_subvolume_release_metadata(fs_info, &block_rsv);
715 
716 	if (async_transid) {
717 		*async_transid = trans->transid;
718 		err = btrfs_commit_transaction_async(trans, 1);
719 		if (err)
720 			err = btrfs_commit_transaction(trans);
721 	} else {
722 		err = btrfs_commit_transaction(trans);
723 	}
724 	if (err && !ret)
725 		ret = err;
726 
727 	if (!ret) {
728 		inode = btrfs_lookup_dentry(dir, dentry);
729 		if (IS_ERR(inode))
730 			return PTR_ERR(inode);
731 		d_instantiate(dentry, inode);
732 	}
733 	return ret;
734 
735 fail_free:
736 	kfree(root_item);
737 	return ret;
738 }
739 
740 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
741 			   struct dentry *dentry,
742 			   u64 *async_transid, bool readonly,
743 			   struct btrfs_qgroup_inherit *inherit)
744 {
745 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
746 	struct inode *inode;
747 	struct btrfs_pending_snapshot *pending_snapshot;
748 	struct btrfs_trans_handle *trans;
749 	int ret;
750 
751 	if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
752 		return -EINVAL;
753 
754 	pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
755 	if (!pending_snapshot)
756 		return -ENOMEM;
757 
758 	pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
759 			GFP_KERNEL);
760 	pending_snapshot->path = btrfs_alloc_path();
761 	if (!pending_snapshot->root_item || !pending_snapshot->path) {
762 		ret = -ENOMEM;
763 		goto free_pending;
764 	}
765 
766 	atomic_inc(&root->will_be_snapshotted);
767 	smp_mb__after_atomic();
768 	/* wait for no snapshot writes */
769 	wait_event(root->subv_writers->wait,
770 		   percpu_counter_sum(&root->subv_writers->counter) == 0);
771 
772 	ret = btrfs_start_delalloc_inodes(root);
773 	if (ret)
774 		goto dec_and_free;
775 
776 	btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
777 
778 	btrfs_init_block_rsv(&pending_snapshot->block_rsv,
779 			     BTRFS_BLOCK_RSV_TEMP);
780 	/*
781 	 * 1 - parent dir inode
782 	 * 2 - dir entries
783 	 * 1 - root item
784 	 * 2 - root ref/backref
785 	 * 1 - root of snapshot
786 	 * 1 - UUID item
787 	 */
788 	ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
789 					&pending_snapshot->block_rsv, 8,
790 					false);
791 	if (ret)
792 		goto dec_and_free;
793 
794 	pending_snapshot->dentry = dentry;
795 	pending_snapshot->root = root;
796 	pending_snapshot->readonly = readonly;
797 	pending_snapshot->dir = dir;
798 	pending_snapshot->inherit = inherit;
799 
800 	trans = btrfs_start_transaction(root, 0);
801 	if (IS_ERR(trans)) {
802 		ret = PTR_ERR(trans);
803 		goto fail;
804 	}
805 
806 	spin_lock(&fs_info->trans_lock);
807 	list_add(&pending_snapshot->list,
808 		 &trans->transaction->pending_snapshots);
809 	spin_unlock(&fs_info->trans_lock);
810 	if (async_transid) {
811 		*async_transid = trans->transid;
812 		ret = btrfs_commit_transaction_async(trans, 1);
813 		if (ret)
814 			ret = btrfs_commit_transaction(trans);
815 	} else {
816 		ret = btrfs_commit_transaction(trans);
817 	}
818 	if (ret)
819 		goto fail;
820 
821 	ret = pending_snapshot->error;
822 	if (ret)
823 		goto fail;
824 
825 	ret = btrfs_orphan_cleanup(pending_snapshot->snap);
826 	if (ret)
827 		goto fail;
828 
829 	inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
830 	if (IS_ERR(inode)) {
831 		ret = PTR_ERR(inode);
832 		goto fail;
833 	}
834 
835 	d_instantiate(dentry, inode);
836 	ret = 0;
837 fail:
838 	btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
839 dec_and_free:
840 	if (atomic_dec_and_test(&root->will_be_snapshotted))
841 		wake_up_var(&root->will_be_snapshotted);
842 free_pending:
843 	kfree(pending_snapshot->root_item);
844 	btrfs_free_path(pending_snapshot->path);
845 	kfree(pending_snapshot);
846 
847 	return ret;
848 }
849 
850 /*  copy of may_delete in fs/namei.c()
851  *	Check whether we can remove a link victim from directory dir, check
852  *  whether the type of victim is right.
853  *  1. We can't do it if dir is read-only (done in permission())
854  *  2. We should have write and exec permissions on dir
855  *  3. We can't remove anything from append-only dir
856  *  4. We can't do anything with immutable dir (done in permission())
857  *  5. If the sticky bit on dir is set we should either
858  *	a. be owner of dir, or
859  *	b. be owner of victim, or
860  *	c. have CAP_FOWNER capability
861  *  6. If the victim is append-only or immutable we can't do anything with
862  *     links pointing to it.
863  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
864  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
865  *  9. We can't remove a root or mountpoint.
866  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
867  *     nfs_async_unlink().
868  */
869 
870 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
871 {
872 	int error;
873 
874 	if (d_really_is_negative(victim))
875 		return -ENOENT;
876 
877 	BUG_ON(d_inode(victim->d_parent) != dir);
878 	audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
879 
880 	error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
881 	if (error)
882 		return error;
883 	if (IS_APPEND(dir))
884 		return -EPERM;
885 	if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
886 	    IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
887 		return -EPERM;
888 	if (isdir) {
889 		if (!d_is_dir(victim))
890 			return -ENOTDIR;
891 		if (IS_ROOT(victim))
892 			return -EBUSY;
893 	} else if (d_is_dir(victim))
894 		return -EISDIR;
895 	if (IS_DEADDIR(dir))
896 		return -ENOENT;
897 	if (victim->d_flags & DCACHE_NFSFS_RENAMED)
898 		return -EBUSY;
899 	return 0;
900 }
901 
902 /* copy of may_create in fs/namei.c() */
903 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
904 {
905 	if (d_really_is_positive(child))
906 		return -EEXIST;
907 	if (IS_DEADDIR(dir))
908 		return -ENOENT;
909 	return inode_permission(dir, MAY_WRITE | MAY_EXEC);
910 }
911 
912 /*
913  * Create a new subvolume below @parent.  This is largely modeled after
914  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
915  * inside this filesystem so it's quite a bit simpler.
916  */
917 static noinline int btrfs_mksubvol(const struct path *parent,
918 				   const char *name, int namelen,
919 				   struct btrfs_root *snap_src,
920 				   u64 *async_transid, bool readonly,
921 				   struct btrfs_qgroup_inherit *inherit)
922 {
923 	struct inode *dir = d_inode(parent->dentry);
924 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
925 	struct dentry *dentry;
926 	int error;
927 
928 	error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
929 	if (error == -EINTR)
930 		return error;
931 
932 	dentry = lookup_one_len(name, parent->dentry, namelen);
933 	error = PTR_ERR(dentry);
934 	if (IS_ERR(dentry))
935 		goto out_unlock;
936 
937 	error = btrfs_may_create(dir, dentry);
938 	if (error)
939 		goto out_dput;
940 
941 	/*
942 	 * even if this name doesn't exist, we may get hash collisions.
943 	 * check for them now when we can safely fail
944 	 */
945 	error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
946 					       dir->i_ino, name,
947 					       namelen);
948 	if (error)
949 		goto out_dput;
950 
951 	down_read(&fs_info->subvol_sem);
952 
953 	if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
954 		goto out_up_read;
955 
956 	if (snap_src) {
957 		error = create_snapshot(snap_src, dir, dentry,
958 					async_transid, readonly, inherit);
959 	} else {
960 		error = create_subvol(dir, dentry, name, namelen,
961 				      async_transid, inherit);
962 	}
963 	if (!error)
964 		fsnotify_mkdir(dir, dentry);
965 out_up_read:
966 	up_read(&fs_info->subvol_sem);
967 out_dput:
968 	dput(dentry);
969 out_unlock:
970 	inode_unlock(dir);
971 	return error;
972 }
973 
974 /*
975  * When we're defragging a range, we don't want to kick it off again
976  * if it is really just waiting for delalloc to send it down.
977  * If we find a nice big extent or delalloc range for the bytes in the
978  * file you want to defrag, we return 0 to let you know to skip this
979  * part of the file
980  */
981 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
982 {
983 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
984 	struct extent_map *em = NULL;
985 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
986 	u64 end;
987 
988 	read_lock(&em_tree->lock);
989 	em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
990 	read_unlock(&em_tree->lock);
991 
992 	if (em) {
993 		end = extent_map_end(em);
994 		free_extent_map(em);
995 		if (end - offset > thresh)
996 			return 0;
997 	}
998 	/* if we already have a nice delalloc here, just stop */
999 	thresh /= 2;
1000 	end = count_range_bits(io_tree, &offset, offset + thresh,
1001 			       thresh, EXTENT_DELALLOC, 1);
1002 	if (end >= thresh)
1003 		return 0;
1004 	return 1;
1005 }
1006 
1007 /*
1008  * helper function to walk through a file and find extents
1009  * newer than a specific transid, and smaller than thresh.
1010  *
1011  * This is used by the defragging code to find new and small
1012  * extents
1013  */
1014 static int find_new_extents(struct btrfs_root *root,
1015 			    struct inode *inode, u64 newer_than,
1016 			    u64 *off, u32 thresh)
1017 {
1018 	struct btrfs_path *path;
1019 	struct btrfs_key min_key;
1020 	struct extent_buffer *leaf;
1021 	struct btrfs_file_extent_item *extent;
1022 	int type;
1023 	int ret;
1024 	u64 ino = btrfs_ino(BTRFS_I(inode));
1025 
1026 	path = btrfs_alloc_path();
1027 	if (!path)
1028 		return -ENOMEM;
1029 
1030 	min_key.objectid = ino;
1031 	min_key.type = BTRFS_EXTENT_DATA_KEY;
1032 	min_key.offset = *off;
1033 
1034 	while (1) {
1035 		ret = btrfs_search_forward(root, &min_key, path, newer_than);
1036 		if (ret != 0)
1037 			goto none;
1038 process_slot:
1039 		if (min_key.objectid != ino)
1040 			goto none;
1041 		if (min_key.type != BTRFS_EXTENT_DATA_KEY)
1042 			goto none;
1043 
1044 		leaf = path->nodes[0];
1045 		extent = btrfs_item_ptr(leaf, path->slots[0],
1046 					struct btrfs_file_extent_item);
1047 
1048 		type = btrfs_file_extent_type(leaf, extent);
1049 		if (type == BTRFS_FILE_EXTENT_REG &&
1050 		    btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
1051 		    check_defrag_in_cache(inode, min_key.offset, thresh)) {
1052 			*off = min_key.offset;
1053 			btrfs_free_path(path);
1054 			return 0;
1055 		}
1056 
1057 		path->slots[0]++;
1058 		if (path->slots[0] < btrfs_header_nritems(leaf)) {
1059 			btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
1060 			goto process_slot;
1061 		}
1062 
1063 		if (min_key.offset == (u64)-1)
1064 			goto none;
1065 
1066 		min_key.offset++;
1067 		btrfs_release_path(path);
1068 	}
1069 none:
1070 	btrfs_free_path(path);
1071 	return -ENOENT;
1072 }
1073 
1074 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1075 {
1076 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1077 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1078 	struct extent_map *em;
1079 	u64 len = PAGE_SIZE;
1080 
1081 	/*
1082 	 * hopefully we have this extent in the tree already, try without
1083 	 * the full extent lock
1084 	 */
1085 	read_lock(&em_tree->lock);
1086 	em = lookup_extent_mapping(em_tree, start, len);
1087 	read_unlock(&em_tree->lock);
1088 
1089 	if (!em) {
1090 		struct extent_state *cached = NULL;
1091 		u64 end = start + len - 1;
1092 
1093 		/* get the big lock and read metadata off disk */
1094 		lock_extent_bits(io_tree, start, end, &cached);
1095 		em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
1096 		unlock_extent_cached(io_tree, start, end, &cached);
1097 
1098 		if (IS_ERR(em))
1099 			return NULL;
1100 	}
1101 
1102 	return em;
1103 }
1104 
1105 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1106 {
1107 	struct extent_map *next;
1108 	bool ret = true;
1109 
1110 	/* this is the last extent */
1111 	if (em->start + em->len >= i_size_read(inode))
1112 		return false;
1113 
1114 	next = defrag_lookup_extent(inode, em->start + em->len);
1115 	if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1116 		ret = false;
1117 	else if ((em->block_start + em->block_len == next->block_start) &&
1118 		 (em->block_len > SZ_128K && next->block_len > SZ_128K))
1119 		ret = false;
1120 
1121 	free_extent_map(next);
1122 	return ret;
1123 }
1124 
1125 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1126 			       u64 *last_len, u64 *skip, u64 *defrag_end,
1127 			       int compress)
1128 {
1129 	struct extent_map *em;
1130 	int ret = 1;
1131 	bool next_mergeable = true;
1132 	bool prev_mergeable = true;
1133 
1134 	/*
1135 	 * make sure that once we start defragging an extent, we keep on
1136 	 * defragging it
1137 	 */
1138 	if (start < *defrag_end)
1139 		return 1;
1140 
1141 	*skip = 0;
1142 
1143 	em = defrag_lookup_extent(inode, start);
1144 	if (!em)
1145 		return 0;
1146 
1147 	/* this will cover holes, and inline extents */
1148 	if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1149 		ret = 0;
1150 		goto out;
1151 	}
1152 
1153 	if (!*defrag_end)
1154 		prev_mergeable = false;
1155 
1156 	next_mergeable = defrag_check_next_extent(inode, em);
1157 	/*
1158 	 * we hit a real extent, if it is big or the next extent is not a
1159 	 * real extent, don't bother defragging it
1160 	 */
1161 	if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1162 	    (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1163 		ret = 0;
1164 out:
1165 	/*
1166 	 * last_len ends up being a counter of how many bytes we've defragged.
1167 	 * every time we choose not to defrag an extent, we reset *last_len
1168 	 * so that the next tiny extent will force a defrag.
1169 	 *
1170 	 * The end result of this is that tiny extents before a single big
1171 	 * extent will force at least part of that big extent to be defragged.
1172 	 */
1173 	if (ret) {
1174 		*defrag_end = extent_map_end(em);
1175 	} else {
1176 		*last_len = 0;
1177 		*skip = extent_map_end(em);
1178 		*defrag_end = 0;
1179 	}
1180 
1181 	free_extent_map(em);
1182 	return ret;
1183 }
1184 
1185 /*
1186  * it doesn't do much good to defrag one or two pages
1187  * at a time.  This pulls in a nice chunk of pages
1188  * to COW and defrag.
1189  *
1190  * It also makes sure the delalloc code has enough
1191  * dirty data to avoid making new small extents as part
1192  * of the defrag
1193  *
1194  * It's a good idea to start RA on this range
1195  * before calling this.
1196  */
1197 static int cluster_pages_for_defrag(struct inode *inode,
1198 				    struct page **pages,
1199 				    unsigned long start_index,
1200 				    unsigned long num_pages)
1201 {
1202 	unsigned long file_end;
1203 	u64 isize = i_size_read(inode);
1204 	u64 page_start;
1205 	u64 page_end;
1206 	u64 page_cnt;
1207 	int ret;
1208 	int i;
1209 	int i_done;
1210 	struct btrfs_ordered_extent *ordered;
1211 	struct extent_state *cached_state = NULL;
1212 	struct extent_io_tree *tree;
1213 	struct extent_changeset *data_reserved = NULL;
1214 	gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1215 
1216 	file_end = (isize - 1) >> PAGE_SHIFT;
1217 	if (!isize || start_index > file_end)
1218 		return 0;
1219 
1220 	page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1221 
1222 	ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
1223 			start_index << PAGE_SHIFT,
1224 			page_cnt << PAGE_SHIFT);
1225 	if (ret)
1226 		return ret;
1227 	i_done = 0;
1228 	tree = &BTRFS_I(inode)->io_tree;
1229 
1230 	/* step one, lock all the pages */
1231 	for (i = 0; i < page_cnt; i++) {
1232 		struct page *page;
1233 again:
1234 		page = find_or_create_page(inode->i_mapping,
1235 					   start_index + i, mask);
1236 		if (!page)
1237 			break;
1238 
1239 		page_start = page_offset(page);
1240 		page_end = page_start + PAGE_SIZE - 1;
1241 		while (1) {
1242 			lock_extent_bits(tree, page_start, page_end,
1243 					 &cached_state);
1244 			ordered = btrfs_lookup_ordered_extent(inode,
1245 							      page_start);
1246 			unlock_extent_cached(tree, page_start, page_end,
1247 					     &cached_state);
1248 			if (!ordered)
1249 				break;
1250 
1251 			unlock_page(page);
1252 			btrfs_start_ordered_extent(inode, ordered, 1);
1253 			btrfs_put_ordered_extent(ordered);
1254 			lock_page(page);
1255 			/*
1256 			 * we unlocked the page above, so we need check if
1257 			 * it was released or not.
1258 			 */
1259 			if (page->mapping != inode->i_mapping) {
1260 				unlock_page(page);
1261 				put_page(page);
1262 				goto again;
1263 			}
1264 		}
1265 
1266 		if (!PageUptodate(page)) {
1267 			btrfs_readpage(NULL, page);
1268 			lock_page(page);
1269 			if (!PageUptodate(page)) {
1270 				unlock_page(page);
1271 				put_page(page);
1272 				ret = -EIO;
1273 				break;
1274 			}
1275 		}
1276 
1277 		if (page->mapping != inode->i_mapping) {
1278 			unlock_page(page);
1279 			put_page(page);
1280 			goto again;
1281 		}
1282 
1283 		pages[i] = page;
1284 		i_done++;
1285 	}
1286 	if (!i_done || ret)
1287 		goto out;
1288 
1289 	if (!(inode->i_sb->s_flags & SB_ACTIVE))
1290 		goto out;
1291 
1292 	/*
1293 	 * so now we have a nice long stream of locked
1294 	 * and up to date pages, lets wait on them
1295 	 */
1296 	for (i = 0; i < i_done; i++)
1297 		wait_on_page_writeback(pages[i]);
1298 
1299 	page_start = page_offset(pages[0]);
1300 	page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1301 
1302 	lock_extent_bits(&BTRFS_I(inode)->io_tree,
1303 			 page_start, page_end - 1, &cached_state);
1304 	clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1305 			  page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1306 			  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1307 			  &cached_state);
1308 
1309 	if (i_done != page_cnt) {
1310 		spin_lock(&BTRFS_I(inode)->lock);
1311 		BTRFS_I(inode)->outstanding_extents++;
1312 		spin_unlock(&BTRFS_I(inode)->lock);
1313 		btrfs_delalloc_release_space(inode, data_reserved,
1314 				start_index << PAGE_SHIFT,
1315 				(page_cnt - i_done) << PAGE_SHIFT, true);
1316 	}
1317 
1318 
1319 	set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1320 			  &cached_state);
1321 
1322 	unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1323 			     page_start, page_end - 1, &cached_state);
1324 
1325 	for (i = 0; i < i_done; i++) {
1326 		clear_page_dirty_for_io(pages[i]);
1327 		ClearPageChecked(pages[i]);
1328 		set_page_extent_mapped(pages[i]);
1329 		set_page_dirty(pages[i]);
1330 		unlock_page(pages[i]);
1331 		put_page(pages[i]);
1332 	}
1333 	btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT,
1334 				       false);
1335 	extent_changeset_free(data_reserved);
1336 	return i_done;
1337 out:
1338 	for (i = 0; i < i_done; i++) {
1339 		unlock_page(pages[i]);
1340 		put_page(pages[i]);
1341 	}
1342 	btrfs_delalloc_release_space(inode, data_reserved,
1343 			start_index << PAGE_SHIFT,
1344 			page_cnt << PAGE_SHIFT, true);
1345 	btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT,
1346 				       true);
1347 	extent_changeset_free(data_reserved);
1348 	return ret;
1349 
1350 }
1351 
1352 int btrfs_defrag_file(struct inode *inode, struct file *file,
1353 		      struct btrfs_ioctl_defrag_range_args *range,
1354 		      u64 newer_than, unsigned long max_to_defrag)
1355 {
1356 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1357 	struct btrfs_root *root = BTRFS_I(inode)->root;
1358 	struct file_ra_state *ra = NULL;
1359 	unsigned long last_index;
1360 	u64 isize = i_size_read(inode);
1361 	u64 last_len = 0;
1362 	u64 skip = 0;
1363 	u64 defrag_end = 0;
1364 	u64 newer_off = range->start;
1365 	unsigned long i;
1366 	unsigned long ra_index = 0;
1367 	int ret;
1368 	int defrag_count = 0;
1369 	int compress_type = BTRFS_COMPRESS_ZLIB;
1370 	u32 extent_thresh = range->extent_thresh;
1371 	unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1372 	unsigned long cluster = max_cluster;
1373 	u64 new_align = ~((u64)SZ_128K - 1);
1374 	struct page **pages = NULL;
1375 	bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1376 
1377 	if (isize == 0)
1378 		return 0;
1379 
1380 	if (range->start >= isize)
1381 		return -EINVAL;
1382 
1383 	if (do_compress) {
1384 		if (range->compress_type > BTRFS_COMPRESS_TYPES)
1385 			return -EINVAL;
1386 		if (range->compress_type)
1387 			compress_type = range->compress_type;
1388 	}
1389 
1390 	if (extent_thresh == 0)
1391 		extent_thresh = SZ_256K;
1392 
1393 	/*
1394 	 * If we were not given a file, allocate a readahead context. As
1395 	 * readahead is just an optimization, defrag will work without it so
1396 	 * we don't error out.
1397 	 */
1398 	if (!file) {
1399 		ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1400 		if (ra)
1401 			file_ra_state_init(ra, inode->i_mapping);
1402 	} else {
1403 		ra = &file->f_ra;
1404 	}
1405 
1406 	pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1407 	if (!pages) {
1408 		ret = -ENOMEM;
1409 		goto out_ra;
1410 	}
1411 
1412 	/* find the last page to defrag */
1413 	if (range->start + range->len > range->start) {
1414 		last_index = min_t(u64, isize - 1,
1415 			 range->start + range->len - 1) >> PAGE_SHIFT;
1416 	} else {
1417 		last_index = (isize - 1) >> PAGE_SHIFT;
1418 	}
1419 
1420 	if (newer_than) {
1421 		ret = find_new_extents(root, inode, newer_than,
1422 				       &newer_off, SZ_64K);
1423 		if (!ret) {
1424 			range->start = newer_off;
1425 			/*
1426 			 * we always align our defrag to help keep
1427 			 * the extents in the file evenly spaced
1428 			 */
1429 			i = (newer_off & new_align) >> PAGE_SHIFT;
1430 		} else
1431 			goto out_ra;
1432 	} else {
1433 		i = range->start >> PAGE_SHIFT;
1434 	}
1435 	if (!max_to_defrag)
1436 		max_to_defrag = last_index - i + 1;
1437 
1438 	/*
1439 	 * make writeback starts from i, so the defrag range can be
1440 	 * written sequentially.
1441 	 */
1442 	if (i < inode->i_mapping->writeback_index)
1443 		inode->i_mapping->writeback_index = i;
1444 
1445 	while (i <= last_index && defrag_count < max_to_defrag &&
1446 	       (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1447 		/*
1448 		 * make sure we stop running if someone unmounts
1449 		 * the FS
1450 		 */
1451 		if (!(inode->i_sb->s_flags & SB_ACTIVE))
1452 			break;
1453 
1454 		if (btrfs_defrag_cancelled(fs_info)) {
1455 			btrfs_debug(fs_info, "defrag_file cancelled");
1456 			ret = -EAGAIN;
1457 			break;
1458 		}
1459 
1460 		if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1461 					 extent_thresh, &last_len, &skip,
1462 					 &defrag_end, do_compress)){
1463 			unsigned long next;
1464 			/*
1465 			 * the should_defrag function tells us how much to skip
1466 			 * bump our counter by the suggested amount
1467 			 */
1468 			next = DIV_ROUND_UP(skip, PAGE_SIZE);
1469 			i = max(i + 1, next);
1470 			continue;
1471 		}
1472 
1473 		if (!newer_than) {
1474 			cluster = (PAGE_ALIGN(defrag_end) >>
1475 				   PAGE_SHIFT) - i;
1476 			cluster = min(cluster, max_cluster);
1477 		} else {
1478 			cluster = max_cluster;
1479 		}
1480 
1481 		if (i + cluster > ra_index) {
1482 			ra_index = max(i, ra_index);
1483 			if (ra)
1484 				page_cache_sync_readahead(inode->i_mapping, ra,
1485 						file, ra_index, cluster);
1486 			ra_index += cluster;
1487 		}
1488 
1489 		inode_lock(inode);
1490 		if (do_compress)
1491 			BTRFS_I(inode)->defrag_compress = compress_type;
1492 		ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1493 		if (ret < 0) {
1494 			inode_unlock(inode);
1495 			goto out_ra;
1496 		}
1497 
1498 		defrag_count += ret;
1499 		balance_dirty_pages_ratelimited(inode->i_mapping);
1500 		inode_unlock(inode);
1501 
1502 		if (newer_than) {
1503 			if (newer_off == (u64)-1)
1504 				break;
1505 
1506 			if (ret > 0)
1507 				i += ret;
1508 
1509 			newer_off = max(newer_off + 1,
1510 					(u64)i << PAGE_SHIFT);
1511 
1512 			ret = find_new_extents(root, inode, newer_than,
1513 					       &newer_off, SZ_64K);
1514 			if (!ret) {
1515 				range->start = newer_off;
1516 				i = (newer_off & new_align) >> PAGE_SHIFT;
1517 			} else {
1518 				break;
1519 			}
1520 		} else {
1521 			if (ret > 0) {
1522 				i += ret;
1523 				last_len += ret << PAGE_SHIFT;
1524 			} else {
1525 				i++;
1526 				last_len = 0;
1527 			}
1528 		}
1529 	}
1530 
1531 	if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1532 		filemap_flush(inode->i_mapping);
1533 		if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1534 			     &BTRFS_I(inode)->runtime_flags))
1535 			filemap_flush(inode->i_mapping);
1536 	}
1537 
1538 	if (range->compress_type == BTRFS_COMPRESS_LZO) {
1539 		btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1540 	} else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1541 		btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1542 	}
1543 
1544 	ret = defrag_count;
1545 
1546 out_ra:
1547 	if (do_compress) {
1548 		inode_lock(inode);
1549 		BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1550 		inode_unlock(inode);
1551 	}
1552 	if (!file)
1553 		kfree(ra);
1554 	kfree(pages);
1555 	return ret;
1556 }
1557 
1558 static noinline int btrfs_ioctl_resize(struct file *file,
1559 					void __user *arg)
1560 {
1561 	struct inode *inode = file_inode(file);
1562 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1563 	u64 new_size;
1564 	u64 old_size;
1565 	u64 devid = 1;
1566 	struct btrfs_root *root = BTRFS_I(inode)->root;
1567 	struct btrfs_ioctl_vol_args *vol_args;
1568 	struct btrfs_trans_handle *trans;
1569 	struct btrfs_device *device = NULL;
1570 	char *sizestr;
1571 	char *retptr;
1572 	char *devstr = NULL;
1573 	int ret = 0;
1574 	int mod = 0;
1575 
1576 	if (!capable(CAP_SYS_ADMIN))
1577 		return -EPERM;
1578 
1579 	ret = mnt_want_write_file(file);
1580 	if (ret)
1581 		return ret;
1582 
1583 	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
1584 		mnt_drop_write_file(file);
1585 		return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1586 	}
1587 
1588 	vol_args = memdup_user(arg, sizeof(*vol_args));
1589 	if (IS_ERR(vol_args)) {
1590 		ret = PTR_ERR(vol_args);
1591 		goto out;
1592 	}
1593 
1594 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1595 
1596 	sizestr = vol_args->name;
1597 	devstr = strchr(sizestr, ':');
1598 	if (devstr) {
1599 		sizestr = devstr + 1;
1600 		*devstr = '\0';
1601 		devstr = vol_args->name;
1602 		ret = kstrtoull(devstr, 10, &devid);
1603 		if (ret)
1604 			goto out_free;
1605 		if (!devid) {
1606 			ret = -EINVAL;
1607 			goto out_free;
1608 		}
1609 		btrfs_info(fs_info, "resizing devid %llu", devid);
1610 	}
1611 
1612 	device = btrfs_find_device(fs_info, devid, NULL, NULL);
1613 	if (!device) {
1614 		btrfs_info(fs_info, "resizer unable to find device %llu",
1615 			   devid);
1616 		ret = -ENODEV;
1617 		goto out_free;
1618 	}
1619 
1620 	if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1621 		btrfs_info(fs_info,
1622 			   "resizer unable to apply on readonly device %llu",
1623 		       devid);
1624 		ret = -EPERM;
1625 		goto out_free;
1626 	}
1627 
1628 	if (!strcmp(sizestr, "max"))
1629 		new_size = device->bdev->bd_inode->i_size;
1630 	else {
1631 		if (sizestr[0] == '-') {
1632 			mod = -1;
1633 			sizestr++;
1634 		} else if (sizestr[0] == '+') {
1635 			mod = 1;
1636 			sizestr++;
1637 		}
1638 		new_size = memparse(sizestr, &retptr);
1639 		if (*retptr != '\0' || new_size == 0) {
1640 			ret = -EINVAL;
1641 			goto out_free;
1642 		}
1643 	}
1644 
1645 	if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1646 		ret = -EPERM;
1647 		goto out_free;
1648 	}
1649 
1650 	old_size = btrfs_device_get_total_bytes(device);
1651 
1652 	if (mod < 0) {
1653 		if (new_size > old_size) {
1654 			ret = -EINVAL;
1655 			goto out_free;
1656 		}
1657 		new_size = old_size - new_size;
1658 	} else if (mod > 0) {
1659 		if (new_size > ULLONG_MAX - old_size) {
1660 			ret = -ERANGE;
1661 			goto out_free;
1662 		}
1663 		new_size = old_size + new_size;
1664 	}
1665 
1666 	if (new_size < SZ_256M) {
1667 		ret = -EINVAL;
1668 		goto out_free;
1669 	}
1670 	if (new_size > device->bdev->bd_inode->i_size) {
1671 		ret = -EFBIG;
1672 		goto out_free;
1673 	}
1674 
1675 	new_size = round_down(new_size, fs_info->sectorsize);
1676 
1677 	btrfs_info_in_rcu(fs_info, "new size for %s is %llu",
1678 			  rcu_str_deref(device->name), new_size);
1679 
1680 	if (new_size > old_size) {
1681 		trans = btrfs_start_transaction(root, 0);
1682 		if (IS_ERR(trans)) {
1683 			ret = PTR_ERR(trans);
1684 			goto out_free;
1685 		}
1686 		ret = btrfs_grow_device(trans, device, new_size);
1687 		btrfs_commit_transaction(trans);
1688 	} else if (new_size < old_size) {
1689 		ret = btrfs_shrink_device(device, new_size);
1690 	} /* equal, nothing need to do */
1691 
1692 out_free:
1693 	kfree(vol_args);
1694 out:
1695 	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
1696 	mnt_drop_write_file(file);
1697 	return ret;
1698 }
1699 
1700 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1701 				const char *name, unsigned long fd, int subvol,
1702 				u64 *transid, bool readonly,
1703 				struct btrfs_qgroup_inherit *inherit)
1704 {
1705 	int namelen;
1706 	int ret = 0;
1707 
1708 	if (!S_ISDIR(file_inode(file)->i_mode))
1709 		return -ENOTDIR;
1710 
1711 	ret = mnt_want_write_file(file);
1712 	if (ret)
1713 		goto out;
1714 
1715 	namelen = strlen(name);
1716 	if (strchr(name, '/')) {
1717 		ret = -EINVAL;
1718 		goto out_drop_write;
1719 	}
1720 
1721 	if (name[0] == '.' &&
1722 	   (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1723 		ret = -EEXIST;
1724 		goto out_drop_write;
1725 	}
1726 
1727 	if (subvol) {
1728 		ret = btrfs_mksubvol(&file->f_path, name, namelen,
1729 				     NULL, transid, readonly, inherit);
1730 	} else {
1731 		struct fd src = fdget(fd);
1732 		struct inode *src_inode;
1733 		if (!src.file) {
1734 			ret = -EINVAL;
1735 			goto out_drop_write;
1736 		}
1737 
1738 		src_inode = file_inode(src.file);
1739 		if (src_inode->i_sb != file_inode(file)->i_sb) {
1740 			btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1741 				   "Snapshot src from another FS");
1742 			ret = -EXDEV;
1743 		} else if (!inode_owner_or_capable(src_inode)) {
1744 			/*
1745 			 * Subvolume creation is not restricted, but snapshots
1746 			 * are limited to own subvolumes only
1747 			 */
1748 			ret = -EPERM;
1749 		} else {
1750 			ret = btrfs_mksubvol(&file->f_path, name, namelen,
1751 					     BTRFS_I(src_inode)->root,
1752 					     transid, readonly, inherit);
1753 		}
1754 		fdput(src);
1755 	}
1756 out_drop_write:
1757 	mnt_drop_write_file(file);
1758 out:
1759 	return ret;
1760 }
1761 
1762 static noinline int btrfs_ioctl_snap_create(struct file *file,
1763 					    void __user *arg, int subvol)
1764 {
1765 	struct btrfs_ioctl_vol_args *vol_args;
1766 	int ret;
1767 
1768 	if (!S_ISDIR(file_inode(file)->i_mode))
1769 		return -ENOTDIR;
1770 
1771 	vol_args = memdup_user(arg, sizeof(*vol_args));
1772 	if (IS_ERR(vol_args))
1773 		return PTR_ERR(vol_args);
1774 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1775 
1776 	ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1777 					      vol_args->fd, subvol,
1778 					      NULL, false, NULL);
1779 
1780 	kfree(vol_args);
1781 	return ret;
1782 }
1783 
1784 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1785 					       void __user *arg, int subvol)
1786 {
1787 	struct btrfs_ioctl_vol_args_v2 *vol_args;
1788 	int ret;
1789 	u64 transid = 0;
1790 	u64 *ptr = NULL;
1791 	bool readonly = false;
1792 	struct btrfs_qgroup_inherit *inherit = NULL;
1793 
1794 	if (!S_ISDIR(file_inode(file)->i_mode))
1795 		return -ENOTDIR;
1796 
1797 	vol_args = memdup_user(arg, sizeof(*vol_args));
1798 	if (IS_ERR(vol_args))
1799 		return PTR_ERR(vol_args);
1800 	vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1801 
1802 	if (vol_args->flags &
1803 	    ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1804 	      BTRFS_SUBVOL_QGROUP_INHERIT)) {
1805 		ret = -EOPNOTSUPP;
1806 		goto free_args;
1807 	}
1808 
1809 	if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1810 		ptr = &transid;
1811 	if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1812 		readonly = true;
1813 	if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1814 		if (vol_args->size > PAGE_SIZE) {
1815 			ret = -EINVAL;
1816 			goto free_args;
1817 		}
1818 		inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1819 		if (IS_ERR(inherit)) {
1820 			ret = PTR_ERR(inherit);
1821 			goto free_args;
1822 		}
1823 	}
1824 
1825 	ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1826 					      vol_args->fd, subvol, ptr,
1827 					      readonly, inherit);
1828 	if (ret)
1829 		goto free_inherit;
1830 
1831 	if (ptr && copy_to_user(arg +
1832 				offsetof(struct btrfs_ioctl_vol_args_v2,
1833 					transid),
1834 				ptr, sizeof(*ptr)))
1835 		ret = -EFAULT;
1836 
1837 free_inherit:
1838 	kfree(inherit);
1839 free_args:
1840 	kfree(vol_args);
1841 	return ret;
1842 }
1843 
1844 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1845 						void __user *arg)
1846 {
1847 	struct inode *inode = file_inode(file);
1848 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1849 	struct btrfs_root *root = BTRFS_I(inode)->root;
1850 	int ret = 0;
1851 	u64 flags = 0;
1852 
1853 	if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1854 		return -EINVAL;
1855 
1856 	down_read(&fs_info->subvol_sem);
1857 	if (btrfs_root_readonly(root))
1858 		flags |= BTRFS_SUBVOL_RDONLY;
1859 	up_read(&fs_info->subvol_sem);
1860 
1861 	if (copy_to_user(arg, &flags, sizeof(flags)))
1862 		ret = -EFAULT;
1863 
1864 	return ret;
1865 }
1866 
1867 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1868 					      void __user *arg)
1869 {
1870 	struct inode *inode = file_inode(file);
1871 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1872 	struct btrfs_root *root = BTRFS_I(inode)->root;
1873 	struct btrfs_trans_handle *trans;
1874 	u64 root_flags;
1875 	u64 flags;
1876 	int ret = 0;
1877 
1878 	if (!inode_owner_or_capable(inode))
1879 		return -EPERM;
1880 
1881 	ret = mnt_want_write_file(file);
1882 	if (ret)
1883 		goto out;
1884 
1885 	if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1886 		ret = -EINVAL;
1887 		goto out_drop_write;
1888 	}
1889 
1890 	if (copy_from_user(&flags, arg, sizeof(flags))) {
1891 		ret = -EFAULT;
1892 		goto out_drop_write;
1893 	}
1894 
1895 	if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1896 		ret = -EINVAL;
1897 		goto out_drop_write;
1898 	}
1899 
1900 	if (flags & ~BTRFS_SUBVOL_RDONLY) {
1901 		ret = -EOPNOTSUPP;
1902 		goto out_drop_write;
1903 	}
1904 
1905 	down_write(&fs_info->subvol_sem);
1906 
1907 	/* nothing to do */
1908 	if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1909 		goto out_drop_sem;
1910 
1911 	root_flags = btrfs_root_flags(&root->root_item);
1912 	if (flags & BTRFS_SUBVOL_RDONLY) {
1913 		btrfs_set_root_flags(&root->root_item,
1914 				     root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1915 	} else {
1916 		/*
1917 		 * Block RO -> RW transition if this subvolume is involved in
1918 		 * send
1919 		 */
1920 		spin_lock(&root->root_item_lock);
1921 		if (root->send_in_progress == 0) {
1922 			btrfs_set_root_flags(&root->root_item,
1923 				     root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1924 			spin_unlock(&root->root_item_lock);
1925 		} else {
1926 			spin_unlock(&root->root_item_lock);
1927 			btrfs_warn(fs_info,
1928 				   "Attempt to set subvolume %llu read-write during send",
1929 				   root->root_key.objectid);
1930 			ret = -EPERM;
1931 			goto out_drop_sem;
1932 		}
1933 	}
1934 
1935 	trans = btrfs_start_transaction(root, 1);
1936 	if (IS_ERR(trans)) {
1937 		ret = PTR_ERR(trans);
1938 		goto out_reset;
1939 	}
1940 
1941 	ret = btrfs_update_root(trans, fs_info->tree_root,
1942 				&root->root_key, &root->root_item);
1943 	if (ret < 0) {
1944 		btrfs_end_transaction(trans);
1945 		goto out_reset;
1946 	}
1947 
1948 	ret = btrfs_commit_transaction(trans);
1949 
1950 out_reset:
1951 	if (ret)
1952 		btrfs_set_root_flags(&root->root_item, root_flags);
1953 out_drop_sem:
1954 	up_write(&fs_info->subvol_sem);
1955 out_drop_write:
1956 	mnt_drop_write_file(file);
1957 out:
1958 	return ret;
1959 }
1960 
1961 static noinline int key_in_sk(struct btrfs_key *key,
1962 			      struct btrfs_ioctl_search_key *sk)
1963 {
1964 	struct btrfs_key test;
1965 	int ret;
1966 
1967 	test.objectid = sk->min_objectid;
1968 	test.type = sk->min_type;
1969 	test.offset = sk->min_offset;
1970 
1971 	ret = btrfs_comp_cpu_keys(key, &test);
1972 	if (ret < 0)
1973 		return 0;
1974 
1975 	test.objectid = sk->max_objectid;
1976 	test.type = sk->max_type;
1977 	test.offset = sk->max_offset;
1978 
1979 	ret = btrfs_comp_cpu_keys(key, &test);
1980 	if (ret > 0)
1981 		return 0;
1982 	return 1;
1983 }
1984 
1985 static noinline int copy_to_sk(struct btrfs_path *path,
1986 			       struct btrfs_key *key,
1987 			       struct btrfs_ioctl_search_key *sk,
1988 			       size_t *buf_size,
1989 			       char __user *ubuf,
1990 			       unsigned long *sk_offset,
1991 			       int *num_found)
1992 {
1993 	u64 found_transid;
1994 	struct extent_buffer *leaf;
1995 	struct btrfs_ioctl_search_header sh;
1996 	struct btrfs_key test;
1997 	unsigned long item_off;
1998 	unsigned long item_len;
1999 	int nritems;
2000 	int i;
2001 	int slot;
2002 	int ret = 0;
2003 
2004 	leaf = path->nodes[0];
2005 	slot = path->slots[0];
2006 	nritems = btrfs_header_nritems(leaf);
2007 
2008 	if (btrfs_header_generation(leaf) > sk->max_transid) {
2009 		i = nritems;
2010 		goto advance_key;
2011 	}
2012 	found_transid = btrfs_header_generation(leaf);
2013 
2014 	for (i = slot; i < nritems; i++) {
2015 		item_off = btrfs_item_ptr_offset(leaf, i);
2016 		item_len = btrfs_item_size_nr(leaf, i);
2017 
2018 		btrfs_item_key_to_cpu(leaf, key, i);
2019 		if (!key_in_sk(key, sk))
2020 			continue;
2021 
2022 		if (sizeof(sh) + item_len > *buf_size) {
2023 			if (*num_found) {
2024 				ret = 1;
2025 				goto out;
2026 			}
2027 
2028 			/*
2029 			 * return one empty item back for v1, which does not
2030 			 * handle -EOVERFLOW
2031 			 */
2032 
2033 			*buf_size = sizeof(sh) + item_len;
2034 			item_len = 0;
2035 			ret = -EOVERFLOW;
2036 		}
2037 
2038 		if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2039 			ret = 1;
2040 			goto out;
2041 		}
2042 
2043 		sh.objectid = key->objectid;
2044 		sh.offset = key->offset;
2045 		sh.type = key->type;
2046 		sh.len = item_len;
2047 		sh.transid = found_transid;
2048 
2049 		/* copy search result header */
2050 		if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
2051 			ret = -EFAULT;
2052 			goto out;
2053 		}
2054 
2055 		*sk_offset += sizeof(sh);
2056 
2057 		if (item_len) {
2058 			char __user *up = ubuf + *sk_offset;
2059 			/* copy the item */
2060 			if (read_extent_buffer_to_user(leaf, up,
2061 						       item_off, item_len)) {
2062 				ret = -EFAULT;
2063 				goto out;
2064 			}
2065 
2066 			*sk_offset += item_len;
2067 		}
2068 		(*num_found)++;
2069 
2070 		if (ret) /* -EOVERFLOW from above */
2071 			goto out;
2072 
2073 		if (*num_found >= sk->nr_items) {
2074 			ret = 1;
2075 			goto out;
2076 		}
2077 	}
2078 advance_key:
2079 	ret = 0;
2080 	test.objectid = sk->max_objectid;
2081 	test.type = sk->max_type;
2082 	test.offset = sk->max_offset;
2083 	if (btrfs_comp_cpu_keys(key, &test) >= 0)
2084 		ret = 1;
2085 	else if (key->offset < (u64)-1)
2086 		key->offset++;
2087 	else if (key->type < (u8)-1) {
2088 		key->offset = 0;
2089 		key->type++;
2090 	} else if (key->objectid < (u64)-1) {
2091 		key->offset = 0;
2092 		key->type = 0;
2093 		key->objectid++;
2094 	} else
2095 		ret = 1;
2096 out:
2097 	/*
2098 	 *  0: all items from this leaf copied, continue with next
2099 	 *  1: * more items can be copied, but unused buffer is too small
2100 	 *     * all items were found
2101 	 *     Either way, it will stops the loop which iterates to the next
2102 	 *     leaf
2103 	 *  -EOVERFLOW: item was to large for buffer
2104 	 *  -EFAULT: could not copy extent buffer back to userspace
2105 	 */
2106 	return ret;
2107 }
2108 
2109 static noinline int search_ioctl(struct inode *inode,
2110 				 struct btrfs_ioctl_search_key *sk,
2111 				 size_t *buf_size,
2112 				 char __user *ubuf)
2113 {
2114 	struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2115 	struct btrfs_root *root;
2116 	struct btrfs_key key;
2117 	struct btrfs_path *path;
2118 	int ret;
2119 	int num_found = 0;
2120 	unsigned long sk_offset = 0;
2121 
2122 	if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2123 		*buf_size = sizeof(struct btrfs_ioctl_search_header);
2124 		return -EOVERFLOW;
2125 	}
2126 
2127 	path = btrfs_alloc_path();
2128 	if (!path)
2129 		return -ENOMEM;
2130 
2131 	if (sk->tree_id == 0) {
2132 		/* search the root of the inode that was passed */
2133 		root = BTRFS_I(inode)->root;
2134 	} else {
2135 		key.objectid = sk->tree_id;
2136 		key.type = BTRFS_ROOT_ITEM_KEY;
2137 		key.offset = (u64)-1;
2138 		root = btrfs_read_fs_root_no_name(info, &key);
2139 		if (IS_ERR(root)) {
2140 			btrfs_free_path(path);
2141 			return PTR_ERR(root);
2142 		}
2143 	}
2144 
2145 	key.objectid = sk->min_objectid;
2146 	key.type = sk->min_type;
2147 	key.offset = sk->min_offset;
2148 
2149 	while (1) {
2150 		ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2151 		if (ret != 0) {
2152 			if (ret > 0)
2153 				ret = 0;
2154 			goto err;
2155 		}
2156 		ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2157 				 &sk_offset, &num_found);
2158 		btrfs_release_path(path);
2159 		if (ret)
2160 			break;
2161 
2162 	}
2163 	if (ret > 0)
2164 		ret = 0;
2165 err:
2166 	sk->nr_items = num_found;
2167 	btrfs_free_path(path);
2168 	return ret;
2169 }
2170 
2171 static noinline int btrfs_ioctl_tree_search(struct file *file,
2172 					   void __user *argp)
2173 {
2174 	struct btrfs_ioctl_search_args __user *uargs;
2175 	struct btrfs_ioctl_search_key sk;
2176 	struct inode *inode;
2177 	int ret;
2178 	size_t buf_size;
2179 
2180 	if (!capable(CAP_SYS_ADMIN))
2181 		return -EPERM;
2182 
2183 	uargs = (struct btrfs_ioctl_search_args __user *)argp;
2184 
2185 	if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2186 		return -EFAULT;
2187 
2188 	buf_size = sizeof(uargs->buf);
2189 
2190 	inode = file_inode(file);
2191 	ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2192 
2193 	/*
2194 	 * In the origin implementation an overflow is handled by returning a
2195 	 * search header with a len of zero, so reset ret.
2196 	 */
2197 	if (ret == -EOVERFLOW)
2198 		ret = 0;
2199 
2200 	if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2201 		ret = -EFAULT;
2202 	return ret;
2203 }
2204 
2205 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2206 					       void __user *argp)
2207 {
2208 	struct btrfs_ioctl_search_args_v2 __user *uarg;
2209 	struct btrfs_ioctl_search_args_v2 args;
2210 	struct inode *inode;
2211 	int ret;
2212 	size_t buf_size;
2213 	const size_t buf_limit = SZ_16M;
2214 
2215 	if (!capable(CAP_SYS_ADMIN))
2216 		return -EPERM;
2217 
2218 	/* copy search header and buffer size */
2219 	uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2220 	if (copy_from_user(&args, uarg, sizeof(args)))
2221 		return -EFAULT;
2222 
2223 	buf_size = args.buf_size;
2224 
2225 	/* limit result size to 16MB */
2226 	if (buf_size > buf_limit)
2227 		buf_size = buf_limit;
2228 
2229 	inode = file_inode(file);
2230 	ret = search_ioctl(inode, &args.key, &buf_size,
2231 			   (char __user *)(&uarg->buf[0]));
2232 	if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2233 		ret = -EFAULT;
2234 	else if (ret == -EOVERFLOW &&
2235 		copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2236 		ret = -EFAULT;
2237 
2238 	return ret;
2239 }
2240 
2241 /*
2242  * Search INODE_REFs to identify path name of 'dirid' directory
2243  * in a 'tree_id' tree. and sets path name to 'name'.
2244  */
2245 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2246 				u64 tree_id, u64 dirid, char *name)
2247 {
2248 	struct btrfs_root *root;
2249 	struct btrfs_key key;
2250 	char *ptr;
2251 	int ret = -1;
2252 	int slot;
2253 	int len;
2254 	int total_len = 0;
2255 	struct btrfs_inode_ref *iref;
2256 	struct extent_buffer *l;
2257 	struct btrfs_path *path;
2258 
2259 	if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2260 		name[0]='\0';
2261 		return 0;
2262 	}
2263 
2264 	path = btrfs_alloc_path();
2265 	if (!path)
2266 		return -ENOMEM;
2267 
2268 	ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2269 
2270 	key.objectid = tree_id;
2271 	key.type = BTRFS_ROOT_ITEM_KEY;
2272 	key.offset = (u64)-1;
2273 	root = btrfs_read_fs_root_no_name(info, &key);
2274 	if (IS_ERR(root)) {
2275 		ret = PTR_ERR(root);
2276 		goto out;
2277 	}
2278 
2279 	key.objectid = dirid;
2280 	key.type = BTRFS_INODE_REF_KEY;
2281 	key.offset = (u64)-1;
2282 
2283 	while (1) {
2284 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2285 		if (ret < 0)
2286 			goto out;
2287 		else if (ret > 0) {
2288 			ret = btrfs_previous_item(root, path, dirid,
2289 						  BTRFS_INODE_REF_KEY);
2290 			if (ret < 0)
2291 				goto out;
2292 			else if (ret > 0) {
2293 				ret = -ENOENT;
2294 				goto out;
2295 			}
2296 		}
2297 
2298 		l = path->nodes[0];
2299 		slot = path->slots[0];
2300 		btrfs_item_key_to_cpu(l, &key, slot);
2301 
2302 		iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2303 		len = btrfs_inode_ref_name_len(l, iref);
2304 		ptr -= len + 1;
2305 		total_len += len + 1;
2306 		if (ptr < name) {
2307 			ret = -ENAMETOOLONG;
2308 			goto out;
2309 		}
2310 
2311 		*(ptr + len) = '/';
2312 		read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2313 
2314 		if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2315 			break;
2316 
2317 		btrfs_release_path(path);
2318 		key.objectid = key.offset;
2319 		key.offset = (u64)-1;
2320 		dirid = key.objectid;
2321 	}
2322 	memmove(name, ptr, total_len);
2323 	name[total_len] = '\0';
2324 	ret = 0;
2325 out:
2326 	btrfs_free_path(path);
2327 	return ret;
2328 }
2329 
2330 static int btrfs_search_path_in_tree_user(struct inode *inode,
2331 				struct btrfs_ioctl_ino_lookup_user_args *args)
2332 {
2333 	struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2334 	struct super_block *sb = inode->i_sb;
2335 	struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2336 	u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2337 	u64 dirid = args->dirid;
2338 	unsigned long item_off;
2339 	unsigned long item_len;
2340 	struct btrfs_inode_ref *iref;
2341 	struct btrfs_root_ref *rref;
2342 	struct btrfs_root *root;
2343 	struct btrfs_path *path;
2344 	struct btrfs_key key, key2;
2345 	struct extent_buffer *leaf;
2346 	struct inode *temp_inode;
2347 	char *ptr;
2348 	int slot;
2349 	int len;
2350 	int total_len = 0;
2351 	int ret;
2352 
2353 	path = btrfs_alloc_path();
2354 	if (!path)
2355 		return -ENOMEM;
2356 
2357 	/*
2358 	 * If the bottom subvolume does not exist directly under upper_limit,
2359 	 * construct the path in from the bottom up.
2360 	 */
2361 	if (dirid != upper_limit.objectid) {
2362 		ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2363 
2364 		key.objectid = treeid;
2365 		key.type = BTRFS_ROOT_ITEM_KEY;
2366 		key.offset = (u64)-1;
2367 		root = btrfs_read_fs_root_no_name(fs_info, &key);
2368 		if (IS_ERR(root)) {
2369 			ret = PTR_ERR(root);
2370 			goto out;
2371 		}
2372 
2373 		key.objectid = dirid;
2374 		key.type = BTRFS_INODE_REF_KEY;
2375 		key.offset = (u64)-1;
2376 		while (1) {
2377 			ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2378 			if (ret < 0) {
2379 				goto out;
2380 			} else if (ret > 0) {
2381 				ret = btrfs_previous_item(root, path, dirid,
2382 							  BTRFS_INODE_REF_KEY);
2383 				if (ret < 0) {
2384 					goto out;
2385 				} else if (ret > 0) {
2386 					ret = -ENOENT;
2387 					goto out;
2388 				}
2389 			}
2390 
2391 			leaf = path->nodes[0];
2392 			slot = path->slots[0];
2393 			btrfs_item_key_to_cpu(leaf, &key, slot);
2394 
2395 			iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2396 			len = btrfs_inode_ref_name_len(leaf, iref);
2397 			ptr -= len + 1;
2398 			total_len += len + 1;
2399 			if (ptr < args->path) {
2400 				ret = -ENAMETOOLONG;
2401 				goto out;
2402 			}
2403 
2404 			*(ptr + len) = '/';
2405 			read_extent_buffer(leaf, ptr,
2406 					(unsigned long)(iref + 1), len);
2407 
2408 			/* Check the read+exec permission of this directory */
2409 			ret = btrfs_previous_item(root, path, dirid,
2410 						  BTRFS_INODE_ITEM_KEY);
2411 			if (ret < 0) {
2412 				goto out;
2413 			} else if (ret > 0) {
2414 				ret = -ENOENT;
2415 				goto out;
2416 			}
2417 
2418 			leaf = path->nodes[0];
2419 			slot = path->slots[0];
2420 			btrfs_item_key_to_cpu(leaf, &key2, slot);
2421 			if (key2.objectid != dirid) {
2422 				ret = -ENOENT;
2423 				goto out;
2424 			}
2425 
2426 			temp_inode = btrfs_iget(sb, &key2, root, NULL);
2427 			if (IS_ERR(temp_inode)) {
2428 				ret = PTR_ERR(temp_inode);
2429 				goto out;
2430 			}
2431 			ret = inode_permission(temp_inode, MAY_READ | MAY_EXEC);
2432 			iput(temp_inode);
2433 			if (ret) {
2434 				ret = -EACCES;
2435 				goto out;
2436 			}
2437 
2438 			if (key.offset == upper_limit.objectid)
2439 				break;
2440 			if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2441 				ret = -EACCES;
2442 				goto out;
2443 			}
2444 
2445 			btrfs_release_path(path);
2446 			key.objectid = key.offset;
2447 			key.offset = (u64)-1;
2448 			dirid = key.objectid;
2449 		}
2450 
2451 		memmove(args->path, ptr, total_len);
2452 		args->path[total_len] = '\0';
2453 		btrfs_release_path(path);
2454 	}
2455 
2456 	/* Get the bottom subvolume's name from ROOT_REF */
2457 	root = fs_info->tree_root;
2458 	key.objectid = treeid;
2459 	key.type = BTRFS_ROOT_REF_KEY;
2460 	key.offset = args->treeid;
2461 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2462 	if (ret < 0) {
2463 		goto out;
2464 	} else if (ret > 0) {
2465 		ret = -ENOENT;
2466 		goto out;
2467 	}
2468 
2469 	leaf = path->nodes[0];
2470 	slot = path->slots[0];
2471 	btrfs_item_key_to_cpu(leaf, &key, slot);
2472 
2473 	item_off = btrfs_item_ptr_offset(leaf, slot);
2474 	item_len = btrfs_item_size_nr(leaf, slot);
2475 	/* Check if dirid in ROOT_REF corresponds to passed dirid */
2476 	rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2477 	if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2478 		ret = -EINVAL;
2479 		goto out;
2480 	}
2481 
2482 	/* Copy subvolume's name */
2483 	item_off += sizeof(struct btrfs_root_ref);
2484 	item_len -= sizeof(struct btrfs_root_ref);
2485 	read_extent_buffer(leaf, args->name, item_off, item_len);
2486 	args->name[item_len] = 0;
2487 
2488 out:
2489 	btrfs_free_path(path);
2490 	return ret;
2491 }
2492 
2493 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2494 					   void __user *argp)
2495 {
2496 	struct btrfs_ioctl_ino_lookup_args *args;
2497 	struct inode *inode;
2498 	int ret = 0;
2499 
2500 	args = memdup_user(argp, sizeof(*args));
2501 	if (IS_ERR(args))
2502 		return PTR_ERR(args);
2503 
2504 	inode = file_inode(file);
2505 
2506 	/*
2507 	 * Unprivileged query to obtain the containing subvolume root id. The
2508 	 * path is reset so it's consistent with btrfs_search_path_in_tree.
2509 	 */
2510 	if (args->treeid == 0)
2511 		args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2512 
2513 	if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2514 		args->name[0] = 0;
2515 		goto out;
2516 	}
2517 
2518 	if (!capable(CAP_SYS_ADMIN)) {
2519 		ret = -EPERM;
2520 		goto out;
2521 	}
2522 
2523 	ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2524 					args->treeid, args->objectid,
2525 					args->name);
2526 
2527 out:
2528 	if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2529 		ret = -EFAULT;
2530 
2531 	kfree(args);
2532 	return ret;
2533 }
2534 
2535 /*
2536  * Version of ino_lookup ioctl (unprivileged)
2537  *
2538  * The main differences from ino_lookup ioctl are:
2539  *
2540  *   1. Read + Exec permission will be checked using inode_permission() during
2541  *      path construction. -EACCES will be returned in case of failure.
2542  *   2. Path construction will be stopped at the inode number which corresponds
2543  *      to the fd with which this ioctl is called. If constructed path does not
2544  *      exist under fd's inode, -EACCES will be returned.
2545  *   3. The name of bottom subvolume is also searched and filled.
2546  */
2547 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2548 {
2549 	struct btrfs_ioctl_ino_lookup_user_args *args;
2550 	struct inode *inode;
2551 	int ret;
2552 
2553 	args = memdup_user(argp, sizeof(*args));
2554 	if (IS_ERR(args))
2555 		return PTR_ERR(args);
2556 
2557 	inode = file_inode(file);
2558 
2559 	if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2560 	    BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2561 		/*
2562 		 * The subvolume does not exist under fd with which this is
2563 		 * called
2564 		 */
2565 		kfree(args);
2566 		return -EACCES;
2567 	}
2568 
2569 	ret = btrfs_search_path_in_tree_user(inode, args);
2570 
2571 	if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2572 		ret = -EFAULT;
2573 
2574 	kfree(args);
2575 	return ret;
2576 }
2577 
2578 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2579 static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp)
2580 {
2581 	struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2582 	struct btrfs_fs_info *fs_info;
2583 	struct btrfs_root *root;
2584 	struct btrfs_path *path;
2585 	struct btrfs_key key;
2586 	struct btrfs_root_item *root_item;
2587 	struct btrfs_root_ref *rref;
2588 	struct extent_buffer *leaf;
2589 	unsigned long item_off;
2590 	unsigned long item_len;
2591 	struct inode *inode;
2592 	int slot;
2593 	int ret = 0;
2594 
2595 	path = btrfs_alloc_path();
2596 	if (!path)
2597 		return -ENOMEM;
2598 
2599 	subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2600 	if (!subvol_info) {
2601 		btrfs_free_path(path);
2602 		return -ENOMEM;
2603 	}
2604 
2605 	inode = file_inode(file);
2606 	fs_info = BTRFS_I(inode)->root->fs_info;
2607 
2608 	/* Get root_item of inode's subvolume */
2609 	key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2610 	key.type = BTRFS_ROOT_ITEM_KEY;
2611 	key.offset = (u64)-1;
2612 	root = btrfs_read_fs_root_no_name(fs_info, &key);
2613 	if (IS_ERR(root)) {
2614 		ret = PTR_ERR(root);
2615 		goto out;
2616 	}
2617 	root_item = &root->root_item;
2618 
2619 	subvol_info->treeid = key.objectid;
2620 
2621 	subvol_info->generation = btrfs_root_generation(root_item);
2622 	subvol_info->flags = btrfs_root_flags(root_item);
2623 
2624 	memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2625 	memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2626 						    BTRFS_UUID_SIZE);
2627 	memcpy(subvol_info->received_uuid, root_item->received_uuid,
2628 						    BTRFS_UUID_SIZE);
2629 
2630 	subvol_info->ctransid = btrfs_root_ctransid(root_item);
2631 	subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2632 	subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2633 
2634 	subvol_info->otransid = btrfs_root_otransid(root_item);
2635 	subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2636 	subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2637 
2638 	subvol_info->stransid = btrfs_root_stransid(root_item);
2639 	subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2640 	subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2641 
2642 	subvol_info->rtransid = btrfs_root_rtransid(root_item);
2643 	subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2644 	subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2645 
2646 	if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2647 		/* Search root tree for ROOT_BACKREF of this subvolume */
2648 		root = fs_info->tree_root;
2649 
2650 		key.type = BTRFS_ROOT_BACKREF_KEY;
2651 		key.offset = 0;
2652 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2653 		if (ret < 0) {
2654 			goto out;
2655 		} else if (path->slots[0] >=
2656 			   btrfs_header_nritems(path->nodes[0])) {
2657 			ret = btrfs_next_leaf(root, path);
2658 			if (ret < 0) {
2659 				goto out;
2660 			} else if (ret > 0) {
2661 				ret = -EUCLEAN;
2662 				goto out;
2663 			}
2664 		}
2665 
2666 		leaf = path->nodes[0];
2667 		slot = path->slots[0];
2668 		btrfs_item_key_to_cpu(leaf, &key, slot);
2669 		if (key.objectid == subvol_info->treeid &&
2670 		    key.type == BTRFS_ROOT_BACKREF_KEY) {
2671 			subvol_info->parent_id = key.offset;
2672 
2673 			rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2674 			subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2675 
2676 			item_off = btrfs_item_ptr_offset(leaf, slot)
2677 					+ sizeof(struct btrfs_root_ref);
2678 			item_len = btrfs_item_size_nr(leaf, slot)
2679 					- sizeof(struct btrfs_root_ref);
2680 			read_extent_buffer(leaf, subvol_info->name,
2681 					   item_off, item_len);
2682 		} else {
2683 			ret = -ENOENT;
2684 			goto out;
2685 		}
2686 	}
2687 
2688 	if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2689 		ret = -EFAULT;
2690 
2691 out:
2692 	btrfs_free_path(path);
2693 	kzfree(subvol_info);
2694 	return ret;
2695 }
2696 
2697 /*
2698  * Return ROOT_REF information of the subvolume containing this inode
2699  * except the subvolume name.
2700  */
2701 static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp)
2702 {
2703 	struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2704 	struct btrfs_root_ref *rref;
2705 	struct btrfs_root *root;
2706 	struct btrfs_path *path;
2707 	struct btrfs_key key;
2708 	struct extent_buffer *leaf;
2709 	struct inode *inode;
2710 	u64 objectid;
2711 	int slot;
2712 	int ret;
2713 	u8 found;
2714 
2715 	path = btrfs_alloc_path();
2716 	if (!path)
2717 		return -ENOMEM;
2718 
2719 	rootrefs = memdup_user(argp, sizeof(*rootrefs));
2720 	if (IS_ERR(rootrefs)) {
2721 		btrfs_free_path(path);
2722 		return PTR_ERR(rootrefs);
2723 	}
2724 
2725 	inode = file_inode(file);
2726 	root = BTRFS_I(inode)->root->fs_info->tree_root;
2727 	objectid = BTRFS_I(inode)->root->root_key.objectid;
2728 
2729 	key.objectid = objectid;
2730 	key.type = BTRFS_ROOT_REF_KEY;
2731 	key.offset = rootrefs->min_treeid;
2732 	found = 0;
2733 
2734 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2735 	if (ret < 0) {
2736 		goto out;
2737 	} else if (path->slots[0] >=
2738 		   btrfs_header_nritems(path->nodes[0])) {
2739 		ret = btrfs_next_leaf(root, path);
2740 		if (ret < 0) {
2741 			goto out;
2742 		} else if (ret > 0) {
2743 			ret = -EUCLEAN;
2744 			goto out;
2745 		}
2746 	}
2747 	while (1) {
2748 		leaf = path->nodes[0];
2749 		slot = path->slots[0];
2750 
2751 		btrfs_item_key_to_cpu(leaf, &key, slot);
2752 		if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2753 			ret = 0;
2754 			goto out;
2755 		}
2756 
2757 		if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2758 			ret = -EOVERFLOW;
2759 			goto out;
2760 		}
2761 
2762 		rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2763 		rootrefs->rootref[found].treeid = key.offset;
2764 		rootrefs->rootref[found].dirid =
2765 				  btrfs_root_ref_dirid(leaf, rref);
2766 		found++;
2767 
2768 		ret = btrfs_next_item(root, path);
2769 		if (ret < 0) {
2770 			goto out;
2771 		} else if (ret > 0) {
2772 			ret = -EUCLEAN;
2773 			goto out;
2774 		}
2775 	}
2776 
2777 out:
2778 	if (!ret || ret == -EOVERFLOW) {
2779 		rootrefs->num_items = found;
2780 		/* update min_treeid for next search */
2781 		if (found)
2782 			rootrefs->min_treeid =
2783 				rootrefs->rootref[found - 1].treeid + 1;
2784 		if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2785 			ret = -EFAULT;
2786 	}
2787 
2788 	kfree(rootrefs);
2789 	btrfs_free_path(path);
2790 
2791 	return ret;
2792 }
2793 
2794 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2795 					     void __user *arg)
2796 {
2797 	struct dentry *parent = file->f_path.dentry;
2798 	struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2799 	struct dentry *dentry;
2800 	struct inode *dir = d_inode(parent);
2801 	struct inode *inode;
2802 	struct btrfs_root *root = BTRFS_I(dir)->root;
2803 	struct btrfs_root *dest = NULL;
2804 	struct btrfs_ioctl_vol_args *vol_args;
2805 	int namelen;
2806 	int err = 0;
2807 
2808 	if (!S_ISDIR(dir->i_mode))
2809 		return -ENOTDIR;
2810 
2811 	vol_args = memdup_user(arg, sizeof(*vol_args));
2812 	if (IS_ERR(vol_args))
2813 		return PTR_ERR(vol_args);
2814 
2815 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2816 	namelen = strlen(vol_args->name);
2817 	if (strchr(vol_args->name, '/') ||
2818 	    strncmp(vol_args->name, "..", namelen) == 0) {
2819 		err = -EINVAL;
2820 		goto out;
2821 	}
2822 
2823 	err = mnt_want_write_file(file);
2824 	if (err)
2825 		goto out;
2826 
2827 
2828 	err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2829 	if (err == -EINTR)
2830 		goto out_drop_write;
2831 	dentry = lookup_one_len(vol_args->name, parent, namelen);
2832 	if (IS_ERR(dentry)) {
2833 		err = PTR_ERR(dentry);
2834 		goto out_unlock_dir;
2835 	}
2836 
2837 	if (d_really_is_negative(dentry)) {
2838 		err = -ENOENT;
2839 		goto out_dput;
2840 	}
2841 
2842 	inode = d_inode(dentry);
2843 	dest = BTRFS_I(inode)->root;
2844 	if (!capable(CAP_SYS_ADMIN)) {
2845 		/*
2846 		 * Regular user.  Only allow this with a special mount
2847 		 * option, when the user has write+exec access to the
2848 		 * subvol root, and when rmdir(2) would have been
2849 		 * allowed.
2850 		 *
2851 		 * Note that this is _not_ check that the subvol is
2852 		 * empty or doesn't contain data that we wouldn't
2853 		 * otherwise be able to delete.
2854 		 *
2855 		 * Users who want to delete empty subvols should try
2856 		 * rmdir(2).
2857 		 */
2858 		err = -EPERM;
2859 		if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2860 			goto out_dput;
2861 
2862 		/*
2863 		 * Do not allow deletion if the parent dir is the same
2864 		 * as the dir to be deleted.  That means the ioctl
2865 		 * must be called on the dentry referencing the root
2866 		 * of the subvol, not a random directory contained
2867 		 * within it.
2868 		 */
2869 		err = -EINVAL;
2870 		if (root == dest)
2871 			goto out_dput;
2872 
2873 		err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2874 		if (err)
2875 			goto out_dput;
2876 	}
2877 
2878 	/* check if subvolume may be deleted by a user */
2879 	err = btrfs_may_delete(dir, dentry, 1);
2880 	if (err)
2881 		goto out_dput;
2882 
2883 	if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2884 		err = -EINVAL;
2885 		goto out_dput;
2886 	}
2887 
2888 	inode_lock(inode);
2889 	err = btrfs_delete_subvolume(dir, dentry);
2890 	inode_unlock(inode);
2891 	if (!err)
2892 		d_delete(dentry);
2893 
2894 out_dput:
2895 	dput(dentry);
2896 out_unlock_dir:
2897 	inode_unlock(dir);
2898 out_drop_write:
2899 	mnt_drop_write_file(file);
2900 out:
2901 	kfree(vol_args);
2902 	return err;
2903 }
2904 
2905 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2906 {
2907 	struct inode *inode = file_inode(file);
2908 	struct btrfs_root *root = BTRFS_I(inode)->root;
2909 	struct btrfs_ioctl_defrag_range_args *range;
2910 	int ret;
2911 
2912 	ret = mnt_want_write_file(file);
2913 	if (ret)
2914 		return ret;
2915 
2916 	if (btrfs_root_readonly(root)) {
2917 		ret = -EROFS;
2918 		goto out;
2919 	}
2920 
2921 	switch (inode->i_mode & S_IFMT) {
2922 	case S_IFDIR:
2923 		if (!capable(CAP_SYS_ADMIN)) {
2924 			ret = -EPERM;
2925 			goto out;
2926 		}
2927 		ret = btrfs_defrag_root(root);
2928 		break;
2929 	case S_IFREG:
2930 		/*
2931 		 * Note that this does not check the file descriptor for write
2932 		 * access. This prevents defragmenting executables that are
2933 		 * running and allows defrag on files open in read-only mode.
2934 		 */
2935 		if (!capable(CAP_SYS_ADMIN) &&
2936 		    inode_permission(inode, MAY_WRITE)) {
2937 			ret = -EPERM;
2938 			goto out;
2939 		}
2940 
2941 		range = kzalloc(sizeof(*range), GFP_KERNEL);
2942 		if (!range) {
2943 			ret = -ENOMEM;
2944 			goto out;
2945 		}
2946 
2947 		if (argp) {
2948 			if (copy_from_user(range, argp,
2949 					   sizeof(*range))) {
2950 				ret = -EFAULT;
2951 				kfree(range);
2952 				goto out;
2953 			}
2954 			/* compression requires us to start the IO */
2955 			if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2956 				range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2957 				range->extent_thresh = (u32)-1;
2958 			}
2959 		} else {
2960 			/* the rest are all set to zero by kzalloc */
2961 			range->len = (u64)-1;
2962 		}
2963 		ret = btrfs_defrag_file(file_inode(file), file,
2964 					range, BTRFS_OLDEST_GENERATION, 0);
2965 		if (ret > 0)
2966 			ret = 0;
2967 		kfree(range);
2968 		break;
2969 	default:
2970 		ret = -EINVAL;
2971 	}
2972 out:
2973 	mnt_drop_write_file(file);
2974 	return ret;
2975 }
2976 
2977 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
2978 {
2979 	struct btrfs_ioctl_vol_args *vol_args;
2980 	int ret;
2981 
2982 	if (!capable(CAP_SYS_ADMIN))
2983 		return -EPERM;
2984 
2985 	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
2986 		return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2987 
2988 	vol_args = memdup_user(arg, sizeof(*vol_args));
2989 	if (IS_ERR(vol_args)) {
2990 		ret = PTR_ERR(vol_args);
2991 		goto out;
2992 	}
2993 
2994 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2995 	ret = btrfs_init_new_device(fs_info, vol_args->name);
2996 
2997 	if (!ret)
2998 		btrfs_info(fs_info, "disk added %s", vol_args->name);
2999 
3000 	kfree(vol_args);
3001 out:
3002 	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3003 	return ret;
3004 }
3005 
3006 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3007 {
3008 	struct inode *inode = file_inode(file);
3009 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3010 	struct btrfs_ioctl_vol_args_v2 *vol_args;
3011 	int ret;
3012 
3013 	if (!capable(CAP_SYS_ADMIN))
3014 		return -EPERM;
3015 
3016 	ret = mnt_want_write_file(file);
3017 	if (ret)
3018 		return ret;
3019 
3020 	vol_args = memdup_user(arg, sizeof(*vol_args));
3021 	if (IS_ERR(vol_args)) {
3022 		ret = PTR_ERR(vol_args);
3023 		goto err_drop;
3024 	}
3025 
3026 	/* Check for compatibility reject unknown flags */
3027 	if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED) {
3028 		ret = -EOPNOTSUPP;
3029 		goto out;
3030 	}
3031 
3032 	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3033 		ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3034 		goto out;
3035 	}
3036 
3037 	if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3038 		ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
3039 	} else {
3040 		vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3041 		ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3042 	}
3043 	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3044 
3045 	if (!ret) {
3046 		if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3047 			btrfs_info(fs_info, "device deleted: id %llu",
3048 					vol_args->devid);
3049 		else
3050 			btrfs_info(fs_info, "device deleted: %s",
3051 					vol_args->name);
3052 	}
3053 out:
3054 	kfree(vol_args);
3055 err_drop:
3056 	mnt_drop_write_file(file);
3057 	return ret;
3058 }
3059 
3060 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3061 {
3062 	struct inode *inode = file_inode(file);
3063 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3064 	struct btrfs_ioctl_vol_args *vol_args;
3065 	int ret;
3066 
3067 	if (!capable(CAP_SYS_ADMIN))
3068 		return -EPERM;
3069 
3070 	ret = mnt_want_write_file(file);
3071 	if (ret)
3072 		return ret;
3073 
3074 	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3075 		ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3076 		goto out_drop_write;
3077 	}
3078 
3079 	vol_args = memdup_user(arg, sizeof(*vol_args));
3080 	if (IS_ERR(vol_args)) {
3081 		ret = PTR_ERR(vol_args);
3082 		goto out;
3083 	}
3084 
3085 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3086 	ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3087 
3088 	if (!ret)
3089 		btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3090 	kfree(vol_args);
3091 out:
3092 	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3093 out_drop_write:
3094 	mnt_drop_write_file(file);
3095 
3096 	return ret;
3097 }
3098 
3099 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3100 				void __user *arg)
3101 {
3102 	struct btrfs_ioctl_fs_info_args *fi_args;
3103 	struct btrfs_device *device;
3104 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3105 	int ret = 0;
3106 
3107 	fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
3108 	if (!fi_args)
3109 		return -ENOMEM;
3110 
3111 	rcu_read_lock();
3112 	fi_args->num_devices = fs_devices->num_devices;
3113 
3114 	list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3115 		if (device->devid > fi_args->max_id)
3116 			fi_args->max_id = device->devid;
3117 	}
3118 	rcu_read_unlock();
3119 
3120 	memcpy(&fi_args->fsid, fs_info->fsid, sizeof(fi_args->fsid));
3121 	fi_args->nodesize = fs_info->nodesize;
3122 	fi_args->sectorsize = fs_info->sectorsize;
3123 	fi_args->clone_alignment = fs_info->sectorsize;
3124 
3125 	if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3126 		ret = -EFAULT;
3127 
3128 	kfree(fi_args);
3129 	return ret;
3130 }
3131 
3132 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3133 				 void __user *arg)
3134 {
3135 	struct btrfs_ioctl_dev_info_args *di_args;
3136 	struct btrfs_device *dev;
3137 	int ret = 0;
3138 	char *s_uuid = NULL;
3139 
3140 	di_args = memdup_user(arg, sizeof(*di_args));
3141 	if (IS_ERR(di_args))
3142 		return PTR_ERR(di_args);
3143 
3144 	if (!btrfs_is_empty_uuid(di_args->uuid))
3145 		s_uuid = di_args->uuid;
3146 
3147 	rcu_read_lock();
3148 	dev = btrfs_find_device(fs_info, di_args->devid, s_uuid, NULL);
3149 
3150 	if (!dev) {
3151 		ret = -ENODEV;
3152 		goto out;
3153 	}
3154 
3155 	di_args->devid = dev->devid;
3156 	di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3157 	di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3158 	memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3159 	if (dev->name) {
3160 		strncpy(di_args->path, rcu_str_deref(dev->name),
3161 				sizeof(di_args->path) - 1);
3162 		di_args->path[sizeof(di_args->path) - 1] = 0;
3163 	} else {
3164 		di_args->path[0] = '\0';
3165 	}
3166 
3167 out:
3168 	rcu_read_unlock();
3169 	if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3170 		ret = -EFAULT;
3171 
3172 	kfree(di_args);
3173 	return ret;
3174 }
3175 
3176 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
3177 {
3178 	struct page *page;
3179 
3180 	page = grab_cache_page(inode->i_mapping, index);
3181 	if (!page)
3182 		return ERR_PTR(-ENOMEM);
3183 
3184 	if (!PageUptodate(page)) {
3185 		int ret;
3186 
3187 		ret = btrfs_readpage(NULL, page);
3188 		if (ret)
3189 			return ERR_PTR(ret);
3190 		lock_page(page);
3191 		if (!PageUptodate(page)) {
3192 			unlock_page(page);
3193 			put_page(page);
3194 			return ERR_PTR(-EIO);
3195 		}
3196 		if (page->mapping != inode->i_mapping) {
3197 			unlock_page(page);
3198 			put_page(page);
3199 			return ERR_PTR(-EAGAIN);
3200 		}
3201 	}
3202 
3203 	return page;
3204 }
3205 
3206 static int gather_extent_pages(struct inode *inode, struct page **pages,
3207 			       int num_pages, u64 off)
3208 {
3209 	int i;
3210 	pgoff_t index = off >> PAGE_SHIFT;
3211 
3212 	for (i = 0; i < num_pages; i++) {
3213 again:
3214 		pages[i] = extent_same_get_page(inode, index + i);
3215 		if (IS_ERR(pages[i])) {
3216 			int err = PTR_ERR(pages[i]);
3217 
3218 			if (err == -EAGAIN)
3219 				goto again;
3220 			pages[i] = NULL;
3221 			return err;
3222 		}
3223 	}
3224 	return 0;
3225 }
3226 
3227 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
3228 			     bool retry_range_locking)
3229 {
3230 	/*
3231 	 * Do any pending delalloc/csum calculations on inode, one way or
3232 	 * another, and lock file content.
3233 	 * The locking order is:
3234 	 *
3235 	 *   1) pages
3236 	 *   2) range in the inode's io tree
3237 	 */
3238 	while (1) {
3239 		struct btrfs_ordered_extent *ordered;
3240 		lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
3241 		ordered = btrfs_lookup_first_ordered_extent(inode,
3242 							    off + len - 1);
3243 		if ((!ordered ||
3244 		     ordered->file_offset + ordered->len <= off ||
3245 		     ordered->file_offset >= off + len) &&
3246 		    !test_range_bit(&BTRFS_I(inode)->io_tree, off,
3247 				    off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
3248 			if (ordered)
3249 				btrfs_put_ordered_extent(ordered);
3250 			break;
3251 		}
3252 		unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
3253 		if (ordered)
3254 			btrfs_put_ordered_extent(ordered);
3255 		if (!retry_range_locking)
3256 			return -EAGAIN;
3257 		btrfs_wait_ordered_range(inode, off, len);
3258 	}
3259 	return 0;
3260 }
3261 
3262 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
3263 {
3264 	inode_unlock(inode1);
3265 	inode_unlock(inode2);
3266 }
3267 
3268 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
3269 {
3270 	if (inode1 < inode2)
3271 		swap(inode1, inode2);
3272 
3273 	inode_lock_nested(inode1, I_MUTEX_PARENT);
3274 	inode_lock_nested(inode2, I_MUTEX_CHILD);
3275 }
3276 
3277 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
3278 				      struct inode *inode2, u64 loff2, u64 len)
3279 {
3280 	unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
3281 	unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
3282 }
3283 
3284 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
3285 				    struct inode *inode2, u64 loff2, u64 len,
3286 				    bool retry_range_locking)
3287 {
3288 	int ret;
3289 
3290 	if (inode1 < inode2) {
3291 		swap(inode1, inode2);
3292 		swap(loff1, loff2);
3293 	}
3294 	ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
3295 	if (ret)
3296 		return ret;
3297 	ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
3298 	if (ret)
3299 		unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
3300 			      loff1 + len - 1);
3301 	return ret;
3302 }
3303 
3304 struct cmp_pages {
3305 	int		num_pages;
3306 	struct page	**src_pages;
3307 	struct page	**dst_pages;
3308 };
3309 
3310 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
3311 {
3312 	int i;
3313 	struct page *pg;
3314 
3315 	for (i = 0; i < cmp->num_pages; i++) {
3316 		pg = cmp->src_pages[i];
3317 		if (pg) {
3318 			unlock_page(pg);
3319 			put_page(pg);
3320 			cmp->src_pages[i] = NULL;
3321 		}
3322 		pg = cmp->dst_pages[i];
3323 		if (pg) {
3324 			unlock_page(pg);
3325 			put_page(pg);
3326 			cmp->dst_pages[i] = NULL;
3327 		}
3328 	}
3329 }
3330 
3331 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
3332 				  struct inode *dst, u64 dst_loff,
3333 				  u64 len, struct cmp_pages *cmp)
3334 {
3335 	int ret;
3336 	int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT;
3337 
3338 	cmp->num_pages = num_pages;
3339 
3340 	ret = gather_extent_pages(src, cmp->src_pages, num_pages, loff);
3341 	if (ret)
3342 		goto out;
3343 
3344 	ret = gather_extent_pages(dst, cmp->dst_pages, num_pages, dst_loff);
3345 
3346 out:
3347 	if (ret)
3348 		btrfs_cmp_data_free(cmp);
3349 	return ret;
3350 }
3351 
3352 static int btrfs_cmp_data(u64 len, struct cmp_pages *cmp)
3353 {
3354 	int ret = 0;
3355 	int i;
3356 	struct page *src_page, *dst_page;
3357 	unsigned int cmp_len = PAGE_SIZE;
3358 	void *addr, *dst_addr;
3359 
3360 	i = 0;
3361 	while (len) {
3362 		if (len < PAGE_SIZE)
3363 			cmp_len = len;
3364 
3365 		BUG_ON(i >= cmp->num_pages);
3366 
3367 		src_page = cmp->src_pages[i];
3368 		dst_page = cmp->dst_pages[i];
3369 		ASSERT(PageLocked(src_page));
3370 		ASSERT(PageLocked(dst_page));
3371 
3372 		addr = kmap_atomic(src_page);
3373 		dst_addr = kmap_atomic(dst_page);
3374 
3375 		flush_dcache_page(src_page);
3376 		flush_dcache_page(dst_page);
3377 
3378 		if (memcmp(addr, dst_addr, cmp_len))
3379 			ret = -EBADE;
3380 
3381 		kunmap_atomic(addr);
3382 		kunmap_atomic(dst_addr);
3383 
3384 		if (ret)
3385 			break;
3386 
3387 		len -= cmp_len;
3388 		i++;
3389 	}
3390 
3391 	return ret;
3392 }
3393 
3394 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3395 				     u64 olen)
3396 {
3397 	u64 len = *plen;
3398 	u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3399 
3400 	if (off + olen > inode->i_size || off + olen < off)
3401 		return -EINVAL;
3402 
3403 	/* if we extend to eof, continue to block boundary */
3404 	if (off + len == inode->i_size)
3405 		*plen = len = ALIGN(inode->i_size, bs) - off;
3406 
3407 	/* Check that we are block aligned - btrfs_clone() requires this */
3408 	if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3409 		return -EINVAL;
3410 
3411 	return 0;
3412 }
3413 
3414 static int btrfs_extent_same_range(struct inode *src, u64 loff, u64 olen,
3415 				   struct inode *dst, u64 dst_loff,
3416 				   struct cmp_pages *cmp)
3417 {
3418 	int ret;
3419 	u64 len = olen;
3420 	bool same_inode = (src == dst);
3421 	u64 same_lock_start = 0;
3422 	u64 same_lock_len = 0;
3423 
3424 	ret = extent_same_check_offsets(src, loff, &len, olen);
3425 	if (ret)
3426 		return ret;
3427 
3428 	ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3429 	if (ret)
3430 		return ret;
3431 
3432 	if (same_inode) {
3433 		/*
3434 		 * Single inode case wants the same checks, except we
3435 		 * don't want our length pushed out past i_size as
3436 		 * comparing that data range makes no sense.
3437 		 *
3438 		 * extent_same_check_offsets() will do this for an
3439 		 * unaligned length at i_size, so catch it here and
3440 		 * reject the request.
3441 		 *
3442 		 * This effectively means we require aligned extents
3443 		 * for the single-inode case, whereas the other cases
3444 		 * allow an unaligned length so long as it ends at
3445 		 * i_size.
3446 		 */
3447 		if (len != olen)
3448 			return -EINVAL;
3449 
3450 		/* Check for overlapping ranges */
3451 		if (dst_loff + len > loff && dst_loff < loff + len)
3452 			return -EINVAL;
3453 
3454 		same_lock_start = min_t(u64, loff, dst_loff);
3455 		same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3456 	}
3457 
3458 again:
3459 	ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, cmp);
3460 	if (ret)
3461 		return ret;
3462 
3463 	if (same_inode)
3464 		ret = lock_extent_range(src, same_lock_start, same_lock_len,
3465 					false);
3466 	else
3467 		ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3468 					       false);
3469 	/*
3470 	 * If one of the inodes has dirty pages in the respective range or
3471 	 * ordered extents, we need to flush dellaloc and wait for all ordered
3472 	 * extents in the range. We must unlock the pages and the ranges in the
3473 	 * io trees to avoid deadlocks when flushing delalloc (requires locking
3474 	 * pages) and when waiting for ordered extents to complete (they require
3475 	 * range locking).
3476 	 */
3477 	if (ret == -EAGAIN) {
3478 		/*
3479 		 * Ranges in the io trees already unlocked. Now unlock all
3480 		 * pages before waiting for all IO to complete.
3481 		 */
3482 		btrfs_cmp_data_free(cmp);
3483 		if (same_inode) {
3484 			btrfs_wait_ordered_range(src, same_lock_start,
3485 						 same_lock_len);
3486 		} else {
3487 			btrfs_wait_ordered_range(src, loff, len);
3488 			btrfs_wait_ordered_range(dst, dst_loff, len);
3489 		}
3490 		goto again;
3491 	}
3492 	ASSERT(ret == 0);
3493 	if (WARN_ON(ret)) {
3494 		/* ranges in the io trees already unlocked */
3495 		btrfs_cmp_data_free(cmp);
3496 		return ret;
3497 	}
3498 
3499 	/* pass original length for comparison so we stay within i_size */
3500 	ret = btrfs_cmp_data(olen, cmp);
3501 	if (ret == 0)
3502 		ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3503 
3504 	if (same_inode)
3505 		unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3506 			      same_lock_start + same_lock_len - 1);
3507 	else
3508 		btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3509 
3510 	btrfs_cmp_data_free(cmp);
3511 
3512 	return ret;
3513 }
3514 
3515 #define BTRFS_MAX_DEDUPE_LEN	SZ_16M
3516 
3517 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3518 			     struct inode *dst, u64 dst_loff)
3519 {
3520 	int ret;
3521 	struct cmp_pages cmp;
3522 	int num_pages = PAGE_ALIGN(BTRFS_MAX_DEDUPE_LEN) >> PAGE_SHIFT;
3523 	bool same_inode = (src == dst);
3524 	u64 i, tail_len, chunk_count;
3525 
3526 	if (olen == 0)
3527 		return 0;
3528 
3529 	if (same_inode)
3530 		inode_lock(src);
3531 	else
3532 		btrfs_double_inode_lock(src, dst);
3533 
3534 	/* don't make the dst file partly checksummed */
3535 	if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3536 	    (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3537 		ret = -EINVAL;
3538 		goto out_unlock;
3539 	}
3540 
3541 	tail_len = olen % BTRFS_MAX_DEDUPE_LEN;
3542 	chunk_count = div_u64(olen, BTRFS_MAX_DEDUPE_LEN);
3543 	if (chunk_count == 0)
3544 		num_pages = PAGE_ALIGN(tail_len) >> PAGE_SHIFT;
3545 
3546 	/*
3547 	 * If deduping ranges in the same inode, locking rules make it
3548 	 * mandatory to always lock pages in ascending order to avoid deadlocks
3549 	 * with concurrent tasks (such as starting writeback/delalloc).
3550 	 */
3551 	if (same_inode && dst_loff < loff)
3552 		swap(loff, dst_loff);
3553 
3554 	/*
3555 	 * We must gather up all the pages before we initiate our extent
3556 	 * locking. We use an array for the page pointers. Size of the array is
3557 	 * bounded by len, which is in turn bounded by BTRFS_MAX_DEDUPE_LEN.
3558 	 */
3559 	cmp.src_pages = kvmalloc_array(num_pages, sizeof(struct page *),
3560 				       GFP_KERNEL | __GFP_ZERO);
3561 	cmp.dst_pages = kvmalloc_array(num_pages, sizeof(struct page *),
3562 				       GFP_KERNEL | __GFP_ZERO);
3563 	if (!cmp.src_pages || !cmp.dst_pages) {
3564 		ret = -ENOMEM;
3565 		goto out_free;
3566 	}
3567 
3568 	for (i = 0; i < chunk_count; i++) {
3569 		ret = btrfs_extent_same_range(src, loff, BTRFS_MAX_DEDUPE_LEN,
3570 					      dst, dst_loff, &cmp);
3571 		if (ret)
3572 			goto out_free;
3573 
3574 		loff += BTRFS_MAX_DEDUPE_LEN;
3575 		dst_loff += BTRFS_MAX_DEDUPE_LEN;
3576 	}
3577 
3578 	if (tail_len > 0)
3579 		ret = btrfs_extent_same_range(src, loff, tail_len, dst,
3580 					      dst_loff, &cmp);
3581 
3582 out_free:
3583 	kvfree(cmp.src_pages);
3584 	kvfree(cmp.dst_pages);
3585 
3586 out_unlock:
3587 	if (same_inode)
3588 		inode_unlock(src);
3589 	else
3590 		btrfs_double_inode_unlock(src, dst);
3591 
3592 	return ret;
3593 }
3594 
3595 int btrfs_dedupe_file_range(struct file *src_file, loff_t src_loff,
3596 			    struct file *dst_file, loff_t dst_loff,
3597 			    u64 olen)
3598 {
3599 	struct inode *src = file_inode(src_file);
3600 	struct inode *dst = file_inode(dst_file);
3601 	u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3602 
3603 	if (WARN_ON_ONCE(bs < PAGE_SIZE)) {
3604 		/*
3605 		 * Btrfs does not support blocksize < page_size. As a
3606 		 * result, btrfs_cmp_data() won't correctly handle
3607 		 * this situation without an update.
3608 		 */
3609 		return -EINVAL;
3610 	}
3611 
3612 	return btrfs_extent_same(src, src_loff, olen, dst, dst_loff);
3613 }
3614 
3615 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3616 				     struct inode *inode,
3617 				     u64 endoff,
3618 				     const u64 destoff,
3619 				     const u64 olen,
3620 				     int no_time_update)
3621 {
3622 	struct btrfs_root *root = BTRFS_I(inode)->root;
3623 	int ret;
3624 
3625 	inode_inc_iversion(inode);
3626 	if (!no_time_update)
3627 		inode->i_mtime = inode->i_ctime = current_time(inode);
3628 	/*
3629 	 * We round up to the block size at eof when determining which
3630 	 * extents to clone above, but shouldn't round up the file size.
3631 	 */
3632 	if (endoff > destoff + olen)
3633 		endoff = destoff + olen;
3634 	if (endoff > inode->i_size)
3635 		btrfs_i_size_write(BTRFS_I(inode), endoff);
3636 
3637 	ret = btrfs_update_inode(trans, root, inode);
3638 	if (ret) {
3639 		btrfs_abort_transaction(trans, ret);
3640 		btrfs_end_transaction(trans);
3641 		goto out;
3642 	}
3643 	ret = btrfs_end_transaction(trans);
3644 out:
3645 	return ret;
3646 }
3647 
3648 static void clone_update_extent_map(struct btrfs_inode *inode,
3649 				    const struct btrfs_trans_handle *trans,
3650 				    const struct btrfs_path *path,
3651 				    const u64 hole_offset,
3652 				    const u64 hole_len)
3653 {
3654 	struct extent_map_tree *em_tree = &inode->extent_tree;
3655 	struct extent_map *em;
3656 	int ret;
3657 
3658 	em = alloc_extent_map();
3659 	if (!em) {
3660 		set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3661 		return;
3662 	}
3663 
3664 	if (path) {
3665 		struct btrfs_file_extent_item *fi;
3666 
3667 		fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3668 				    struct btrfs_file_extent_item);
3669 		btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3670 		em->generation = -1;
3671 		if (btrfs_file_extent_type(path->nodes[0], fi) ==
3672 		    BTRFS_FILE_EXTENT_INLINE)
3673 			set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3674 					&inode->runtime_flags);
3675 	} else {
3676 		em->start = hole_offset;
3677 		em->len = hole_len;
3678 		em->ram_bytes = em->len;
3679 		em->orig_start = hole_offset;
3680 		em->block_start = EXTENT_MAP_HOLE;
3681 		em->block_len = 0;
3682 		em->orig_block_len = 0;
3683 		em->compress_type = BTRFS_COMPRESS_NONE;
3684 		em->generation = trans->transid;
3685 	}
3686 
3687 	while (1) {
3688 		write_lock(&em_tree->lock);
3689 		ret = add_extent_mapping(em_tree, em, 1);
3690 		write_unlock(&em_tree->lock);
3691 		if (ret != -EEXIST) {
3692 			free_extent_map(em);
3693 			break;
3694 		}
3695 		btrfs_drop_extent_cache(inode, em->start,
3696 					em->start + em->len - 1, 0);
3697 	}
3698 
3699 	if (ret)
3700 		set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3701 }
3702 
3703 /*
3704  * Make sure we do not end up inserting an inline extent into a file that has
3705  * already other (non-inline) extents. If a file has an inline extent it can
3706  * not have any other extents and the (single) inline extent must start at the
3707  * file offset 0. Failing to respect these rules will lead to file corruption,
3708  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3709  *
3710  * We can have extents that have been already written to disk or we can have
3711  * dirty ranges still in delalloc, in which case the extent maps and items are
3712  * created only when we run delalloc, and the delalloc ranges might fall outside
3713  * the range we are currently locking in the inode's io tree. So we check the
3714  * inode's i_size because of that (i_size updates are done while holding the
3715  * i_mutex, which we are holding here).
3716  * We also check to see if the inode has a size not greater than "datal" but has
3717  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3718  * protected against such concurrent fallocate calls by the i_mutex).
3719  *
3720  * If the file has no extents but a size greater than datal, do not allow the
3721  * copy because we would need turn the inline extent into a non-inline one (even
3722  * with NO_HOLES enabled). If we find our destination inode only has one inline
3723  * extent, just overwrite it with the source inline extent if its size is less
3724  * than the source extent's size, or we could copy the source inline extent's
3725  * data into the destination inode's inline extent if the later is greater then
3726  * the former.
3727  */
3728 static int clone_copy_inline_extent(struct inode *dst,
3729 				    struct btrfs_trans_handle *trans,
3730 				    struct btrfs_path *path,
3731 				    struct btrfs_key *new_key,
3732 				    const u64 drop_start,
3733 				    const u64 datal,
3734 				    const u64 skip,
3735 				    const u64 size,
3736 				    char *inline_data)
3737 {
3738 	struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb);
3739 	struct btrfs_root *root = BTRFS_I(dst)->root;
3740 	const u64 aligned_end = ALIGN(new_key->offset + datal,
3741 				      fs_info->sectorsize);
3742 	int ret;
3743 	struct btrfs_key key;
3744 
3745 	if (new_key->offset > 0)
3746 		return -EOPNOTSUPP;
3747 
3748 	key.objectid = btrfs_ino(BTRFS_I(dst));
3749 	key.type = BTRFS_EXTENT_DATA_KEY;
3750 	key.offset = 0;
3751 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3752 	if (ret < 0) {
3753 		return ret;
3754 	} else if (ret > 0) {
3755 		if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3756 			ret = btrfs_next_leaf(root, path);
3757 			if (ret < 0)
3758 				return ret;
3759 			else if (ret > 0)
3760 				goto copy_inline_extent;
3761 		}
3762 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3763 		if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3764 		    key.type == BTRFS_EXTENT_DATA_KEY) {
3765 			ASSERT(key.offset > 0);
3766 			return -EOPNOTSUPP;
3767 		}
3768 	} else if (i_size_read(dst) <= datal) {
3769 		struct btrfs_file_extent_item *ei;
3770 		u64 ext_len;
3771 
3772 		/*
3773 		 * If the file size is <= datal, make sure there are no other
3774 		 * extents following (can happen do to an fallocate call with
3775 		 * the flag FALLOC_FL_KEEP_SIZE).
3776 		 */
3777 		ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3778 				    struct btrfs_file_extent_item);
3779 		/*
3780 		 * If it's an inline extent, it can not have other extents
3781 		 * following it.
3782 		 */
3783 		if (btrfs_file_extent_type(path->nodes[0], ei) ==
3784 		    BTRFS_FILE_EXTENT_INLINE)
3785 			goto copy_inline_extent;
3786 
3787 		ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3788 		if (ext_len > aligned_end)
3789 			return -EOPNOTSUPP;
3790 
3791 		ret = btrfs_next_item(root, path);
3792 		if (ret < 0) {
3793 			return ret;
3794 		} else if (ret == 0) {
3795 			btrfs_item_key_to_cpu(path->nodes[0], &key,
3796 					      path->slots[0]);
3797 			if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3798 			    key.type == BTRFS_EXTENT_DATA_KEY)
3799 				return -EOPNOTSUPP;
3800 		}
3801 	}
3802 
3803 copy_inline_extent:
3804 	/*
3805 	 * We have no extent items, or we have an extent at offset 0 which may
3806 	 * or may not be inlined. All these cases are dealt the same way.
3807 	 */
3808 	if (i_size_read(dst) > datal) {
3809 		/*
3810 		 * If the destination inode has an inline extent...
3811 		 * This would require copying the data from the source inline
3812 		 * extent into the beginning of the destination's inline extent.
3813 		 * But this is really complex, both extents can be compressed
3814 		 * or just one of them, which would require decompressing and
3815 		 * re-compressing data (which could increase the new compressed
3816 		 * size, not allowing the compressed data to fit anymore in an
3817 		 * inline extent).
3818 		 * So just don't support this case for now (it should be rare,
3819 		 * we are not really saving space when cloning inline extents).
3820 		 */
3821 		return -EOPNOTSUPP;
3822 	}
3823 
3824 	btrfs_release_path(path);
3825 	ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3826 	if (ret)
3827 		return ret;
3828 	ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3829 	if (ret)
3830 		return ret;
3831 
3832 	if (skip) {
3833 		const u32 start = btrfs_file_extent_calc_inline_size(0);
3834 
3835 		memmove(inline_data + start, inline_data + start + skip, datal);
3836 	}
3837 
3838 	write_extent_buffer(path->nodes[0], inline_data,
3839 			    btrfs_item_ptr_offset(path->nodes[0],
3840 						  path->slots[0]),
3841 			    size);
3842 	inode_add_bytes(dst, datal);
3843 
3844 	return 0;
3845 }
3846 
3847 /**
3848  * btrfs_clone() - clone a range from inode file to another
3849  *
3850  * @src: Inode to clone from
3851  * @inode: Inode to clone to
3852  * @off: Offset within source to start clone from
3853  * @olen: Original length, passed by user, of range to clone
3854  * @olen_aligned: Block-aligned value of olen
3855  * @destoff: Offset within @inode to start clone
3856  * @no_time_update: Whether to update mtime/ctime on the target inode
3857  */
3858 static int btrfs_clone(struct inode *src, struct inode *inode,
3859 		       const u64 off, const u64 olen, const u64 olen_aligned,
3860 		       const u64 destoff, int no_time_update)
3861 {
3862 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3863 	struct btrfs_root *root = BTRFS_I(inode)->root;
3864 	struct btrfs_path *path = NULL;
3865 	struct extent_buffer *leaf;
3866 	struct btrfs_trans_handle *trans;
3867 	char *buf = NULL;
3868 	struct btrfs_key key;
3869 	u32 nritems;
3870 	int slot;
3871 	int ret;
3872 	const u64 len = olen_aligned;
3873 	u64 last_dest_end = destoff;
3874 
3875 	ret = -ENOMEM;
3876 	buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
3877 	if (!buf)
3878 		return ret;
3879 
3880 	path = btrfs_alloc_path();
3881 	if (!path) {
3882 		kvfree(buf);
3883 		return ret;
3884 	}
3885 
3886 	path->reada = READA_FORWARD;
3887 	/* clone data */
3888 	key.objectid = btrfs_ino(BTRFS_I(src));
3889 	key.type = BTRFS_EXTENT_DATA_KEY;
3890 	key.offset = off;
3891 
3892 	while (1) {
3893 		u64 next_key_min_offset = key.offset + 1;
3894 
3895 		/*
3896 		 * note the key will change type as we walk through the
3897 		 * tree.
3898 		 */
3899 		path->leave_spinning = 1;
3900 		ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3901 				0, 0);
3902 		if (ret < 0)
3903 			goto out;
3904 		/*
3905 		 * First search, if no extent item that starts at offset off was
3906 		 * found but the previous item is an extent item, it's possible
3907 		 * it might overlap our target range, therefore process it.
3908 		 */
3909 		if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3910 			btrfs_item_key_to_cpu(path->nodes[0], &key,
3911 					      path->slots[0] - 1);
3912 			if (key.type == BTRFS_EXTENT_DATA_KEY)
3913 				path->slots[0]--;
3914 		}
3915 
3916 		nritems = btrfs_header_nritems(path->nodes[0]);
3917 process_slot:
3918 		if (path->slots[0] >= nritems) {
3919 			ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3920 			if (ret < 0)
3921 				goto out;
3922 			if (ret > 0)
3923 				break;
3924 			nritems = btrfs_header_nritems(path->nodes[0]);
3925 		}
3926 		leaf = path->nodes[0];
3927 		slot = path->slots[0];
3928 
3929 		btrfs_item_key_to_cpu(leaf, &key, slot);
3930 		if (key.type > BTRFS_EXTENT_DATA_KEY ||
3931 		    key.objectid != btrfs_ino(BTRFS_I(src)))
3932 			break;
3933 
3934 		if (key.type == BTRFS_EXTENT_DATA_KEY) {
3935 			struct btrfs_file_extent_item *extent;
3936 			int type;
3937 			u32 size;
3938 			struct btrfs_key new_key;
3939 			u64 disko = 0, diskl = 0;
3940 			u64 datao = 0, datal = 0;
3941 			u8 comp;
3942 			u64 drop_start;
3943 
3944 			extent = btrfs_item_ptr(leaf, slot,
3945 						struct btrfs_file_extent_item);
3946 			comp = btrfs_file_extent_compression(leaf, extent);
3947 			type = btrfs_file_extent_type(leaf, extent);
3948 			if (type == BTRFS_FILE_EXTENT_REG ||
3949 			    type == BTRFS_FILE_EXTENT_PREALLOC) {
3950 				disko = btrfs_file_extent_disk_bytenr(leaf,
3951 								      extent);
3952 				diskl = btrfs_file_extent_disk_num_bytes(leaf,
3953 								 extent);
3954 				datao = btrfs_file_extent_offset(leaf, extent);
3955 				datal = btrfs_file_extent_num_bytes(leaf,
3956 								    extent);
3957 			} else if (type == BTRFS_FILE_EXTENT_INLINE) {
3958 				/* take upper bound, may be compressed */
3959 				datal = btrfs_file_extent_ram_bytes(leaf,
3960 								    extent);
3961 			}
3962 
3963 			/*
3964 			 * The first search might have left us at an extent
3965 			 * item that ends before our target range's start, can
3966 			 * happen if we have holes and NO_HOLES feature enabled.
3967 			 */
3968 			if (key.offset + datal <= off) {
3969 				path->slots[0]++;
3970 				goto process_slot;
3971 			} else if (key.offset >= off + len) {
3972 				break;
3973 			}
3974 			next_key_min_offset = key.offset + datal;
3975 			size = btrfs_item_size_nr(leaf, slot);
3976 			read_extent_buffer(leaf, buf,
3977 					   btrfs_item_ptr_offset(leaf, slot),
3978 					   size);
3979 
3980 			btrfs_release_path(path);
3981 			path->leave_spinning = 0;
3982 
3983 			memcpy(&new_key, &key, sizeof(new_key));
3984 			new_key.objectid = btrfs_ino(BTRFS_I(inode));
3985 			if (off <= key.offset)
3986 				new_key.offset = key.offset + destoff - off;
3987 			else
3988 				new_key.offset = destoff;
3989 
3990 			/*
3991 			 * Deal with a hole that doesn't have an extent item
3992 			 * that represents it (NO_HOLES feature enabled).
3993 			 * This hole is either in the middle of the cloning
3994 			 * range or at the beginning (fully overlaps it or
3995 			 * partially overlaps it).
3996 			 */
3997 			if (new_key.offset != last_dest_end)
3998 				drop_start = last_dest_end;
3999 			else
4000 				drop_start = new_key.offset;
4001 
4002 			/*
4003 			 * 1 - adjusting old extent (we may have to split it)
4004 			 * 1 - add new extent
4005 			 * 1 - inode update
4006 			 */
4007 			trans = btrfs_start_transaction(root, 3);
4008 			if (IS_ERR(trans)) {
4009 				ret = PTR_ERR(trans);
4010 				goto out;
4011 			}
4012 
4013 			if (type == BTRFS_FILE_EXTENT_REG ||
4014 			    type == BTRFS_FILE_EXTENT_PREALLOC) {
4015 				/*
4016 				 *    a  | --- range to clone ---|  b
4017 				 * | ------------- extent ------------- |
4018 				 */
4019 
4020 				/* subtract range b */
4021 				if (key.offset + datal > off + len)
4022 					datal = off + len - key.offset;
4023 
4024 				/* subtract range a */
4025 				if (off > key.offset) {
4026 					datao += off - key.offset;
4027 					datal -= off - key.offset;
4028 				}
4029 
4030 				ret = btrfs_drop_extents(trans, root, inode,
4031 							 drop_start,
4032 							 new_key.offset + datal,
4033 							 1);
4034 				if (ret) {
4035 					if (ret != -EOPNOTSUPP)
4036 						btrfs_abort_transaction(trans,
4037 									ret);
4038 					btrfs_end_transaction(trans);
4039 					goto out;
4040 				}
4041 
4042 				ret = btrfs_insert_empty_item(trans, root, path,
4043 							      &new_key, size);
4044 				if (ret) {
4045 					btrfs_abort_transaction(trans, ret);
4046 					btrfs_end_transaction(trans);
4047 					goto out;
4048 				}
4049 
4050 				leaf = path->nodes[0];
4051 				slot = path->slots[0];
4052 				write_extent_buffer(leaf, buf,
4053 					    btrfs_item_ptr_offset(leaf, slot),
4054 					    size);
4055 
4056 				extent = btrfs_item_ptr(leaf, slot,
4057 						struct btrfs_file_extent_item);
4058 
4059 				/* disko == 0 means it's a hole */
4060 				if (!disko)
4061 					datao = 0;
4062 
4063 				btrfs_set_file_extent_offset(leaf, extent,
4064 							     datao);
4065 				btrfs_set_file_extent_num_bytes(leaf, extent,
4066 								datal);
4067 
4068 				if (disko) {
4069 					inode_add_bytes(inode, datal);
4070 					ret = btrfs_inc_extent_ref(trans,
4071 							root,
4072 							disko, diskl, 0,
4073 							root->root_key.objectid,
4074 							btrfs_ino(BTRFS_I(inode)),
4075 							new_key.offset - datao);
4076 					if (ret) {
4077 						btrfs_abort_transaction(trans,
4078 									ret);
4079 						btrfs_end_transaction(trans);
4080 						goto out;
4081 
4082 					}
4083 				}
4084 			} else if (type == BTRFS_FILE_EXTENT_INLINE) {
4085 				u64 skip = 0;
4086 				u64 trim = 0;
4087 
4088 				if (off > key.offset) {
4089 					skip = off - key.offset;
4090 					new_key.offset += skip;
4091 				}
4092 
4093 				if (key.offset + datal > off + len)
4094 					trim = key.offset + datal - (off + len);
4095 
4096 				if (comp && (skip || trim)) {
4097 					ret = -EINVAL;
4098 					btrfs_end_transaction(trans);
4099 					goto out;
4100 				}
4101 				size -= skip + trim;
4102 				datal -= skip + trim;
4103 
4104 				ret = clone_copy_inline_extent(inode,
4105 							       trans, path,
4106 							       &new_key,
4107 							       drop_start,
4108 							       datal,
4109 							       skip, size, buf);
4110 				if (ret) {
4111 					if (ret != -EOPNOTSUPP)
4112 						btrfs_abort_transaction(trans,
4113 									ret);
4114 					btrfs_end_transaction(trans);
4115 					goto out;
4116 				}
4117 				leaf = path->nodes[0];
4118 				slot = path->slots[0];
4119 			}
4120 
4121 			/* If we have an implicit hole (NO_HOLES feature). */
4122 			if (drop_start < new_key.offset)
4123 				clone_update_extent_map(BTRFS_I(inode), trans,
4124 						NULL, drop_start,
4125 						new_key.offset - drop_start);
4126 
4127 			clone_update_extent_map(BTRFS_I(inode), trans,
4128 					path, 0, 0);
4129 
4130 			btrfs_mark_buffer_dirty(leaf);
4131 			btrfs_release_path(path);
4132 
4133 			last_dest_end = ALIGN(new_key.offset + datal,
4134 					      fs_info->sectorsize);
4135 			ret = clone_finish_inode_update(trans, inode,
4136 							last_dest_end,
4137 							destoff, olen,
4138 							no_time_update);
4139 			if (ret)
4140 				goto out;
4141 			if (new_key.offset + datal >= destoff + len)
4142 				break;
4143 		}
4144 		btrfs_release_path(path);
4145 		key.offset = next_key_min_offset;
4146 
4147 		if (fatal_signal_pending(current)) {
4148 			ret = -EINTR;
4149 			goto out;
4150 		}
4151 	}
4152 	ret = 0;
4153 
4154 	if (last_dest_end < destoff + len) {
4155 		/*
4156 		 * We have an implicit hole (NO_HOLES feature is enabled) that
4157 		 * fully or partially overlaps our cloning range at its end.
4158 		 */
4159 		btrfs_release_path(path);
4160 
4161 		/*
4162 		 * 1 - remove extent(s)
4163 		 * 1 - inode update
4164 		 */
4165 		trans = btrfs_start_transaction(root, 2);
4166 		if (IS_ERR(trans)) {
4167 			ret = PTR_ERR(trans);
4168 			goto out;
4169 		}
4170 		ret = btrfs_drop_extents(trans, root, inode,
4171 					 last_dest_end, destoff + len, 1);
4172 		if (ret) {
4173 			if (ret != -EOPNOTSUPP)
4174 				btrfs_abort_transaction(trans, ret);
4175 			btrfs_end_transaction(trans);
4176 			goto out;
4177 		}
4178 		clone_update_extent_map(BTRFS_I(inode), trans, NULL,
4179 				last_dest_end,
4180 				destoff + len - last_dest_end);
4181 		ret = clone_finish_inode_update(trans, inode, destoff + len,
4182 						destoff, olen, no_time_update);
4183 	}
4184 
4185 out:
4186 	btrfs_free_path(path);
4187 	kvfree(buf);
4188 	return ret;
4189 }
4190 
4191 static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
4192 					u64 off, u64 olen, u64 destoff)
4193 {
4194 	struct inode *inode = file_inode(file);
4195 	struct inode *src = file_inode(file_src);
4196 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4197 	struct btrfs_root *root = BTRFS_I(inode)->root;
4198 	int ret;
4199 	u64 len = olen;
4200 	u64 bs = fs_info->sb->s_blocksize;
4201 	int same_inode = src == inode;
4202 
4203 	/*
4204 	 * TODO:
4205 	 * - split compressed inline extents.  annoying: we need to
4206 	 *   decompress into destination's address_space (the file offset
4207 	 *   may change, so source mapping won't do), then recompress (or
4208 	 *   otherwise reinsert) a subrange.
4209 	 *
4210 	 * - split destination inode's inline extents.  The inline extents can
4211 	 *   be either compressed or non-compressed.
4212 	 */
4213 
4214 	if (btrfs_root_readonly(root))
4215 		return -EROFS;
4216 
4217 	if (file_src->f_path.mnt != file->f_path.mnt ||
4218 	    src->i_sb != inode->i_sb)
4219 		return -EXDEV;
4220 
4221 	if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
4222 		return -EISDIR;
4223 
4224 	if (!same_inode) {
4225 		btrfs_double_inode_lock(src, inode);
4226 	} else {
4227 		inode_lock(src);
4228 	}
4229 
4230 	/* don't make the dst file partly checksummed */
4231 	if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
4232 	    (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
4233 		ret = -EINVAL;
4234 		goto out_unlock;
4235 	}
4236 
4237 	/* determine range to clone */
4238 	ret = -EINVAL;
4239 	if (off + len > src->i_size || off + len < off)
4240 		goto out_unlock;
4241 	if (len == 0)
4242 		olen = len = src->i_size - off;
4243 	/* if we extend to eof, continue to block boundary */
4244 	if (off + len == src->i_size)
4245 		len = ALIGN(src->i_size, bs) - off;
4246 
4247 	if (len == 0) {
4248 		ret = 0;
4249 		goto out_unlock;
4250 	}
4251 
4252 	/* verify the end result is block aligned */
4253 	if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
4254 	    !IS_ALIGNED(destoff, bs))
4255 		goto out_unlock;
4256 
4257 	/* verify if ranges are overlapped within the same file */
4258 	if (same_inode) {
4259 		if (destoff + len > off && destoff < off + len)
4260 			goto out_unlock;
4261 	}
4262 
4263 	if (destoff > inode->i_size) {
4264 		ret = btrfs_cont_expand(inode, inode->i_size, destoff);
4265 		if (ret)
4266 			goto out_unlock;
4267 	}
4268 
4269 	/*
4270 	 * Lock the target range too. Right after we replace the file extent
4271 	 * items in the fs tree (which now point to the cloned data), we might
4272 	 * have a worker replace them with extent items relative to a write
4273 	 * operation that was issued before this clone operation (i.e. confront
4274 	 * with inode.c:btrfs_finish_ordered_io).
4275 	 */
4276 	if (same_inode) {
4277 		u64 lock_start = min_t(u64, off, destoff);
4278 		u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
4279 
4280 		ret = lock_extent_range(src, lock_start, lock_len, true);
4281 	} else {
4282 		ret = btrfs_double_extent_lock(src, off, inode, destoff, len,
4283 					       true);
4284 	}
4285 	ASSERT(ret == 0);
4286 	if (WARN_ON(ret)) {
4287 		/* ranges in the io trees already unlocked */
4288 		goto out_unlock;
4289 	}
4290 
4291 	ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
4292 
4293 	if (same_inode) {
4294 		u64 lock_start = min_t(u64, off, destoff);
4295 		u64 lock_end = max_t(u64, off, destoff) + len - 1;
4296 
4297 		unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
4298 	} else {
4299 		btrfs_double_extent_unlock(src, off, inode, destoff, len);
4300 	}
4301 	/*
4302 	 * Truncate page cache pages so that future reads will see the cloned
4303 	 * data immediately and not the previous data.
4304 	 */
4305 	truncate_inode_pages_range(&inode->i_data,
4306 				round_down(destoff, PAGE_SIZE),
4307 				round_up(destoff + len, PAGE_SIZE) - 1);
4308 out_unlock:
4309 	if (!same_inode)
4310 		btrfs_double_inode_unlock(src, inode);
4311 	else
4312 		inode_unlock(src);
4313 	return ret;
4314 }
4315 
4316 int btrfs_clone_file_range(struct file *src_file, loff_t off,
4317 		struct file *dst_file, loff_t destoff, u64 len)
4318 {
4319 	return btrfs_clone_files(dst_file, src_file, off, len, destoff);
4320 }
4321 
4322 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
4323 {
4324 	struct inode *inode = file_inode(file);
4325 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4326 	struct btrfs_root *root = BTRFS_I(inode)->root;
4327 	struct btrfs_root *new_root;
4328 	struct btrfs_dir_item *di;
4329 	struct btrfs_trans_handle *trans;
4330 	struct btrfs_path *path;
4331 	struct btrfs_key location;
4332 	struct btrfs_disk_key disk_key;
4333 	u64 objectid = 0;
4334 	u64 dir_id;
4335 	int ret;
4336 
4337 	if (!capable(CAP_SYS_ADMIN))
4338 		return -EPERM;
4339 
4340 	ret = mnt_want_write_file(file);
4341 	if (ret)
4342 		return ret;
4343 
4344 	if (copy_from_user(&objectid, argp, sizeof(objectid))) {
4345 		ret = -EFAULT;
4346 		goto out;
4347 	}
4348 
4349 	if (!objectid)
4350 		objectid = BTRFS_FS_TREE_OBJECTID;
4351 
4352 	location.objectid = objectid;
4353 	location.type = BTRFS_ROOT_ITEM_KEY;
4354 	location.offset = (u64)-1;
4355 
4356 	new_root = btrfs_read_fs_root_no_name(fs_info, &location);
4357 	if (IS_ERR(new_root)) {
4358 		ret = PTR_ERR(new_root);
4359 		goto out;
4360 	}
4361 	if (!is_fstree(new_root->objectid)) {
4362 		ret = -ENOENT;
4363 		goto out;
4364 	}
4365 
4366 	path = btrfs_alloc_path();
4367 	if (!path) {
4368 		ret = -ENOMEM;
4369 		goto out;
4370 	}
4371 	path->leave_spinning = 1;
4372 
4373 	trans = btrfs_start_transaction(root, 1);
4374 	if (IS_ERR(trans)) {
4375 		btrfs_free_path(path);
4376 		ret = PTR_ERR(trans);
4377 		goto out;
4378 	}
4379 
4380 	dir_id = btrfs_super_root_dir(fs_info->super_copy);
4381 	di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
4382 				   dir_id, "default", 7, 1);
4383 	if (IS_ERR_OR_NULL(di)) {
4384 		btrfs_free_path(path);
4385 		btrfs_end_transaction(trans);
4386 		btrfs_err(fs_info,
4387 			  "Umm, you don't have the default diritem, this isn't going to work");
4388 		ret = -ENOENT;
4389 		goto out;
4390 	}
4391 
4392 	btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
4393 	btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
4394 	btrfs_mark_buffer_dirty(path->nodes[0]);
4395 	btrfs_free_path(path);
4396 
4397 	btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
4398 	btrfs_end_transaction(trans);
4399 out:
4400 	mnt_drop_write_file(file);
4401 	return ret;
4402 }
4403 
4404 static void get_block_group_info(struct list_head *groups_list,
4405 				 struct btrfs_ioctl_space_info *space)
4406 {
4407 	struct btrfs_block_group_cache *block_group;
4408 
4409 	space->total_bytes = 0;
4410 	space->used_bytes = 0;
4411 	space->flags = 0;
4412 	list_for_each_entry(block_group, groups_list, list) {
4413 		space->flags = block_group->flags;
4414 		space->total_bytes += block_group->key.offset;
4415 		space->used_bytes +=
4416 			btrfs_block_group_used(&block_group->item);
4417 	}
4418 }
4419 
4420 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
4421 				   void __user *arg)
4422 {
4423 	struct btrfs_ioctl_space_args space_args;
4424 	struct btrfs_ioctl_space_info space;
4425 	struct btrfs_ioctl_space_info *dest;
4426 	struct btrfs_ioctl_space_info *dest_orig;
4427 	struct btrfs_ioctl_space_info __user *user_dest;
4428 	struct btrfs_space_info *info;
4429 	static const u64 types[] = {
4430 		BTRFS_BLOCK_GROUP_DATA,
4431 		BTRFS_BLOCK_GROUP_SYSTEM,
4432 		BTRFS_BLOCK_GROUP_METADATA,
4433 		BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
4434 	};
4435 	int num_types = 4;
4436 	int alloc_size;
4437 	int ret = 0;
4438 	u64 slot_count = 0;
4439 	int i, c;
4440 
4441 	if (copy_from_user(&space_args,
4442 			   (struct btrfs_ioctl_space_args __user *)arg,
4443 			   sizeof(space_args)))
4444 		return -EFAULT;
4445 
4446 	for (i = 0; i < num_types; i++) {
4447 		struct btrfs_space_info *tmp;
4448 
4449 		info = NULL;
4450 		rcu_read_lock();
4451 		list_for_each_entry_rcu(tmp, &fs_info->space_info,
4452 					list) {
4453 			if (tmp->flags == types[i]) {
4454 				info = tmp;
4455 				break;
4456 			}
4457 		}
4458 		rcu_read_unlock();
4459 
4460 		if (!info)
4461 			continue;
4462 
4463 		down_read(&info->groups_sem);
4464 		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4465 			if (!list_empty(&info->block_groups[c]))
4466 				slot_count++;
4467 		}
4468 		up_read(&info->groups_sem);
4469 	}
4470 
4471 	/*
4472 	 * Global block reserve, exported as a space_info
4473 	 */
4474 	slot_count++;
4475 
4476 	/* space_slots == 0 means they are asking for a count */
4477 	if (space_args.space_slots == 0) {
4478 		space_args.total_spaces = slot_count;
4479 		goto out;
4480 	}
4481 
4482 	slot_count = min_t(u64, space_args.space_slots, slot_count);
4483 
4484 	alloc_size = sizeof(*dest) * slot_count;
4485 
4486 	/* we generally have at most 6 or so space infos, one for each raid
4487 	 * level.  So, a whole page should be more than enough for everyone
4488 	 */
4489 	if (alloc_size > PAGE_SIZE)
4490 		return -ENOMEM;
4491 
4492 	space_args.total_spaces = 0;
4493 	dest = kmalloc(alloc_size, GFP_KERNEL);
4494 	if (!dest)
4495 		return -ENOMEM;
4496 	dest_orig = dest;
4497 
4498 	/* now we have a buffer to copy into */
4499 	for (i = 0; i < num_types; i++) {
4500 		struct btrfs_space_info *tmp;
4501 
4502 		if (!slot_count)
4503 			break;
4504 
4505 		info = NULL;
4506 		rcu_read_lock();
4507 		list_for_each_entry_rcu(tmp, &fs_info->space_info,
4508 					list) {
4509 			if (tmp->flags == types[i]) {
4510 				info = tmp;
4511 				break;
4512 			}
4513 		}
4514 		rcu_read_unlock();
4515 
4516 		if (!info)
4517 			continue;
4518 		down_read(&info->groups_sem);
4519 		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4520 			if (!list_empty(&info->block_groups[c])) {
4521 				get_block_group_info(&info->block_groups[c],
4522 						     &space);
4523 				memcpy(dest, &space, sizeof(space));
4524 				dest++;
4525 				space_args.total_spaces++;
4526 				slot_count--;
4527 			}
4528 			if (!slot_count)
4529 				break;
4530 		}
4531 		up_read(&info->groups_sem);
4532 	}
4533 
4534 	/*
4535 	 * Add global block reserve
4536 	 */
4537 	if (slot_count) {
4538 		struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4539 
4540 		spin_lock(&block_rsv->lock);
4541 		space.total_bytes = block_rsv->size;
4542 		space.used_bytes = block_rsv->size - block_rsv->reserved;
4543 		spin_unlock(&block_rsv->lock);
4544 		space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4545 		memcpy(dest, &space, sizeof(space));
4546 		space_args.total_spaces++;
4547 	}
4548 
4549 	user_dest = (struct btrfs_ioctl_space_info __user *)
4550 		(arg + sizeof(struct btrfs_ioctl_space_args));
4551 
4552 	if (copy_to_user(user_dest, dest_orig, alloc_size))
4553 		ret = -EFAULT;
4554 
4555 	kfree(dest_orig);
4556 out:
4557 	if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4558 		ret = -EFAULT;
4559 
4560 	return ret;
4561 }
4562 
4563 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4564 					    void __user *argp)
4565 {
4566 	struct btrfs_trans_handle *trans;
4567 	u64 transid;
4568 	int ret;
4569 
4570 	trans = btrfs_attach_transaction_barrier(root);
4571 	if (IS_ERR(trans)) {
4572 		if (PTR_ERR(trans) != -ENOENT)
4573 			return PTR_ERR(trans);
4574 
4575 		/* No running transaction, don't bother */
4576 		transid = root->fs_info->last_trans_committed;
4577 		goto out;
4578 	}
4579 	transid = trans->transid;
4580 	ret = btrfs_commit_transaction_async(trans, 0);
4581 	if (ret) {
4582 		btrfs_end_transaction(trans);
4583 		return ret;
4584 	}
4585 out:
4586 	if (argp)
4587 		if (copy_to_user(argp, &transid, sizeof(transid)))
4588 			return -EFAULT;
4589 	return 0;
4590 }
4591 
4592 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
4593 					   void __user *argp)
4594 {
4595 	u64 transid;
4596 
4597 	if (argp) {
4598 		if (copy_from_user(&transid, argp, sizeof(transid)))
4599 			return -EFAULT;
4600 	} else {
4601 		transid = 0;  /* current trans */
4602 	}
4603 	return btrfs_wait_for_commit(fs_info, transid);
4604 }
4605 
4606 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4607 {
4608 	struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
4609 	struct btrfs_ioctl_scrub_args *sa;
4610 	int ret;
4611 
4612 	if (!capable(CAP_SYS_ADMIN))
4613 		return -EPERM;
4614 
4615 	sa = memdup_user(arg, sizeof(*sa));
4616 	if (IS_ERR(sa))
4617 		return PTR_ERR(sa);
4618 
4619 	if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4620 		ret = mnt_want_write_file(file);
4621 		if (ret)
4622 			goto out;
4623 	}
4624 
4625 	ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
4626 			      &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4627 			      0);
4628 
4629 	if (copy_to_user(arg, sa, sizeof(*sa)))
4630 		ret = -EFAULT;
4631 
4632 	if (!(sa->flags & BTRFS_SCRUB_READONLY))
4633 		mnt_drop_write_file(file);
4634 out:
4635 	kfree(sa);
4636 	return ret;
4637 }
4638 
4639 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
4640 {
4641 	if (!capable(CAP_SYS_ADMIN))
4642 		return -EPERM;
4643 
4644 	return btrfs_scrub_cancel(fs_info);
4645 }
4646 
4647 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
4648 				       void __user *arg)
4649 {
4650 	struct btrfs_ioctl_scrub_args *sa;
4651 	int ret;
4652 
4653 	if (!capable(CAP_SYS_ADMIN))
4654 		return -EPERM;
4655 
4656 	sa = memdup_user(arg, sizeof(*sa));
4657 	if (IS_ERR(sa))
4658 		return PTR_ERR(sa);
4659 
4660 	ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
4661 
4662 	if (copy_to_user(arg, sa, sizeof(*sa)))
4663 		ret = -EFAULT;
4664 
4665 	kfree(sa);
4666 	return ret;
4667 }
4668 
4669 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
4670 				      void __user *arg)
4671 {
4672 	struct btrfs_ioctl_get_dev_stats *sa;
4673 	int ret;
4674 
4675 	sa = memdup_user(arg, sizeof(*sa));
4676 	if (IS_ERR(sa))
4677 		return PTR_ERR(sa);
4678 
4679 	if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4680 		kfree(sa);
4681 		return -EPERM;
4682 	}
4683 
4684 	ret = btrfs_get_dev_stats(fs_info, sa);
4685 
4686 	if (copy_to_user(arg, sa, sizeof(*sa)))
4687 		ret = -EFAULT;
4688 
4689 	kfree(sa);
4690 	return ret;
4691 }
4692 
4693 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
4694 				    void __user *arg)
4695 {
4696 	struct btrfs_ioctl_dev_replace_args *p;
4697 	int ret;
4698 
4699 	if (!capable(CAP_SYS_ADMIN))
4700 		return -EPERM;
4701 
4702 	p = memdup_user(arg, sizeof(*p));
4703 	if (IS_ERR(p))
4704 		return PTR_ERR(p);
4705 
4706 	switch (p->cmd) {
4707 	case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4708 		if (sb_rdonly(fs_info->sb)) {
4709 			ret = -EROFS;
4710 			goto out;
4711 		}
4712 		if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4713 			ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4714 		} else {
4715 			ret = btrfs_dev_replace_by_ioctl(fs_info, p);
4716 			clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4717 		}
4718 		break;
4719 	case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4720 		btrfs_dev_replace_status(fs_info, p);
4721 		ret = 0;
4722 		break;
4723 	case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4724 		p->result = btrfs_dev_replace_cancel(fs_info);
4725 		ret = 0;
4726 		break;
4727 	default:
4728 		ret = -EINVAL;
4729 		break;
4730 	}
4731 
4732 	if (copy_to_user(arg, p, sizeof(*p)))
4733 		ret = -EFAULT;
4734 out:
4735 	kfree(p);
4736 	return ret;
4737 }
4738 
4739 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4740 {
4741 	int ret = 0;
4742 	int i;
4743 	u64 rel_ptr;
4744 	int size;
4745 	struct btrfs_ioctl_ino_path_args *ipa = NULL;
4746 	struct inode_fs_paths *ipath = NULL;
4747 	struct btrfs_path *path;
4748 
4749 	if (!capable(CAP_DAC_READ_SEARCH))
4750 		return -EPERM;
4751 
4752 	path = btrfs_alloc_path();
4753 	if (!path) {
4754 		ret = -ENOMEM;
4755 		goto out;
4756 	}
4757 
4758 	ipa = memdup_user(arg, sizeof(*ipa));
4759 	if (IS_ERR(ipa)) {
4760 		ret = PTR_ERR(ipa);
4761 		ipa = NULL;
4762 		goto out;
4763 	}
4764 
4765 	size = min_t(u32, ipa->size, 4096);
4766 	ipath = init_ipath(size, root, path);
4767 	if (IS_ERR(ipath)) {
4768 		ret = PTR_ERR(ipath);
4769 		ipath = NULL;
4770 		goto out;
4771 	}
4772 
4773 	ret = paths_from_inode(ipa->inum, ipath);
4774 	if (ret < 0)
4775 		goto out;
4776 
4777 	for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4778 		rel_ptr = ipath->fspath->val[i] -
4779 			  (u64)(unsigned long)ipath->fspath->val;
4780 		ipath->fspath->val[i] = rel_ptr;
4781 	}
4782 
4783 	ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
4784 			   ipath->fspath, size);
4785 	if (ret) {
4786 		ret = -EFAULT;
4787 		goto out;
4788 	}
4789 
4790 out:
4791 	btrfs_free_path(path);
4792 	free_ipath(ipath);
4793 	kfree(ipa);
4794 
4795 	return ret;
4796 }
4797 
4798 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4799 {
4800 	struct btrfs_data_container *inodes = ctx;
4801 	const size_t c = 3 * sizeof(u64);
4802 
4803 	if (inodes->bytes_left >= c) {
4804 		inodes->bytes_left -= c;
4805 		inodes->val[inodes->elem_cnt] = inum;
4806 		inodes->val[inodes->elem_cnt + 1] = offset;
4807 		inodes->val[inodes->elem_cnt + 2] = root;
4808 		inodes->elem_cnt += 3;
4809 	} else {
4810 		inodes->bytes_missing += c - inodes->bytes_left;
4811 		inodes->bytes_left = 0;
4812 		inodes->elem_missed += 3;
4813 	}
4814 
4815 	return 0;
4816 }
4817 
4818 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
4819 					void __user *arg, int version)
4820 {
4821 	int ret = 0;
4822 	int size;
4823 	struct btrfs_ioctl_logical_ino_args *loi;
4824 	struct btrfs_data_container *inodes = NULL;
4825 	struct btrfs_path *path = NULL;
4826 	bool ignore_offset;
4827 
4828 	if (!capable(CAP_SYS_ADMIN))
4829 		return -EPERM;
4830 
4831 	loi = memdup_user(arg, sizeof(*loi));
4832 	if (IS_ERR(loi))
4833 		return PTR_ERR(loi);
4834 
4835 	if (version == 1) {
4836 		ignore_offset = false;
4837 		size = min_t(u32, loi->size, SZ_64K);
4838 	} else {
4839 		/* All reserved bits must be 0 for now */
4840 		if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
4841 			ret = -EINVAL;
4842 			goto out_loi;
4843 		}
4844 		/* Only accept flags we have defined so far */
4845 		if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
4846 			ret = -EINVAL;
4847 			goto out_loi;
4848 		}
4849 		ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
4850 		size = min_t(u32, loi->size, SZ_16M);
4851 	}
4852 
4853 	path = btrfs_alloc_path();
4854 	if (!path) {
4855 		ret = -ENOMEM;
4856 		goto out;
4857 	}
4858 
4859 	inodes = init_data_container(size);
4860 	if (IS_ERR(inodes)) {
4861 		ret = PTR_ERR(inodes);
4862 		inodes = NULL;
4863 		goto out;
4864 	}
4865 
4866 	ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
4867 					  build_ino_list, inodes, ignore_offset);
4868 	if (ret == -EINVAL)
4869 		ret = -ENOENT;
4870 	if (ret < 0)
4871 		goto out;
4872 
4873 	ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
4874 			   size);
4875 	if (ret)
4876 		ret = -EFAULT;
4877 
4878 out:
4879 	btrfs_free_path(path);
4880 	kvfree(inodes);
4881 out_loi:
4882 	kfree(loi);
4883 
4884 	return ret;
4885 }
4886 
4887 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
4888 			       struct btrfs_ioctl_balance_args *bargs)
4889 {
4890 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4891 
4892 	bargs->flags = bctl->flags;
4893 
4894 	if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
4895 		bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4896 	if (atomic_read(&fs_info->balance_pause_req))
4897 		bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4898 	if (atomic_read(&fs_info->balance_cancel_req))
4899 		bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4900 
4901 	memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4902 	memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4903 	memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4904 
4905 	spin_lock(&fs_info->balance_lock);
4906 	memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4907 	spin_unlock(&fs_info->balance_lock);
4908 }
4909 
4910 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4911 {
4912 	struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4913 	struct btrfs_fs_info *fs_info = root->fs_info;
4914 	struct btrfs_ioctl_balance_args *bargs;
4915 	struct btrfs_balance_control *bctl;
4916 	bool need_unlock; /* for mut. excl. ops lock */
4917 	int ret;
4918 
4919 	if (!capable(CAP_SYS_ADMIN))
4920 		return -EPERM;
4921 
4922 	ret = mnt_want_write_file(file);
4923 	if (ret)
4924 		return ret;
4925 
4926 again:
4927 	if (!test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4928 		mutex_lock(&fs_info->balance_mutex);
4929 		need_unlock = true;
4930 		goto locked;
4931 	}
4932 
4933 	/*
4934 	 * mut. excl. ops lock is locked.  Three possibilities:
4935 	 *   (1) some other op is running
4936 	 *   (2) balance is running
4937 	 *   (3) balance is paused -- special case (think resume)
4938 	 */
4939 	mutex_lock(&fs_info->balance_mutex);
4940 	if (fs_info->balance_ctl) {
4941 		/* this is either (2) or (3) */
4942 		if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4943 			mutex_unlock(&fs_info->balance_mutex);
4944 			/*
4945 			 * Lock released to allow other waiters to continue,
4946 			 * we'll reexamine the status again.
4947 			 */
4948 			mutex_lock(&fs_info->balance_mutex);
4949 
4950 			if (fs_info->balance_ctl &&
4951 			    !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4952 				/* this is (3) */
4953 				need_unlock = false;
4954 				goto locked;
4955 			}
4956 
4957 			mutex_unlock(&fs_info->balance_mutex);
4958 			goto again;
4959 		} else {
4960 			/* this is (2) */
4961 			mutex_unlock(&fs_info->balance_mutex);
4962 			ret = -EINPROGRESS;
4963 			goto out;
4964 		}
4965 	} else {
4966 		/* this is (1) */
4967 		mutex_unlock(&fs_info->balance_mutex);
4968 		ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4969 		goto out;
4970 	}
4971 
4972 locked:
4973 	BUG_ON(!test_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4974 
4975 	if (arg) {
4976 		bargs = memdup_user(arg, sizeof(*bargs));
4977 		if (IS_ERR(bargs)) {
4978 			ret = PTR_ERR(bargs);
4979 			goto out_unlock;
4980 		}
4981 
4982 		if (bargs->flags & BTRFS_BALANCE_RESUME) {
4983 			if (!fs_info->balance_ctl) {
4984 				ret = -ENOTCONN;
4985 				goto out_bargs;
4986 			}
4987 
4988 			bctl = fs_info->balance_ctl;
4989 			spin_lock(&fs_info->balance_lock);
4990 			bctl->flags |= BTRFS_BALANCE_RESUME;
4991 			spin_unlock(&fs_info->balance_lock);
4992 
4993 			goto do_balance;
4994 		}
4995 	} else {
4996 		bargs = NULL;
4997 	}
4998 
4999 	if (fs_info->balance_ctl) {
5000 		ret = -EINPROGRESS;
5001 		goto out_bargs;
5002 	}
5003 
5004 	bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
5005 	if (!bctl) {
5006 		ret = -ENOMEM;
5007 		goto out_bargs;
5008 	}
5009 
5010 	if (arg) {
5011 		memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
5012 		memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
5013 		memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
5014 
5015 		bctl->flags = bargs->flags;
5016 	} else {
5017 		/* balance everything - no filters */
5018 		bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
5019 	}
5020 
5021 	if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
5022 		ret = -EINVAL;
5023 		goto out_bctl;
5024 	}
5025 
5026 do_balance:
5027 	/*
5028 	 * Ownership of bctl and filesystem flag BTRFS_FS_EXCL_OP goes to
5029 	 * btrfs_balance.  bctl is freed in reset_balance_state, or, if
5030 	 * restriper was paused all the way until unmount, in free_fs_info.
5031 	 * The flag should be cleared after reset_balance_state.
5032 	 */
5033 	need_unlock = false;
5034 
5035 	ret = btrfs_balance(fs_info, bctl, bargs);
5036 	bctl = NULL;
5037 
5038 	if (arg) {
5039 		if (copy_to_user(arg, bargs, sizeof(*bargs)))
5040 			ret = -EFAULT;
5041 	}
5042 
5043 out_bctl:
5044 	kfree(bctl);
5045 out_bargs:
5046 	kfree(bargs);
5047 out_unlock:
5048 	mutex_unlock(&fs_info->balance_mutex);
5049 	if (need_unlock)
5050 		clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
5051 out:
5052 	mnt_drop_write_file(file);
5053 	return ret;
5054 }
5055 
5056 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
5057 {
5058 	if (!capable(CAP_SYS_ADMIN))
5059 		return -EPERM;
5060 
5061 	switch (cmd) {
5062 	case BTRFS_BALANCE_CTL_PAUSE:
5063 		return btrfs_pause_balance(fs_info);
5064 	case BTRFS_BALANCE_CTL_CANCEL:
5065 		return btrfs_cancel_balance(fs_info);
5066 	}
5067 
5068 	return -EINVAL;
5069 }
5070 
5071 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
5072 					 void __user *arg)
5073 {
5074 	struct btrfs_ioctl_balance_args *bargs;
5075 	int ret = 0;
5076 
5077 	if (!capable(CAP_SYS_ADMIN))
5078 		return -EPERM;
5079 
5080 	mutex_lock(&fs_info->balance_mutex);
5081 	if (!fs_info->balance_ctl) {
5082 		ret = -ENOTCONN;
5083 		goto out;
5084 	}
5085 
5086 	bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
5087 	if (!bargs) {
5088 		ret = -ENOMEM;
5089 		goto out;
5090 	}
5091 
5092 	btrfs_update_ioctl_balance_args(fs_info, bargs);
5093 
5094 	if (copy_to_user(arg, bargs, sizeof(*bargs)))
5095 		ret = -EFAULT;
5096 
5097 	kfree(bargs);
5098 out:
5099 	mutex_unlock(&fs_info->balance_mutex);
5100 	return ret;
5101 }
5102 
5103 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
5104 {
5105 	struct inode *inode = file_inode(file);
5106 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5107 	struct btrfs_ioctl_quota_ctl_args *sa;
5108 	int ret;
5109 
5110 	if (!capable(CAP_SYS_ADMIN))
5111 		return -EPERM;
5112 
5113 	ret = mnt_want_write_file(file);
5114 	if (ret)
5115 		return ret;
5116 
5117 	sa = memdup_user(arg, sizeof(*sa));
5118 	if (IS_ERR(sa)) {
5119 		ret = PTR_ERR(sa);
5120 		goto drop_write;
5121 	}
5122 
5123 	down_write(&fs_info->subvol_sem);
5124 
5125 	switch (sa->cmd) {
5126 	case BTRFS_QUOTA_CTL_ENABLE:
5127 		ret = btrfs_quota_enable(fs_info);
5128 		break;
5129 	case BTRFS_QUOTA_CTL_DISABLE:
5130 		ret = btrfs_quota_disable(fs_info);
5131 		break;
5132 	default:
5133 		ret = -EINVAL;
5134 		break;
5135 	}
5136 
5137 	kfree(sa);
5138 	up_write(&fs_info->subvol_sem);
5139 drop_write:
5140 	mnt_drop_write_file(file);
5141 	return ret;
5142 }
5143 
5144 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
5145 {
5146 	struct inode *inode = file_inode(file);
5147 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5148 	struct btrfs_root *root = BTRFS_I(inode)->root;
5149 	struct btrfs_ioctl_qgroup_assign_args *sa;
5150 	struct btrfs_trans_handle *trans;
5151 	int ret;
5152 	int err;
5153 
5154 	if (!capable(CAP_SYS_ADMIN))
5155 		return -EPERM;
5156 
5157 	ret = mnt_want_write_file(file);
5158 	if (ret)
5159 		return ret;
5160 
5161 	sa = memdup_user(arg, sizeof(*sa));
5162 	if (IS_ERR(sa)) {
5163 		ret = PTR_ERR(sa);
5164 		goto drop_write;
5165 	}
5166 
5167 	trans = btrfs_join_transaction(root);
5168 	if (IS_ERR(trans)) {
5169 		ret = PTR_ERR(trans);
5170 		goto out;
5171 	}
5172 
5173 	if (sa->assign) {
5174 		ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
5175 	} else {
5176 		ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
5177 	}
5178 
5179 	/* update qgroup status and info */
5180 	err = btrfs_run_qgroups(trans);
5181 	if (err < 0)
5182 		btrfs_handle_fs_error(fs_info, err,
5183 				      "failed to update qgroup status and info");
5184 	err = btrfs_end_transaction(trans);
5185 	if (err && !ret)
5186 		ret = err;
5187 
5188 out:
5189 	kfree(sa);
5190 drop_write:
5191 	mnt_drop_write_file(file);
5192 	return ret;
5193 }
5194 
5195 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
5196 {
5197 	struct inode *inode = file_inode(file);
5198 	struct btrfs_root *root = BTRFS_I(inode)->root;
5199 	struct btrfs_ioctl_qgroup_create_args *sa;
5200 	struct btrfs_trans_handle *trans;
5201 	int ret;
5202 	int err;
5203 
5204 	if (!capable(CAP_SYS_ADMIN))
5205 		return -EPERM;
5206 
5207 	ret = mnt_want_write_file(file);
5208 	if (ret)
5209 		return ret;
5210 
5211 	sa = memdup_user(arg, sizeof(*sa));
5212 	if (IS_ERR(sa)) {
5213 		ret = PTR_ERR(sa);
5214 		goto drop_write;
5215 	}
5216 
5217 	if (!sa->qgroupid) {
5218 		ret = -EINVAL;
5219 		goto out;
5220 	}
5221 
5222 	trans = btrfs_join_transaction(root);
5223 	if (IS_ERR(trans)) {
5224 		ret = PTR_ERR(trans);
5225 		goto out;
5226 	}
5227 
5228 	if (sa->create) {
5229 		ret = btrfs_create_qgroup(trans, sa->qgroupid);
5230 	} else {
5231 		ret = btrfs_remove_qgroup(trans, sa->qgroupid);
5232 	}
5233 
5234 	err = btrfs_end_transaction(trans);
5235 	if (err && !ret)
5236 		ret = err;
5237 
5238 out:
5239 	kfree(sa);
5240 drop_write:
5241 	mnt_drop_write_file(file);
5242 	return ret;
5243 }
5244 
5245 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
5246 {
5247 	struct inode *inode = file_inode(file);
5248 	struct btrfs_root *root = BTRFS_I(inode)->root;
5249 	struct btrfs_ioctl_qgroup_limit_args *sa;
5250 	struct btrfs_trans_handle *trans;
5251 	int ret;
5252 	int err;
5253 	u64 qgroupid;
5254 
5255 	if (!capable(CAP_SYS_ADMIN))
5256 		return -EPERM;
5257 
5258 	ret = mnt_want_write_file(file);
5259 	if (ret)
5260 		return ret;
5261 
5262 	sa = memdup_user(arg, sizeof(*sa));
5263 	if (IS_ERR(sa)) {
5264 		ret = PTR_ERR(sa);
5265 		goto drop_write;
5266 	}
5267 
5268 	trans = btrfs_join_transaction(root);
5269 	if (IS_ERR(trans)) {
5270 		ret = PTR_ERR(trans);
5271 		goto out;
5272 	}
5273 
5274 	qgroupid = sa->qgroupid;
5275 	if (!qgroupid) {
5276 		/* take the current subvol as qgroup */
5277 		qgroupid = root->root_key.objectid;
5278 	}
5279 
5280 	ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
5281 
5282 	err = btrfs_end_transaction(trans);
5283 	if (err && !ret)
5284 		ret = err;
5285 
5286 out:
5287 	kfree(sa);
5288 drop_write:
5289 	mnt_drop_write_file(file);
5290 	return ret;
5291 }
5292 
5293 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
5294 {
5295 	struct inode *inode = file_inode(file);
5296 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5297 	struct btrfs_ioctl_quota_rescan_args *qsa;
5298 	int ret;
5299 
5300 	if (!capable(CAP_SYS_ADMIN))
5301 		return -EPERM;
5302 
5303 	ret = mnt_want_write_file(file);
5304 	if (ret)
5305 		return ret;
5306 
5307 	qsa = memdup_user(arg, sizeof(*qsa));
5308 	if (IS_ERR(qsa)) {
5309 		ret = PTR_ERR(qsa);
5310 		goto drop_write;
5311 	}
5312 
5313 	if (qsa->flags) {
5314 		ret = -EINVAL;
5315 		goto out;
5316 	}
5317 
5318 	ret = btrfs_qgroup_rescan(fs_info);
5319 
5320 out:
5321 	kfree(qsa);
5322 drop_write:
5323 	mnt_drop_write_file(file);
5324 	return ret;
5325 }
5326 
5327 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
5328 {
5329 	struct inode *inode = file_inode(file);
5330 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5331 	struct btrfs_ioctl_quota_rescan_args *qsa;
5332 	int ret = 0;
5333 
5334 	if (!capable(CAP_SYS_ADMIN))
5335 		return -EPERM;
5336 
5337 	qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
5338 	if (!qsa)
5339 		return -ENOMEM;
5340 
5341 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
5342 		qsa->flags = 1;
5343 		qsa->progress = fs_info->qgroup_rescan_progress.objectid;
5344 	}
5345 
5346 	if (copy_to_user(arg, qsa, sizeof(*qsa)))
5347 		ret = -EFAULT;
5348 
5349 	kfree(qsa);
5350 	return ret;
5351 }
5352 
5353 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
5354 {
5355 	struct inode *inode = file_inode(file);
5356 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5357 
5358 	if (!capable(CAP_SYS_ADMIN))
5359 		return -EPERM;
5360 
5361 	return btrfs_qgroup_wait_for_completion(fs_info, true);
5362 }
5363 
5364 static long _btrfs_ioctl_set_received_subvol(struct file *file,
5365 					    struct btrfs_ioctl_received_subvol_args *sa)
5366 {
5367 	struct inode *inode = file_inode(file);
5368 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5369 	struct btrfs_root *root = BTRFS_I(inode)->root;
5370 	struct btrfs_root_item *root_item = &root->root_item;
5371 	struct btrfs_trans_handle *trans;
5372 	struct timespec64 ct = current_time(inode);
5373 	int ret = 0;
5374 	int received_uuid_changed;
5375 
5376 	if (!inode_owner_or_capable(inode))
5377 		return -EPERM;
5378 
5379 	ret = mnt_want_write_file(file);
5380 	if (ret < 0)
5381 		return ret;
5382 
5383 	down_write(&fs_info->subvol_sem);
5384 
5385 	if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
5386 		ret = -EINVAL;
5387 		goto out;
5388 	}
5389 
5390 	if (btrfs_root_readonly(root)) {
5391 		ret = -EROFS;
5392 		goto out;
5393 	}
5394 
5395 	/*
5396 	 * 1 - root item
5397 	 * 2 - uuid items (received uuid + subvol uuid)
5398 	 */
5399 	trans = btrfs_start_transaction(root, 3);
5400 	if (IS_ERR(trans)) {
5401 		ret = PTR_ERR(trans);
5402 		trans = NULL;
5403 		goto out;
5404 	}
5405 
5406 	sa->rtransid = trans->transid;
5407 	sa->rtime.sec = ct.tv_sec;
5408 	sa->rtime.nsec = ct.tv_nsec;
5409 
5410 	received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5411 				       BTRFS_UUID_SIZE);
5412 	if (received_uuid_changed &&
5413 	    !btrfs_is_empty_uuid(root_item->received_uuid)) {
5414 		ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
5415 					  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5416 					  root->root_key.objectid);
5417 		if (ret && ret != -ENOENT) {
5418 		        btrfs_abort_transaction(trans, ret);
5419 		        btrfs_end_transaction(trans);
5420 		        goto out;
5421 		}
5422 	}
5423 	memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5424 	btrfs_set_root_stransid(root_item, sa->stransid);
5425 	btrfs_set_root_rtransid(root_item, sa->rtransid);
5426 	btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5427 	btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5428 	btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5429 	btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5430 
5431 	ret = btrfs_update_root(trans, fs_info->tree_root,
5432 				&root->root_key, &root->root_item);
5433 	if (ret < 0) {
5434 		btrfs_end_transaction(trans);
5435 		goto out;
5436 	}
5437 	if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5438 		ret = btrfs_uuid_tree_add(trans, sa->uuid,
5439 					  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5440 					  root->root_key.objectid);
5441 		if (ret < 0 && ret != -EEXIST) {
5442 			btrfs_abort_transaction(trans, ret);
5443 			btrfs_end_transaction(trans);
5444 			goto out;
5445 		}
5446 	}
5447 	ret = btrfs_commit_transaction(trans);
5448 out:
5449 	up_write(&fs_info->subvol_sem);
5450 	mnt_drop_write_file(file);
5451 	return ret;
5452 }
5453 
5454 #ifdef CONFIG_64BIT
5455 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5456 						void __user *arg)
5457 {
5458 	struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5459 	struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5460 	int ret = 0;
5461 
5462 	args32 = memdup_user(arg, sizeof(*args32));
5463 	if (IS_ERR(args32))
5464 		return PTR_ERR(args32);
5465 
5466 	args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
5467 	if (!args64) {
5468 		ret = -ENOMEM;
5469 		goto out;
5470 	}
5471 
5472 	memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5473 	args64->stransid = args32->stransid;
5474 	args64->rtransid = args32->rtransid;
5475 	args64->stime.sec = args32->stime.sec;
5476 	args64->stime.nsec = args32->stime.nsec;
5477 	args64->rtime.sec = args32->rtime.sec;
5478 	args64->rtime.nsec = args32->rtime.nsec;
5479 	args64->flags = args32->flags;
5480 
5481 	ret = _btrfs_ioctl_set_received_subvol(file, args64);
5482 	if (ret)
5483 		goto out;
5484 
5485 	memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5486 	args32->stransid = args64->stransid;
5487 	args32->rtransid = args64->rtransid;
5488 	args32->stime.sec = args64->stime.sec;
5489 	args32->stime.nsec = args64->stime.nsec;
5490 	args32->rtime.sec = args64->rtime.sec;
5491 	args32->rtime.nsec = args64->rtime.nsec;
5492 	args32->flags = args64->flags;
5493 
5494 	ret = copy_to_user(arg, args32, sizeof(*args32));
5495 	if (ret)
5496 		ret = -EFAULT;
5497 
5498 out:
5499 	kfree(args32);
5500 	kfree(args64);
5501 	return ret;
5502 }
5503 #endif
5504 
5505 static long btrfs_ioctl_set_received_subvol(struct file *file,
5506 					    void __user *arg)
5507 {
5508 	struct btrfs_ioctl_received_subvol_args *sa = NULL;
5509 	int ret = 0;
5510 
5511 	sa = memdup_user(arg, sizeof(*sa));
5512 	if (IS_ERR(sa))
5513 		return PTR_ERR(sa);
5514 
5515 	ret = _btrfs_ioctl_set_received_subvol(file, sa);
5516 
5517 	if (ret)
5518 		goto out;
5519 
5520 	ret = copy_to_user(arg, sa, sizeof(*sa));
5521 	if (ret)
5522 		ret = -EFAULT;
5523 
5524 out:
5525 	kfree(sa);
5526 	return ret;
5527 }
5528 
5529 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5530 {
5531 	struct inode *inode = file_inode(file);
5532 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5533 	size_t len;
5534 	int ret;
5535 	char label[BTRFS_LABEL_SIZE];
5536 
5537 	spin_lock(&fs_info->super_lock);
5538 	memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5539 	spin_unlock(&fs_info->super_lock);
5540 
5541 	len = strnlen(label, BTRFS_LABEL_SIZE);
5542 
5543 	if (len == BTRFS_LABEL_SIZE) {
5544 		btrfs_warn(fs_info,
5545 			   "label is too long, return the first %zu bytes",
5546 			   --len);
5547 	}
5548 
5549 	ret = copy_to_user(arg, label, len);
5550 
5551 	return ret ? -EFAULT : 0;
5552 }
5553 
5554 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5555 {
5556 	struct inode *inode = file_inode(file);
5557 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5558 	struct btrfs_root *root = BTRFS_I(inode)->root;
5559 	struct btrfs_super_block *super_block = fs_info->super_copy;
5560 	struct btrfs_trans_handle *trans;
5561 	char label[BTRFS_LABEL_SIZE];
5562 	int ret;
5563 
5564 	if (!capable(CAP_SYS_ADMIN))
5565 		return -EPERM;
5566 
5567 	if (copy_from_user(label, arg, sizeof(label)))
5568 		return -EFAULT;
5569 
5570 	if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5571 		btrfs_err(fs_info,
5572 			  "unable to set label with more than %d bytes",
5573 			  BTRFS_LABEL_SIZE - 1);
5574 		return -EINVAL;
5575 	}
5576 
5577 	ret = mnt_want_write_file(file);
5578 	if (ret)
5579 		return ret;
5580 
5581 	trans = btrfs_start_transaction(root, 0);
5582 	if (IS_ERR(trans)) {
5583 		ret = PTR_ERR(trans);
5584 		goto out_unlock;
5585 	}
5586 
5587 	spin_lock(&fs_info->super_lock);
5588 	strcpy(super_block->label, label);
5589 	spin_unlock(&fs_info->super_lock);
5590 	ret = btrfs_commit_transaction(trans);
5591 
5592 out_unlock:
5593 	mnt_drop_write_file(file);
5594 	return ret;
5595 }
5596 
5597 #define INIT_FEATURE_FLAGS(suffix) \
5598 	{ .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5599 	  .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5600 	  .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5601 
5602 int btrfs_ioctl_get_supported_features(void __user *arg)
5603 {
5604 	static const struct btrfs_ioctl_feature_flags features[3] = {
5605 		INIT_FEATURE_FLAGS(SUPP),
5606 		INIT_FEATURE_FLAGS(SAFE_SET),
5607 		INIT_FEATURE_FLAGS(SAFE_CLEAR)
5608 	};
5609 
5610 	if (copy_to_user(arg, &features, sizeof(features)))
5611 		return -EFAULT;
5612 
5613 	return 0;
5614 }
5615 
5616 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5617 {
5618 	struct inode *inode = file_inode(file);
5619 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5620 	struct btrfs_super_block *super_block = fs_info->super_copy;
5621 	struct btrfs_ioctl_feature_flags features;
5622 
5623 	features.compat_flags = btrfs_super_compat_flags(super_block);
5624 	features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5625 	features.incompat_flags = btrfs_super_incompat_flags(super_block);
5626 
5627 	if (copy_to_user(arg, &features, sizeof(features)))
5628 		return -EFAULT;
5629 
5630 	return 0;
5631 }
5632 
5633 static int check_feature_bits(struct btrfs_fs_info *fs_info,
5634 			      enum btrfs_feature_set set,
5635 			      u64 change_mask, u64 flags, u64 supported_flags,
5636 			      u64 safe_set, u64 safe_clear)
5637 {
5638 	const char *type = btrfs_feature_set_names[set];
5639 	char *names;
5640 	u64 disallowed, unsupported;
5641 	u64 set_mask = flags & change_mask;
5642 	u64 clear_mask = ~flags & change_mask;
5643 
5644 	unsupported = set_mask & ~supported_flags;
5645 	if (unsupported) {
5646 		names = btrfs_printable_features(set, unsupported);
5647 		if (names) {
5648 			btrfs_warn(fs_info,
5649 				   "this kernel does not support the %s feature bit%s",
5650 				   names, strchr(names, ',') ? "s" : "");
5651 			kfree(names);
5652 		} else
5653 			btrfs_warn(fs_info,
5654 				   "this kernel does not support %s bits 0x%llx",
5655 				   type, unsupported);
5656 		return -EOPNOTSUPP;
5657 	}
5658 
5659 	disallowed = set_mask & ~safe_set;
5660 	if (disallowed) {
5661 		names = btrfs_printable_features(set, disallowed);
5662 		if (names) {
5663 			btrfs_warn(fs_info,
5664 				   "can't set the %s feature bit%s while mounted",
5665 				   names, strchr(names, ',') ? "s" : "");
5666 			kfree(names);
5667 		} else
5668 			btrfs_warn(fs_info,
5669 				   "can't set %s bits 0x%llx while mounted",
5670 				   type, disallowed);
5671 		return -EPERM;
5672 	}
5673 
5674 	disallowed = clear_mask & ~safe_clear;
5675 	if (disallowed) {
5676 		names = btrfs_printable_features(set, disallowed);
5677 		if (names) {
5678 			btrfs_warn(fs_info,
5679 				   "can't clear the %s feature bit%s while mounted",
5680 				   names, strchr(names, ',') ? "s" : "");
5681 			kfree(names);
5682 		} else
5683 			btrfs_warn(fs_info,
5684 				   "can't clear %s bits 0x%llx while mounted",
5685 				   type, disallowed);
5686 		return -EPERM;
5687 	}
5688 
5689 	return 0;
5690 }
5691 
5692 #define check_feature(fs_info, change_mask, flags, mask_base)	\
5693 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,	\
5694 		   BTRFS_FEATURE_ ## mask_base ## _SUPP,	\
5695 		   BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,	\
5696 		   BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5697 
5698 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5699 {
5700 	struct inode *inode = file_inode(file);
5701 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5702 	struct btrfs_root *root = BTRFS_I(inode)->root;
5703 	struct btrfs_super_block *super_block = fs_info->super_copy;
5704 	struct btrfs_ioctl_feature_flags flags[2];
5705 	struct btrfs_trans_handle *trans;
5706 	u64 newflags;
5707 	int ret;
5708 
5709 	if (!capable(CAP_SYS_ADMIN))
5710 		return -EPERM;
5711 
5712 	if (copy_from_user(flags, arg, sizeof(flags)))
5713 		return -EFAULT;
5714 
5715 	/* Nothing to do */
5716 	if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5717 	    !flags[0].incompat_flags)
5718 		return 0;
5719 
5720 	ret = check_feature(fs_info, flags[0].compat_flags,
5721 			    flags[1].compat_flags, COMPAT);
5722 	if (ret)
5723 		return ret;
5724 
5725 	ret = check_feature(fs_info, flags[0].compat_ro_flags,
5726 			    flags[1].compat_ro_flags, COMPAT_RO);
5727 	if (ret)
5728 		return ret;
5729 
5730 	ret = check_feature(fs_info, flags[0].incompat_flags,
5731 			    flags[1].incompat_flags, INCOMPAT);
5732 	if (ret)
5733 		return ret;
5734 
5735 	ret = mnt_want_write_file(file);
5736 	if (ret)
5737 		return ret;
5738 
5739 	trans = btrfs_start_transaction(root, 0);
5740 	if (IS_ERR(trans)) {
5741 		ret = PTR_ERR(trans);
5742 		goto out_drop_write;
5743 	}
5744 
5745 	spin_lock(&fs_info->super_lock);
5746 	newflags = btrfs_super_compat_flags(super_block);
5747 	newflags |= flags[0].compat_flags & flags[1].compat_flags;
5748 	newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5749 	btrfs_set_super_compat_flags(super_block, newflags);
5750 
5751 	newflags = btrfs_super_compat_ro_flags(super_block);
5752 	newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5753 	newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5754 	btrfs_set_super_compat_ro_flags(super_block, newflags);
5755 
5756 	newflags = btrfs_super_incompat_flags(super_block);
5757 	newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5758 	newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5759 	btrfs_set_super_incompat_flags(super_block, newflags);
5760 	spin_unlock(&fs_info->super_lock);
5761 
5762 	ret = btrfs_commit_transaction(trans);
5763 out_drop_write:
5764 	mnt_drop_write_file(file);
5765 
5766 	return ret;
5767 }
5768 
5769 static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
5770 {
5771 	struct btrfs_ioctl_send_args *arg;
5772 	int ret;
5773 
5774 	if (compat) {
5775 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5776 		struct btrfs_ioctl_send_args_32 args32;
5777 
5778 		ret = copy_from_user(&args32, argp, sizeof(args32));
5779 		if (ret)
5780 			return -EFAULT;
5781 		arg = kzalloc(sizeof(*arg), GFP_KERNEL);
5782 		if (!arg)
5783 			return -ENOMEM;
5784 		arg->send_fd = args32.send_fd;
5785 		arg->clone_sources_count = args32.clone_sources_count;
5786 		arg->clone_sources = compat_ptr(args32.clone_sources);
5787 		arg->parent_root = args32.parent_root;
5788 		arg->flags = args32.flags;
5789 		memcpy(arg->reserved, args32.reserved,
5790 		       sizeof(args32.reserved));
5791 #else
5792 		return -ENOTTY;
5793 #endif
5794 	} else {
5795 		arg = memdup_user(argp, sizeof(*arg));
5796 		if (IS_ERR(arg))
5797 			return PTR_ERR(arg);
5798 	}
5799 	ret = btrfs_ioctl_send(file, arg);
5800 	kfree(arg);
5801 	return ret;
5802 }
5803 
5804 long btrfs_ioctl(struct file *file, unsigned int
5805 		cmd, unsigned long arg)
5806 {
5807 	struct inode *inode = file_inode(file);
5808 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5809 	struct btrfs_root *root = BTRFS_I(inode)->root;
5810 	void __user *argp = (void __user *)arg;
5811 
5812 	switch (cmd) {
5813 	case FS_IOC_GETFLAGS:
5814 		return btrfs_ioctl_getflags(file, argp);
5815 	case FS_IOC_SETFLAGS:
5816 		return btrfs_ioctl_setflags(file, argp);
5817 	case FS_IOC_GETVERSION:
5818 		return btrfs_ioctl_getversion(file, argp);
5819 	case FITRIM:
5820 		return btrfs_ioctl_fitrim(file, argp);
5821 	case BTRFS_IOC_SNAP_CREATE:
5822 		return btrfs_ioctl_snap_create(file, argp, 0);
5823 	case BTRFS_IOC_SNAP_CREATE_V2:
5824 		return btrfs_ioctl_snap_create_v2(file, argp, 0);
5825 	case BTRFS_IOC_SUBVOL_CREATE:
5826 		return btrfs_ioctl_snap_create(file, argp, 1);
5827 	case BTRFS_IOC_SUBVOL_CREATE_V2:
5828 		return btrfs_ioctl_snap_create_v2(file, argp, 1);
5829 	case BTRFS_IOC_SNAP_DESTROY:
5830 		return btrfs_ioctl_snap_destroy(file, argp);
5831 	case BTRFS_IOC_SUBVOL_GETFLAGS:
5832 		return btrfs_ioctl_subvol_getflags(file, argp);
5833 	case BTRFS_IOC_SUBVOL_SETFLAGS:
5834 		return btrfs_ioctl_subvol_setflags(file, argp);
5835 	case BTRFS_IOC_DEFAULT_SUBVOL:
5836 		return btrfs_ioctl_default_subvol(file, argp);
5837 	case BTRFS_IOC_DEFRAG:
5838 		return btrfs_ioctl_defrag(file, NULL);
5839 	case BTRFS_IOC_DEFRAG_RANGE:
5840 		return btrfs_ioctl_defrag(file, argp);
5841 	case BTRFS_IOC_RESIZE:
5842 		return btrfs_ioctl_resize(file, argp);
5843 	case BTRFS_IOC_ADD_DEV:
5844 		return btrfs_ioctl_add_dev(fs_info, argp);
5845 	case BTRFS_IOC_RM_DEV:
5846 		return btrfs_ioctl_rm_dev(file, argp);
5847 	case BTRFS_IOC_RM_DEV_V2:
5848 		return btrfs_ioctl_rm_dev_v2(file, argp);
5849 	case BTRFS_IOC_FS_INFO:
5850 		return btrfs_ioctl_fs_info(fs_info, argp);
5851 	case BTRFS_IOC_DEV_INFO:
5852 		return btrfs_ioctl_dev_info(fs_info, argp);
5853 	case BTRFS_IOC_BALANCE:
5854 		return btrfs_ioctl_balance(file, NULL);
5855 	case BTRFS_IOC_TREE_SEARCH:
5856 		return btrfs_ioctl_tree_search(file, argp);
5857 	case BTRFS_IOC_TREE_SEARCH_V2:
5858 		return btrfs_ioctl_tree_search_v2(file, argp);
5859 	case BTRFS_IOC_INO_LOOKUP:
5860 		return btrfs_ioctl_ino_lookup(file, argp);
5861 	case BTRFS_IOC_INO_PATHS:
5862 		return btrfs_ioctl_ino_to_path(root, argp);
5863 	case BTRFS_IOC_LOGICAL_INO:
5864 		return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
5865 	case BTRFS_IOC_LOGICAL_INO_V2:
5866 		return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
5867 	case BTRFS_IOC_SPACE_INFO:
5868 		return btrfs_ioctl_space_info(fs_info, argp);
5869 	case BTRFS_IOC_SYNC: {
5870 		int ret;
5871 
5872 		ret = btrfs_start_delalloc_roots(fs_info, -1);
5873 		if (ret)
5874 			return ret;
5875 		ret = btrfs_sync_fs(inode->i_sb, 1);
5876 		/*
5877 		 * The transaction thread may want to do more work,
5878 		 * namely it pokes the cleaner kthread that will start
5879 		 * processing uncleaned subvols.
5880 		 */
5881 		wake_up_process(fs_info->transaction_kthread);
5882 		return ret;
5883 	}
5884 	case BTRFS_IOC_START_SYNC:
5885 		return btrfs_ioctl_start_sync(root, argp);
5886 	case BTRFS_IOC_WAIT_SYNC:
5887 		return btrfs_ioctl_wait_sync(fs_info, argp);
5888 	case BTRFS_IOC_SCRUB:
5889 		return btrfs_ioctl_scrub(file, argp);
5890 	case BTRFS_IOC_SCRUB_CANCEL:
5891 		return btrfs_ioctl_scrub_cancel(fs_info);
5892 	case BTRFS_IOC_SCRUB_PROGRESS:
5893 		return btrfs_ioctl_scrub_progress(fs_info, argp);
5894 	case BTRFS_IOC_BALANCE_V2:
5895 		return btrfs_ioctl_balance(file, argp);
5896 	case BTRFS_IOC_BALANCE_CTL:
5897 		return btrfs_ioctl_balance_ctl(fs_info, arg);
5898 	case BTRFS_IOC_BALANCE_PROGRESS:
5899 		return btrfs_ioctl_balance_progress(fs_info, argp);
5900 	case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5901 		return btrfs_ioctl_set_received_subvol(file, argp);
5902 #ifdef CONFIG_64BIT
5903 	case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5904 		return btrfs_ioctl_set_received_subvol_32(file, argp);
5905 #endif
5906 	case BTRFS_IOC_SEND:
5907 		return _btrfs_ioctl_send(file, argp, false);
5908 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5909 	case BTRFS_IOC_SEND_32:
5910 		return _btrfs_ioctl_send(file, argp, true);
5911 #endif
5912 	case BTRFS_IOC_GET_DEV_STATS:
5913 		return btrfs_ioctl_get_dev_stats(fs_info, argp);
5914 	case BTRFS_IOC_QUOTA_CTL:
5915 		return btrfs_ioctl_quota_ctl(file, argp);
5916 	case BTRFS_IOC_QGROUP_ASSIGN:
5917 		return btrfs_ioctl_qgroup_assign(file, argp);
5918 	case BTRFS_IOC_QGROUP_CREATE:
5919 		return btrfs_ioctl_qgroup_create(file, argp);
5920 	case BTRFS_IOC_QGROUP_LIMIT:
5921 		return btrfs_ioctl_qgroup_limit(file, argp);
5922 	case BTRFS_IOC_QUOTA_RESCAN:
5923 		return btrfs_ioctl_quota_rescan(file, argp);
5924 	case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5925 		return btrfs_ioctl_quota_rescan_status(file, argp);
5926 	case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5927 		return btrfs_ioctl_quota_rescan_wait(file, argp);
5928 	case BTRFS_IOC_DEV_REPLACE:
5929 		return btrfs_ioctl_dev_replace(fs_info, argp);
5930 	case BTRFS_IOC_GET_FSLABEL:
5931 		return btrfs_ioctl_get_fslabel(file, argp);
5932 	case BTRFS_IOC_SET_FSLABEL:
5933 		return btrfs_ioctl_set_fslabel(file, argp);
5934 	case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5935 		return btrfs_ioctl_get_supported_features(argp);
5936 	case BTRFS_IOC_GET_FEATURES:
5937 		return btrfs_ioctl_get_features(file, argp);
5938 	case BTRFS_IOC_SET_FEATURES:
5939 		return btrfs_ioctl_set_features(file, argp);
5940 	case FS_IOC_FSGETXATTR:
5941 		return btrfs_ioctl_fsgetxattr(file, argp);
5942 	case FS_IOC_FSSETXATTR:
5943 		return btrfs_ioctl_fssetxattr(file, argp);
5944 	case BTRFS_IOC_GET_SUBVOL_INFO:
5945 		return btrfs_ioctl_get_subvol_info(file, argp);
5946 	case BTRFS_IOC_GET_SUBVOL_ROOTREF:
5947 		return btrfs_ioctl_get_subvol_rootref(file, argp);
5948 	case BTRFS_IOC_INO_LOOKUP_USER:
5949 		return btrfs_ioctl_ino_lookup_user(file, argp);
5950 	}
5951 
5952 	return -ENOTTY;
5953 }
5954 
5955 #ifdef CONFIG_COMPAT
5956 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5957 {
5958 	/*
5959 	 * These all access 32-bit values anyway so no further
5960 	 * handling is necessary.
5961 	 */
5962 	switch (cmd) {
5963 	case FS_IOC32_GETFLAGS:
5964 		cmd = FS_IOC_GETFLAGS;
5965 		break;
5966 	case FS_IOC32_SETFLAGS:
5967 		cmd = FS_IOC_SETFLAGS;
5968 		break;
5969 	case FS_IOC32_GETVERSION:
5970 		cmd = FS_IOC_GETVERSION;
5971 		break;
5972 	}
5973 
5974 	return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5975 }
5976 #endif
5977