xref: /openbmc/linux/fs/btrfs/ioctl.c (revision fed8b7e366e7c8f81e957ef91aa8f0a38e038c66)
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 	int ret;
495 
496 	if (!capable(CAP_SYS_ADMIN))
497 		return -EPERM;
498 
499 	rcu_read_lock();
500 	list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
501 				dev_list) {
502 		if (!device->bdev)
503 			continue;
504 		q = bdev_get_queue(device->bdev);
505 		if (blk_queue_discard(q)) {
506 			num_devices++;
507 			minlen = min_t(u64, q->limits.discard_granularity,
508 				     minlen);
509 		}
510 	}
511 	rcu_read_unlock();
512 
513 	if (!num_devices)
514 		return -EOPNOTSUPP;
515 	if (copy_from_user(&range, arg, sizeof(range)))
516 		return -EFAULT;
517 
518 	/*
519 	 * NOTE: Don't truncate the range using super->total_bytes.  Bytenr of
520 	 * block group is in the logical address space, which can be any
521 	 * sectorsize aligned bytenr in  the range [0, U64_MAX].
522 	 */
523 	if (range.len < fs_info->sb->s_blocksize)
524 		return -EINVAL;
525 
526 	range.minlen = max(range.minlen, minlen);
527 	ret = btrfs_trim_fs(fs_info, &range);
528 	if (ret < 0)
529 		return ret;
530 
531 	if (copy_to_user(arg, &range, sizeof(range)))
532 		return -EFAULT;
533 
534 	return 0;
535 }
536 
537 int btrfs_is_empty_uuid(u8 *uuid)
538 {
539 	int i;
540 
541 	for (i = 0; i < BTRFS_UUID_SIZE; i++) {
542 		if (uuid[i])
543 			return 0;
544 	}
545 	return 1;
546 }
547 
548 static noinline int create_subvol(struct inode *dir,
549 				  struct dentry *dentry,
550 				  const char *name, int namelen,
551 				  u64 *async_transid,
552 				  struct btrfs_qgroup_inherit *inherit)
553 {
554 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
555 	struct btrfs_trans_handle *trans;
556 	struct btrfs_key key;
557 	struct btrfs_root_item *root_item;
558 	struct btrfs_inode_item *inode_item;
559 	struct extent_buffer *leaf;
560 	struct btrfs_root *root = BTRFS_I(dir)->root;
561 	struct btrfs_root *new_root;
562 	struct btrfs_block_rsv block_rsv;
563 	struct timespec64 cur_time = current_time(dir);
564 	struct inode *inode;
565 	int ret;
566 	int err;
567 	u64 objectid;
568 	u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
569 	u64 index = 0;
570 	uuid_le new_uuid;
571 
572 	root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
573 	if (!root_item)
574 		return -ENOMEM;
575 
576 	ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid);
577 	if (ret)
578 		goto fail_free;
579 
580 	/*
581 	 * Don't create subvolume whose level is not zero. Or qgroup will be
582 	 * screwed up since it assumes subvolume qgroup's level to be 0.
583 	 */
584 	if (btrfs_qgroup_level(objectid)) {
585 		ret = -ENOSPC;
586 		goto fail_free;
587 	}
588 
589 	btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
590 	/*
591 	 * The same as the snapshot creation, please see the comment
592 	 * of create_snapshot().
593 	 */
594 	ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 8, false);
595 	if (ret)
596 		goto fail_free;
597 
598 	trans = btrfs_start_transaction(root, 0);
599 	if (IS_ERR(trans)) {
600 		ret = PTR_ERR(trans);
601 		btrfs_subvolume_release_metadata(fs_info, &block_rsv);
602 		goto fail_free;
603 	}
604 	trans->block_rsv = &block_rsv;
605 	trans->bytes_reserved = block_rsv.size;
606 
607 	ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
608 	if (ret)
609 		goto fail;
610 
611 	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
612 	if (IS_ERR(leaf)) {
613 		ret = PTR_ERR(leaf);
614 		goto fail;
615 	}
616 
617 	btrfs_mark_buffer_dirty(leaf);
618 
619 	inode_item = &root_item->inode;
620 	btrfs_set_stack_inode_generation(inode_item, 1);
621 	btrfs_set_stack_inode_size(inode_item, 3);
622 	btrfs_set_stack_inode_nlink(inode_item, 1);
623 	btrfs_set_stack_inode_nbytes(inode_item,
624 				     fs_info->nodesize);
625 	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
626 
627 	btrfs_set_root_flags(root_item, 0);
628 	btrfs_set_root_limit(root_item, 0);
629 	btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
630 
631 	btrfs_set_root_bytenr(root_item, leaf->start);
632 	btrfs_set_root_generation(root_item, trans->transid);
633 	btrfs_set_root_level(root_item, 0);
634 	btrfs_set_root_refs(root_item, 1);
635 	btrfs_set_root_used(root_item, leaf->len);
636 	btrfs_set_root_last_snapshot(root_item, 0);
637 
638 	btrfs_set_root_generation_v2(root_item,
639 			btrfs_root_generation(root_item));
640 	uuid_le_gen(&new_uuid);
641 	memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
642 	btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
643 	btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
644 	root_item->ctime = root_item->otime;
645 	btrfs_set_root_ctransid(root_item, trans->transid);
646 	btrfs_set_root_otransid(root_item, trans->transid);
647 
648 	btrfs_tree_unlock(leaf);
649 	free_extent_buffer(leaf);
650 	leaf = NULL;
651 
652 	btrfs_set_root_dirid(root_item, new_dirid);
653 
654 	key.objectid = objectid;
655 	key.offset = 0;
656 	key.type = BTRFS_ROOT_ITEM_KEY;
657 	ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
658 				root_item);
659 	if (ret)
660 		goto fail;
661 
662 	key.offset = (u64)-1;
663 	new_root = btrfs_read_fs_root_no_name(fs_info, &key);
664 	if (IS_ERR(new_root)) {
665 		ret = PTR_ERR(new_root);
666 		btrfs_abort_transaction(trans, ret);
667 		goto fail;
668 	}
669 
670 	btrfs_record_root_in_trans(trans, new_root);
671 
672 	ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
673 	if (ret) {
674 		/* We potentially lose an unused inode item here */
675 		btrfs_abort_transaction(trans, ret);
676 		goto fail;
677 	}
678 
679 	mutex_lock(&new_root->objectid_mutex);
680 	new_root->highest_objectid = new_dirid;
681 	mutex_unlock(&new_root->objectid_mutex);
682 
683 	/*
684 	 * insert the directory item
685 	 */
686 	ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
687 	if (ret) {
688 		btrfs_abort_transaction(trans, ret);
689 		goto fail;
690 	}
691 
692 	ret = btrfs_insert_dir_item(trans, name, namelen, BTRFS_I(dir), &key,
693 				    BTRFS_FT_DIR, index);
694 	if (ret) {
695 		btrfs_abort_transaction(trans, ret);
696 		goto fail;
697 	}
698 
699 	btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
700 	ret = btrfs_update_inode(trans, root, dir);
701 	BUG_ON(ret);
702 
703 	ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid,
704 				 btrfs_ino(BTRFS_I(dir)), index, name, namelen);
705 	BUG_ON(ret);
706 
707 	ret = btrfs_uuid_tree_add(trans, root_item->uuid,
708 				  BTRFS_UUID_KEY_SUBVOL, objectid);
709 	if (ret)
710 		btrfs_abort_transaction(trans, ret);
711 
712 fail:
713 	kfree(root_item);
714 	trans->block_rsv = NULL;
715 	trans->bytes_reserved = 0;
716 	btrfs_subvolume_release_metadata(fs_info, &block_rsv);
717 
718 	if (async_transid) {
719 		*async_transid = trans->transid;
720 		err = btrfs_commit_transaction_async(trans, 1);
721 		if (err)
722 			err = btrfs_commit_transaction(trans);
723 	} else {
724 		err = btrfs_commit_transaction(trans);
725 	}
726 	if (err && !ret)
727 		ret = err;
728 
729 	if (!ret) {
730 		inode = btrfs_lookup_dentry(dir, dentry);
731 		if (IS_ERR(inode))
732 			return PTR_ERR(inode);
733 		d_instantiate(dentry, inode);
734 	}
735 	return ret;
736 
737 fail_free:
738 	kfree(root_item);
739 	return ret;
740 }
741 
742 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
743 			   struct dentry *dentry,
744 			   u64 *async_transid, bool readonly,
745 			   struct btrfs_qgroup_inherit *inherit)
746 {
747 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
748 	struct inode *inode;
749 	struct btrfs_pending_snapshot *pending_snapshot;
750 	struct btrfs_trans_handle *trans;
751 	int ret;
752 	bool snapshot_force_cow = false;
753 
754 	if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
755 		return -EINVAL;
756 
757 	pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
758 	if (!pending_snapshot)
759 		return -ENOMEM;
760 
761 	pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
762 			GFP_KERNEL);
763 	pending_snapshot->path = btrfs_alloc_path();
764 	if (!pending_snapshot->root_item || !pending_snapshot->path) {
765 		ret = -ENOMEM;
766 		goto free_pending;
767 	}
768 
769 	/*
770 	 * Force new buffered writes to reserve space even when NOCOW is
771 	 * possible. This is to avoid later writeback (running dealloc) to
772 	 * fallback to COW mode and unexpectedly fail with ENOSPC.
773 	 */
774 	atomic_inc(&root->will_be_snapshotted);
775 	smp_mb__after_atomic();
776 	/* wait for no snapshot writes */
777 	wait_event(root->subv_writers->wait,
778 		   percpu_counter_sum(&root->subv_writers->counter) == 0);
779 
780 	ret = btrfs_start_delalloc_inodes(root);
781 	if (ret)
782 		goto dec_and_free;
783 
784 	/*
785 	 * All previous writes have started writeback in NOCOW mode, so now
786 	 * we force future writes to fallback to COW mode during snapshot
787 	 * creation.
788 	 */
789 	atomic_inc(&root->snapshot_force_cow);
790 	snapshot_force_cow = true;
791 
792 	btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
793 
794 	btrfs_init_block_rsv(&pending_snapshot->block_rsv,
795 			     BTRFS_BLOCK_RSV_TEMP);
796 	/*
797 	 * 1 - parent dir inode
798 	 * 2 - dir entries
799 	 * 1 - root item
800 	 * 2 - root ref/backref
801 	 * 1 - root of snapshot
802 	 * 1 - UUID item
803 	 */
804 	ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
805 					&pending_snapshot->block_rsv, 8,
806 					false);
807 	if (ret)
808 		goto dec_and_free;
809 
810 	pending_snapshot->dentry = dentry;
811 	pending_snapshot->root = root;
812 	pending_snapshot->readonly = readonly;
813 	pending_snapshot->dir = dir;
814 	pending_snapshot->inherit = inherit;
815 
816 	trans = btrfs_start_transaction(root, 0);
817 	if (IS_ERR(trans)) {
818 		ret = PTR_ERR(trans);
819 		goto fail;
820 	}
821 
822 	spin_lock(&fs_info->trans_lock);
823 	list_add(&pending_snapshot->list,
824 		 &trans->transaction->pending_snapshots);
825 	spin_unlock(&fs_info->trans_lock);
826 	if (async_transid) {
827 		*async_transid = trans->transid;
828 		ret = btrfs_commit_transaction_async(trans, 1);
829 		if (ret)
830 			ret = btrfs_commit_transaction(trans);
831 	} else {
832 		ret = btrfs_commit_transaction(trans);
833 	}
834 	if (ret)
835 		goto fail;
836 
837 	ret = pending_snapshot->error;
838 	if (ret)
839 		goto fail;
840 
841 	ret = btrfs_orphan_cleanup(pending_snapshot->snap);
842 	if (ret)
843 		goto fail;
844 
845 	inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
846 	if (IS_ERR(inode)) {
847 		ret = PTR_ERR(inode);
848 		goto fail;
849 	}
850 
851 	d_instantiate(dentry, inode);
852 	ret = 0;
853 fail:
854 	btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
855 dec_and_free:
856 	if (snapshot_force_cow)
857 		atomic_dec(&root->snapshot_force_cow);
858 	if (atomic_dec_and_test(&root->will_be_snapshotted))
859 		wake_up_var(&root->will_be_snapshotted);
860 free_pending:
861 	kfree(pending_snapshot->root_item);
862 	btrfs_free_path(pending_snapshot->path);
863 	kfree(pending_snapshot);
864 
865 	return ret;
866 }
867 
868 /*  copy of may_delete in fs/namei.c()
869  *	Check whether we can remove a link victim from directory dir, check
870  *  whether the type of victim is right.
871  *  1. We can't do it if dir is read-only (done in permission())
872  *  2. We should have write and exec permissions on dir
873  *  3. We can't remove anything from append-only dir
874  *  4. We can't do anything with immutable dir (done in permission())
875  *  5. If the sticky bit on dir is set we should either
876  *	a. be owner of dir, or
877  *	b. be owner of victim, or
878  *	c. have CAP_FOWNER capability
879  *  6. If the victim is append-only or immutable we can't do anything with
880  *     links pointing to it.
881  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
882  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
883  *  9. We can't remove a root or mountpoint.
884  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
885  *     nfs_async_unlink().
886  */
887 
888 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
889 {
890 	int error;
891 
892 	if (d_really_is_negative(victim))
893 		return -ENOENT;
894 
895 	BUG_ON(d_inode(victim->d_parent) != dir);
896 	audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
897 
898 	error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
899 	if (error)
900 		return error;
901 	if (IS_APPEND(dir))
902 		return -EPERM;
903 	if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
904 	    IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
905 		return -EPERM;
906 	if (isdir) {
907 		if (!d_is_dir(victim))
908 			return -ENOTDIR;
909 		if (IS_ROOT(victim))
910 			return -EBUSY;
911 	} else if (d_is_dir(victim))
912 		return -EISDIR;
913 	if (IS_DEADDIR(dir))
914 		return -ENOENT;
915 	if (victim->d_flags & DCACHE_NFSFS_RENAMED)
916 		return -EBUSY;
917 	return 0;
918 }
919 
920 /* copy of may_create in fs/namei.c() */
921 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
922 {
923 	if (d_really_is_positive(child))
924 		return -EEXIST;
925 	if (IS_DEADDIR(dir))
926 		return -ENOENT;
927 	return inode_permission(dir, MAY_WRITE | MAY_EXEC);
928 }
929 
930 /*
931  * Create a new subvolume below @parent.  This is largely modeled after
932  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
933  * inside this filesystem so it's quite a bit simpler.
934  */
935 static noinline int btrfs_mksubvol(const struct path *parent,
936 				   const char *name, int namelen,
937 				   struct btrfs_root *snap_src,
938 				   u64 *async_transid, bool readonly,
939 				   struct btrfs_qgroup_inherit *inherit)
940 {
941 	struct inode *dir = d_inode(parent->dentry);
942 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
943 	struct dentry *dentry;
944 	int error;
945 
946 	error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
947 	if (error == -EINTR)
948 		return error;
949 
950 	dentry = lookup_one_len(name, parent->dentry, namelen);
951 	error = PTR_ERR(dentry);
952 	if (IS_ERR(dentry))
953 		goto out_unlock;
954 
955 	error = btrfs_may_create(dir, dentry);
956 	if (error)
957 		goto out_dput;
958 
959 	/*
960 	 * even if this name doesn't exist, we may get hash collisions.
961 	 * check for them now when we can safely fail
962 	 */
963 	error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
964 					       dir->i_ino, name,
965 					       namelen);
966 	if (error)
967 		goto out_dput;
968 
969 	down_read(&fs_info->subvol_sem);
970 
971 	if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
972 		goto out_up_read;
973 
974 	if (snap_src) {
975 		error = create_snapshot(snap_src, dir, dentry,
976 					async_transid, readonly, inherit);
977 	} else {
978 		error = create_subvol(dir, dentry, name, namelen,
979 				      async_transid, inherit);
980 	}
981 	if (!error)
982 		fsnotify_mkdir(dir, dentry);
983 out_up_read:
984 	up_read(&fs_info->subvol_sem);
985 out_dput:
986 	dput(dentry);
987 out_unlock:
988 	inode_unlock(dir);
989 	return error;
990 }
991 
992 /*
993  * When we're defragging a range, we don't want to kick it off again
994  * if it is really just waiting for delalloc to send it down.
995  * If we find a nice big extent or delalloc range for the bytes in the
996  * file you want to defrag, we return 0 to let you know to skip this
997  * part of the file
998  */
999 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
1000 {
1001 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1002 	struct extent_map *em = NULL;
1003 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1004 	u64 end;
1005 
1006 	read_lock(&em_tree->lock);
1007 	em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
1008 	read_unlock(&em_tree->lock);
1009 
1010 	if (em) {
1011 		end = extent_map_end(em);
1012 		free_extent_map(em);
1013 		if (end - offset > thresh)
1014 			return 0;
1015 	}
1016 	/* if we already have a nice delalloc here, just stop */
1017 	thresh /= 2;
1018 	end = count_range_bits(io_tree, &offset, offset + thresh,
1019 			       thresh, EXTENT_DELALLOC, 1);
1020 	if (end >= thresh)
1021 		return 0;
1022 	return 1;
1023 }
1024 
1025 /*
1026  * helper function to walk through a file and find extents
1027  * newer than a specific transid, and smaller than thresh.
1028  *
1029  * This is used by the defragging code to find new and small
1030  * extents
1031  */
1032 static int find_new_extents(struct btrfs_root *root,
1033 			    struct inode *inode, u64 newer_than,
1034 			    u64 *off, u32 thresh)
1035 {
1036 	struct btrfs_path *path;
1037 	struct btrfs_key min_key;
1038 	struct extent_buffer *leaf;
1039 	struct btrfs_file_extent_item *extent;
1040 	int type;
1041 	int ret;
1042 	u64 ino = btrfs_ino(BTRFS_I(inode));
1043 
1044 	path = btrfs_alloc_path();
1045 	if (!path)
1046 		return -ENOMEM;
1047 
1048 	min_key.objectid = ino;
1049 	min_key.type = BTRFS_EXTENT_DATA_KEY;
1050 	min_key.offset = *off;
1051 
1052 	while (1) {
1053 		ret = btrfs_search_forward(root, &min_key, path, newer_than);
1054 		if (ret != 0)
1055 			goto none;
1056 process_slot:
1057 		if (min_key.objectid != ino)
1058 			goto none;
1059 		if (min_key.type != BTRFS_EXTENT_DATA_KEY)
1060 			goto none;
1061 
1062 		leaf = path->nodes[0];
1063 		extent = btrfs_item_ptr(leaf, path->slots[0],
1064 					struct btrfs_file_extent_item);
1065 
1066 		type = btrfs_file_extent_type(leaf, extent);
1067 		if (type == BTRFS_FILE_EXTENT_REG &&
1068 		    btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
1069 		    check_defrag_in_cache(inode, min_key.offset, thresh)) {
1070 			*off = min_key.offset;
1071 			btrfs_free_path(path);
1072 			return 0;
1073 		}
1074 
1075 		path->slots[0]++;
1076 		if (path->slots[0] < btrfs_header_nritems(leaf)) {
1077 			btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
1078 			goto process_slot;
1079 		}
1080 
1081 		if (min_key.offset == (u64)-1)
1082 			goto none;
1083 
1084 		min_key.offset++;
1085 		btrfs_release_path(path);
1086 	}
1087 none:
1088 	btrfs_free_path(path);
1089 	return -ENOENT;
1090 }
1091 
1092 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1093 {
1094 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1095 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1096 	struct extent_map *em;
1097 	u64 len = PAGE_SIZE;
1098 
1099 	/*
1100 	 * hopefully we have this extent in the tree already, try without
1101 	 * the full extent lock
1102 	 */
1103 	read_lock(&em_tree->lock);
1104 	em = lookup_extent_mapping(em_tree, start, len);
1105 	read_unlock(&em_tree->lock);
1106 
1107 	if (!em) {
1108 		struct extent_state *cached = NULL;
1109 		u64 end = start + len - 1;
1110 
1111 		/* get the big lock and read metadata off disk */
1112 		lock_extent_bits(io_tree, start, end, &cached);
1113 		em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
1114 		unlock_extent_cached(io_tree, start, end, &cached);
1115 
1116 		if (IS_ERR(em))
1117 			return NULL;
1118 	}
1119 
1120 	return em;
1121 }
1122 
1123 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1124 {
1125 	struct extent_map *next;
1126 	bool ret = true;
1127 
1128 	/* this is the last extent */
1129 	if (em->start + em->len >= i_size_read(inode))
1130 		return false;
1131 
1132 	next = defrag_lookup_extent(inode, em->start + em->len);
1133 	if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1134 		ret = false;
1135 	else if ((em->block_start + em->block_len == next->block_start) &&
1136 		 (em->block_len > SZ_128K && next->block_len > SZ_128K))
1137 		ret = false;
1138 
1139 	free_extent_map(next);
1140 	return ret;
1141 }
1142 
1143 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1144 			       u64 *last_len, u64 *skip, u64 *defrag_end,
1145 			       int compress)
1146 {
1147 	struct extent_map *em;
1148 	int ret = 1;
1149 	bool next_mergeable = true;
1150 	bool prev_mergeable = true;
1151 
1152 	/*
1153 	 * make sure that once we start defragging an extent, we keep on
1154 	 * defragging it
1155 	 */
1156 	if (start < *defrag_end)
1157 		return 1;
1158 
1159 	*skip = 0;
1160 
1161 	em = defrag_lookup_extent(inode, start);
1162 	if (!em)
1163 		return 0;
1164 
1165 	/* this will cover holes, and inline extents */
1166 	if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1167 		ret = 0;
1168 		goto out;
1169 	}
1170 
1171 	if (!*defrag_end)
1172 		prev_mergeable = false;
1173 
1174 	next_mergeable = defrag_check_next_extent(inode, em);
1175 	/*
1176 	 * we hit a real extent, if it is big or the next extent is not a
1177 	 * real extent, don't bother defragging it
1178 	 */
1179 	if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1180 	    (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1181 		ret = 0;
1182 out:
1183 	/*
1184 	 * last_len ends up being a counter of how many bytes we've defragged.
1185 	 * every time we choose not to defrag an extent, we reset *last_len
1186 	 * so that the next tiny extent will force a defrag.
1187 	 *
1188 	 * The end result of this is that tiny extents before a single big
1189 	 * extent will force at least part of that big extent to be defragged.
1190 	 */
1191 	if (ret) {
1192 		*defrag_end = extent_map_end(em);
1193 	} else {
1194 		*last_len = 0;
1195 		*skip = extent_map_end(em);
1196 		*defrag_end = 0;
1197 	}
1198 
1199 	free_extent_map(em);
1200 	return ret;
1201 }
1202 
1203 /*
1204  * it doesn't do much good to defrag one or two pages
1205  * at a time.  This pulls in a nice chunk of pages
1206  * to COW and defrag.
1207  *
1208  * It also makes sure the delalloc code has enough
1209  * dirty data to avoid making new small extents as part
1210  * of the defrag
1211  *
1212  * It's a good idea to start RA on this range
1213  * before calling this.
1214  */
1215 static int cluster_pages_for_defrag(struct inode *inode,
1216 				    struct page **pages,
1217 				    unsigned long start_index,
1218 				    unsigned long num_pages)
1219 {
1220 	unsigned long file_end;
1221 	u64 isize = i_size_read(inode);
1222 	u64 page_start;
1223 	u64 page_end;
1224 	u64 page_cnt;
1225 	int ret;
1226 	int i;
1227 	int i_done;
1228 	struct btrfs_ordered_extent *ordered;
1229 	struct extent_state *cached_state = NULL;
1230 	struct extent_io_tree *tree;
1231 	struct extent_changeset *data_reserved = NULL;
1232 	gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1233 
1234 	file_end = (isize - 1) >> PAGE_SHIFT;
1235 	if (!isize || start_index > file_end)
1236 		return 0;
1237 
1238 	page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1239 
1240 	ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
1241 			start_index << PAGE_SHIFT,
1242 			page_cnt << PAGE_SHIFT);
1243 	if (ret)
1244 		return ret;
1245 	i_done = 0;
1246 	tree = &BTRFS_I(inode)->io_tree;
1247 
1248 	/* step one, lock all the pages */
1249 	for (i = 0; i < page_cnt; i++) {
1250 		struct page *page;
1251 again:
1252 		page = find_or_create_page(inode->i_mapping,
1253 					   start_index + i, mask);
1254 		if (!page)
1255 			break;
1256 
1257 		page_start = page_offset(page);
1258 		page_end = page_start + PAGE_SIZE - 1;
1259 		while (1) {
1260 			lock_extent_bits(tree, page_start, page_end,
1261 					 &cached_state);
1262 			ordered = btrfs_lookup_ordered_extent(inode,
1263 							      page_start);
1264 			unlock_extent_cached(tree, page_start, page_end,
1265 					     &cached_state);
1266 			if (!ordered)
1267 				break;
1268 
1269 			unlock_page(page);
1270 			btrfs_start_ordered_extent(inode, ordered, 1);
1271 			btrfs_put_ordered_extent(ordered);
1272 			lock_page(page);
1273 			/*
1274 			 * we unlocked the page above, so we need check if
1275 			 * it was released or not.
1276 			 */
1277 			if (page->mapping != inode->i_mapping) {
1278 				unlock_page(page);
1279 				put_page(page);
1280 				goto again;
1281 			}
1282 		}
1283 
1284 		if (!PageUptodate(page)) {
1285 			btrfs_readpage(NULL, page);
1286 			lock_page(page);
1287 			if (!PageUptodate(page)) {
1288 				unlock_page(page);
1289 				put_page(page);
1290 				ret = -EIO;
1291 				break;
1292 			}
1293 		}
1294 
1295 		if (page->mapping != inode->i_mapping) {
1296 			unlock_page(page);
1297 			put_page(page);
1298 			goto again;
1299 		}
1300 
1301 		pages[i] = page;
1302 		i_done++;
1303 	}
1304 	if (!i_done || ret)
1305 		goto out;
1306 
1307 	if (!(inode->i_sb->s_flags & SB_ACTIVE))
1308 		goto out;
1309 
1310 	/*
1311 	 * so now we have a nice long stream of locked
1312 	 * and up to date pages, lets wait on them
1313 	 */
1314 	for (i = 0; i < i_done; i++)
1315 		wait_on_page_writeback(pages[i]);
1316 
1317 	page_start = page_offset(pages[0]);
1318 	page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1319 
1320 	lock_extent_bits(&BTRFS_I(inode)->io_tree,
1321 			 page_start, page_end - 1, &cached_state);
1322 	clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1323 			  page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1324 			  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1325 			  &cached_state);
1326 
1327 	if (i_done != page_cnt) {
1328 		spin_lock(&BTRFS_I(inode)->lock);
1329 		btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1330 		spin_unlock(&BTRFS_I(inode)->lock);
1331 		btrfs_delalloc_release_space(inode, data_reserved,
1332 				start_index << PAGE_SHIFT,
1333 				(page_cnt - i_done) << PAGE_SHIFT, true);
1334 	}
1335 
1336 
1337 	set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1338 			  &cached_state);
1339 
1340 	unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1341 			     page_start, page_end - 1, &cached_state);
1342 
1343 	for (i = 0; i < i_done; i++) {
1344 		clear_page_dirty_for_io(pages[i]);
1345 		ClearPageChecked(pages[i]);
1346 		set_page_extent_mapped(pages[i]);
1347 		set_page_dirty(pages[i]);
1348 		unlock_page(pages[i]);
1349 		put_page(pages[i]);
1350 	}
1351 	btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT,
1352 				       false);
1353 	extent_changeset_free(data_reserved);
1354 	return i_done;
1355 out:
1356 	for (i = 0; i < i_done; i++) {
1357 		unlock_page(pages[i]);
1358 		put_page(pages[i]);
1359 	}
1360 	btrfs_delalloc_release_space(inode, data_reserved,
1361 			start_index << PAGE_SHIFT,
1362 			page_cnt << PAGE_SHIFT, true);
1363 	btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT,
1364 				       true);
1365 	extent_changeset_free(data_reserved);
1366 	return ret;
1367 
1368 }
1369 
1370 int btrfs_defrag_file(struct inode *inode, struct file *file,
1371 		      struct btrfs_ioctl_defrag_range_args *range,
1372 		      u64 newer_than, unsigned long max_to_defrag)
1373 {
1374 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1375 	struct btrfs_root *root = BTRFS_I(inode)->root;
1376 	struct file_ra_state *ra = NULL;
1377 	unsigned long last_index;
1378 	u64 isize = i_size_read(inode);
1379 	u64 last_len = 0;
1380 	u64 skip = 0;
1381 	u64 defrag_end = 0;
1382 	u64 newer_off = range->start;
1383 	unsigned long i;
1384 	unsigned long ra_index = 0;
1385 	int ret;
1386 	int defrag_count = 0;
1387 	int compress_type = BTRFS_COMPRESS_ZLIB;
1388 	u32 extent_thresh = range->extent_thresh;
1389 	unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1390 	unsigned long cluster = max_cluster;
1391 	u64 new_align = ~((u64)SZ_128K - 1);
1392 	struct page **pages = NULL;
1393 	bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1394 
1395 	if (isize == 0)
1396 		return 0;
1397 
1398 	if (range->start >= isize)
1399 		return -EINVAL;
1400 
1401 	if (do_compress) {
1402 		if (range->compress_type > BTRFS_COMPRESS_TYPES)
1403 			return -EINVAL;
1404 		if (range->compress_type)
1405 			compress_type = range->compress_type;
1406 	}
1407 
1408 	if (extent_thresh == 0)
1409 		extent_thresh = SZ_256K;
1410 
1411 	/*
1412 	 * If we were not given a file, allocate a readahead context. As
1413 	 * readahead is just an optimization, defrag will work without it so
1414 	 * we don't error out.
1415 	 */
1416 	if (!file) {
1417 		ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1418 		if (ra)
1419 			file_ra_state_init(ra, inode->i_mapping);
1420 	} else {
1421 		ra = &file->f_ra;
1422 	}
1423 
1424 	pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1425 	if (!pages) {
1426 		ret = -ENOMEM;
1427 		goto out_ra;
1428 	}
1429 
1430 	/* find the last page to defrag */
1431 	if (range->start + range->len > range->start) {
1432 		last_index = min_t(u64, isize - 1,
1433 			 range->start + range->len - 1) >> PAGE_SHIFT;
1434 	} else {
1435 		last_index = (isize - 1) >> PAGE_SHIFT;
1436 	}
1437 
1438 	if (newer_than) {
1439 		ret = find_new_extents(root, inode, newer_than,
1440 				       &newer_off, SZ_64K);
1441 		if (!ret) {
1442 			range->start = newer_off;
1443 			/*
1444 			 * we always align our defrag to help keep
1445 			 * the extents in the file evenly spaced
1446 			 */
1447 			i = (newer_off & new_align) >> PAGE_SHIFT;
1448 		} else
1449 			goto out_ra;
1450 	} else {
1451 		i = range->start >> PAGE_SHIFT;
1452 	}
1453 	if (!max_to_defrag)
1454 		max_to_defrag = last_index - i + 1;
1455 
1456 	/*
1457 	 * make writeback starts from i, so the defrag range can be
1458 	 * written sequentially.
1459 	 */
1460 	if (i < inode->i_mapping->writeback_index)
1461 		inode->i_mapping->writeback_index = i;
1462 
1463 	while (i <= last_index && defrag_count < max_to_defrag &&
1464 	       (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1465 		/*
1466 		 * make sure we stop running if someone unmounts
1467 		 * the FS
1468 		 */
1469 		if (!(inode->i_sb->s_flags & SB_ACTIVE))
1470 			break;
1471 
1472 		if (btrfs_defrag_cancelled(fs_info)) {
1473 			btrfs_debug(fs_info, "defrag_file cancelled");
1474 			ret = -EAGAIN;
1475 			break;
1476 		}
1477 
1478 		if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1479 					 extent_thresh, &last_len, &skip,
1480 					 &defrag_end, do_compress)){
1481 			unsigned long next;
1482 			/*
1483 			 * the should_defrag function tells us how much to skip
1484 			 * bump our counter by the suggested amount
1485 			 */
1486 			next = DIV_ROUND_UP(skip, PAGE_SIZE);
1487 			i = max(i + 1, next);
1488 			continue;
1489 		}
1490 
1491 		if (!newer_than) {
1492 			cluster = (PAGE_ALIGN(defrag_end) >>
1493 				   PAGE_SHIFT) - i;
1494 			cluster = min(cluster, max_cluster);
1495 		} else {
1496 			cluster = max_cluster;
1497 		}
1498 
1499 		if (i + cluster > ra_index) {
1500 			ra_index = max(i, ra_index);
1501 			if (ra)
1502 				page_cache_sync_readahead(inode->i_mapping, ra,
1503 						file, ra_index, cluster);
1504 			ra_index += cluster;
1505 		}
1506 
1507 		inode_lock(inode);
1508 		if (do_compress)
1509 			BTRFS_I(inode)->defrag_compress = compress_type;
1510 		ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1511 		if (ret < 0) {
1512 			inode_unlock(inode);
1513 			goto out_ra;
1514 		}
1515 
1516 		defrag_count += ret;
1517 		balance_dirty_pages_ratelimited(inode->i_mapping);
1518 		inode_unlock(inode);
1519 
1520 		if (newer_than) {
1521 			if (newer_off == (u64)-1)
1522 				break;
1523 
1524 			if (ret > 0)
1525 				i += ret;
1526 
1527 			newer_off = max(newer_off + 1,
1528 					(u64)i << PAGE_SHIFT);
1529 
1530 			ret = find_new_extents(root, inode, newer_than,
1531 					       &newer_off, SZ_64K);
1532 			if (!ret) {
1533 				range->start = newer_off;
1534 				i = (newer_off & new_align) >> PAGE_SHIFT;
1535 			} else {
1536 				break;
1537 			}
1538 		} else {
1539 			if (ret > 0) {
1540 				i += ret;
1541 				last_len += ret << PAGE_SHIFT;
1542 			} else {
1543 				i++;
1544 				last_len = 0;
1545 			}
1546 		}
1547 	}
1548 
1549 	if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1550 		filemap_flush(inode->i_mapping);
1551 		if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1552 			     &BTRFS_I(inode)->runtime_flags))
1553 			filemap_flush(inode->i_mapping);
1554 	}
1555 
1556 	if (range->compress_type == BTRFS_COMPRESS_LZO) {
1557 		btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1558 	} else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1559 		btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1560 	}
1561 
1562 	ret = defrag_count;
1563 
1564 out_ra:
1565 	if (do_compress) {
1566 		inode_lock(inode);
1567 		BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1568 		inode_unlock(inode);
1569 	}
1570 	if (!file)
1571 		kfree(ra);
1572 	kfree(pages);
1573 	return ret;
1574 }
1575 
1576 static noinline int btrfs_ioctl_resize(struct file *file,
1577 					void __user *arg)
1578 {
1579 	struct inode *inode = file_inode(file);
1580 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1581 	u64 new_size;
1582 	u64 old_size;
1583 	u64 devid = 1;
1584 	struct btrfs_root *root = BTRFS_I(inode)->root;
1585 	struct btrfs_ioctl_vol_args *vol_args;
1586 	struct btrfs_trans_handle *trans;
1587 	struct btrfs_device *device = NULL;
1588 	char *sizestr;
1589 	char *retptr;
1590 	char *devstr = NULL;
1591 	int ret = 0;
1592 	int mod = 0;
1593 
1594 	if (!capable(CAP_SYS_ADMIN))
1595 		return -EPERM;
1596 
1597 	ret = mnt_want_write_file(file);
1598 	if (ret)
1599 		return ret;
1600 
1601 	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
1602 		mnt_drop_write_file(file);
1603 		return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1604 	}
1605 
1606 	vol_args = memdup_user(arg, sizeof(*vol_args));
1607 	if (IS_ERR(vol_args)) {
1608 		ret = PTR_ERR(vol_args);
1609 		goto out;
1610 	}
1611 
1612 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1613 
1614 	sizestr = vol_args->name;
1615 	devstr = strchr(sizestr, ':');
1616 	if (devstr) {
1617 		sizestr = devstr + 1;
1618 		*devstr = '\0';
1619 		devstr = vol_args->name;
1620 		ret = kstrtoull(devstr, 10, &devid);
1621 		if (ret)
1622 			goto out_free;
1623 		if (!devid) {
1624 			ret = -EINVAL;
1625 			goto out_free;
1626 		}
1627 		btrfs_info(fs_info, "resizing devid %llu", devid);
1628 	}
1629 
1630 	device = btrfs_find_device(fs_info, devid, NULL, NULL);
1631 	if (!device) {
1632 		btrfs_info(fs_info, "resizer unable to find device %llu",
1633 			   devid);
1634 		ret = -ENODEV;
1635 		goto out_free;
1636 	}
1637 
1638 	if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1639 		btrfs_info(fs_info,
1640 			   "resizer unable to apply on readonly device %llu",
1641 		       devid);
1642 		ret = -EPERM;
1643 		goto out_free;
1644 	}
1645 
1646 	if (!strcmp(sizestr, "max"))
1647 		new_size = device->bdev->bd_inode->i_size;
1648 	else {
1649 		if (sizestr[0] == '-') {
1650 			mod = -1;
1651 			sizestr++;
1652 		} else if (sizestr[0] == '+') {
1653 			mod = 1;
1654 			sizestr++;
1655 		}
1656 		new_size = memparse(sizestr, &retptr);
1657 		if (*retptr != '\0' || new_size == 0) {
1658 			ret = -EINVAL;
1659 			goto out_free;
1660 		}
1661 	}
1662 
1663 	if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1664 		ret = -EPERM;
1665 		goto out_free;
1666 	}
1667 
1668 	old_size = btrfs_device_get_total_bytes(device);
1669 
1670 	if (mod < 0) {
1671 		if (new_size > old_size) {
1672 			ret = -EINVAL;
1673 			goto out_free;
1674 		}
1675 		new_size = old_size - new_size;
1676 	} else if (mod > 0) {
1677 		if (new_size > ULLONG_MAX - old_size) {
1678 			ret = -ERANGE;
1679 			goto out_free;
1680 		}
1681 		new_size = old_size + new_size;
1682 	}
1683 
1684 	if (new_size < SZ_256M) {
1685 		ret = -EINVAL;
1686 		goto out_free;
1687 	}
1688 	if (new_size > device->bdev->bd_inode->i_size) {
1689 		ret = -EFBIG;
1690 		goto out_free;
1691 	}
1692 
1693 	new_size = round_down(new_size, fs_info->sectorsize);
1694 
1695 	btrfs_info_in_rcu(fs_info, "new size for %s is %llu",
1696 			  rcu_str_deref(device->name), new_size);
1697 
1698 	if (new_size > old_size) {
1699 		trans = btrfs_start_transaction(root, 0);
1700 		if (IS_ERR(trans)) {
1701 			ret = PTR_ERR(trans);
1702 			goto out_free;
1703 		}
1704 		ret = btrfs_grow_device(trans, device, new_size);
1705 		btrfs_commit_transaction(trans);
1706 	} else if (new_size < old_size) {
1707 		ret = btrfs_shrink_device(device, new_size);
1708 	} /* equal, nothing need to do */
1709 
1710 out_free:
1711 	kfree(vol_args);
1712 out:
1713 	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
1714 	mnt_drop_write_file(file);
1715 	return ret;
1716 }
1717 
1718 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1719 				const char *name, unsigned long fd, int subvol,
1720 				u64 *transid, bool readonly,
1721 				struct btrfs_qgroup_inherit *inherit)
1722 {
1723 	int namelen;
1724 	int ret = 0;
1725 
1726 	if (!S_ISDIR(file_inode(file)->i_mode))
1727 		return -ENOTDIR;
1728 
1729 	ret = mnt_want_write_file(file);
1730 	if (ret)
1731 		goto out;
1732 
1733 	namelen = strlen(name);
1734 	if (strchr(name, '/')) {
1735 		ret = -EINVAL;
1736 		goto out_drop_write;
1737 	}
1738 
1739 	if (name[0] == '.' &&
1740 	   (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1741 		ret = -EEXIST;
1742 		goto out_drop_write;
1743 	}
1744 
1745 	if (subvol) {
1746 		ret = btrfs_mksubvol(&file->f_path, name, namelen,
1747 				     NULL, transid, readonly, inherit);
1748 	} else {
1749 		struct fd src = fdget(fd);
1750 		struct inode *src_inode;
1751 		if (!src.file) {
1752 			ret = -EINVAL;
1753 			goto out_drop_write;
1754 		}
1755 
1756 		src_inode = file_inode(src.file);
1757 		if (src_inode->i_sb != file_inode(file)->i_sb) {
1758 			btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1759 				   "Snapshot src from another FS");
1760 			ret = -EXDEV;
1761 		} else if (!inode_owner_or_capable(src_inode)) {
1762 			/*
1763 			 * Subvolume creation is not restricted, but snapshots
1764 			 * are limited to own subvolumes only
1765 			 */
1766 			ret = -EPERM;
1767 		} else {
1768 			ret = btrfs_mksubvol(&file->f_path, name, namelen,
1769 					     BTRFS_I(src_inode)->root,
1770 					     transid, readonly, inherit);
1771 		}
1772 		fdput(src);
1773 	}
1774 out_drop_write:
1775 	mnt_drop_write_file(file);
1776 out:
1777 	return ret;
1778 }
1779 
1780 static noinline int btrfs_ioctl_snap_create(struct file *file,
1781 					    void __user *arg, int subvol)
1782 {
1783 	struct btrfs_ioctl_vol_args *vol_args;
1784 	int ret;
1785 
1786 	if (!S_ISDIR(file_inode(file)->i_mode))
1787 		return -ENOTDIR;
1788 
1789 	vol_args = memdup_user(arg, sizeof(*vol_args));
1790 	if (IS_ERR(vol_args))
1791 		return PTR_ERR(vol_args);
1792 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1793 
1794 	ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1795 					      vol_args->fd, subvol,
1796 					      NULL, false, NULL);
1797 
1798 	kfree(vol_args);
1799 	return ret;
1800 }
1801 
1802 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1803 					       void __user *arg, int subvol)
1804 {
1805 	struct btrfs_ioctl_vol_args_v2 *vol_args;
1806 	int ret;
1807 	u64 transid = 0;
1808 	u64 *ptr = NULL;
1809 	bool readonly = false;
1810 	struct btrfs_qgroup_inherit *inherit = NULL;
1811 
1812 	if (!S_ISDIR(file_inode(file)->i_mode))
1813 		return -ENOTDIR;
1814 
1815 	vol_args = memdup_user(arg, sizeof(*vol_args));
1816 	if (IS_ERR(vol_args))
1817 		return PTR_ERR(vol_args);
1818 	vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1819 
1820 	if (vol_args->flags &
1821 	    ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1822 	      BTRFS_SUBVOL_QGROUP_INHERIT)) {
1823 		ret = -EOPNOTSUPP;
1824 		goto free_args;
1825 	}
1826 
1827 	if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1828 		ptr = &transid;
1829 	if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1830 		readonly = true;
1831 	if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1832 		if (vol_args->size > PAGE_SIZE) {
1833 			ret = -EINVAL;
1834 			goto free_args;
1835 		}
1836 		inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1837 		if (IS_ERR(inherit)) {
1838 			ret = PTR_ERR(inherit);
1839 			goto free_args;
1840 		}
1841 	}
1842 
1843 	ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1844 					      vol_args->fd, subvol, ptr,
1845 					      readonly, inherit);
1846 	if (ret)
1847 		goto free_inherit;
1848 
1849 	if (ptr && copy_to_user(arg +
1850 				offsetof(struct btrfs_ioctl_vol_args_v2,
1851 					transid),
1852 				ptr, sizeof(*ptr)))
1853 		ret = -EFAULT;
1854 
1855 free_inherit:
1856 	kfree(inherit);
1857 free_args:
1858 	kfree(vol_args);
1859 	return ret;
1860 }
1861 
1862 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1863 						void __user *arg)
1864 {
1865 	struct inode *inode = file_inode(file);
1866 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1867 	struct btrfs_root *root = BTRFS_I(inode)->root;
1868 	int ret = 0;
1869 	u64 flags = 0;
1870 
1871 	if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1872 		return -EINVAL;
1873 
1874 	down_read(&fs_info->subvol_sem);
1875 	if (btrfs_root_readonly(root))
1876 		flags |= BTRFS_SUBVOL_RDONLY;
1877 	up_read(&fs_info->subvol_sem);
1878 
1879 	if (copy_to_user(arg, &flags, sizeof(flags)))
1880 		ret = -EFAULT;
1881 
1882 	return ret;
1883 }
1884 
1885 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1886 					      void __user *arg)
1887 {
1888 	struct inode *inode = file_inode(file);
1889 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1890 	struct btrfs_root *root = BTRFS_I(inode)->root;
1891 	struct btrfs_trans_handle *trans;
1892 	u64 root_flags;
1893 	u64 flags;
1894 	int ret = 0;
1895 
1896 	if (!inode_owner_or_capable(inode))
1897 		return -EPERM;
1898 
1899 	ret = mnt_want_write_file(file);
1900 	if (ret)
1901 		goto out;
1902 
1903 	if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1904 		ret = -EINVAL;
1905 		goto out_drop_write;
1906 	}
1907 
1908 	if (copy_from_user(&flags, arg, sizeof(flags))) {
1909 		ret = -EFAULT;
1910 		goto out_drop_write;
1911 	}
1912 
1913 	if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1914 		ret = -EINVAL;
1915 		goto out_drop_write;
1916 	}
1917 
1918 	if (flags & ~BTRFS_SUBVOL_RDONLY) {
1919 		ret = -EOPNOTSUPP;
1920 		goto out_drop_write;
1921 	}
1922 
1923 	down_write(&fs_info->subvol_sem);
1924 
1925 	/* nothing to do */
1926 	if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1927 		goto out_drop_sem;
1928 
1929 	root_flags = btrfs_root_flags(&root->root_item);
1930 	if (flags & BTRFS_SUBVOL_RDONLY) {
1931 		btrfs_set_root_flags(&root->root_item,
1932 				     root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1933 	} else {
1934 		/*
1935 		 * Block RO -> RW transition if this subvolume is involved in
1936 		 * send
1937 		 */
1938 		spin_lock(&root->root_item_lock);
1939 		if (root->send_in_progress == 0) {
1940 			btrfs_set_root_flags(&root->root_item,
1941 				     root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1942 			spin_unlock(&root->root_item_lock);
1943 		} else {
1944 			spin_unlock(&root->root_item_lock);
1945 			btrfs_warn(fs_info,
1946 				   "Attempt to set subvolume %llu read-write during send",
1947 				   root->root_key.objectid);
1948 			ret = -EPERM;
1949 			goto out_drop_sem;
1950 		}
1951 	}
1952 
1953 	trans = btrfs_start_transaction(root, 1);
1954 	if (IS_ERR(trans)) {
1955 		ret = PTR_ERR(trans);
1956 		goto out_reset;
1957 	}
1958 
1959 	ret = btrfs_update_root(trans, fs_info->tree_root,
1960 				&root->root_key, &root->root_item);
1961 	if (ret < 0) {
1962 		btrfs_end_transaction(trans);
1963 		goto out_reset;
1964 	}
1965 
1966 	ret = btrfs_commit_transaction(trans);
1967 
1968 out_reset:
1969 	if (ret)
1970 		btrfs_set_root_flags(&root->root_item, root_flags);
1971 out_drop_sem:
1972 	up_write(&fs_info->subvol_sem);
1973 out_drop_write:
1974 	mnt_drop_write_file(file);
1975 out:
1976 	return ret;
1977 }
1978 
1979 static noinline int key_in_sk(struct btrfs_key *key,
1980 			      struct btrfs_ioctl_search_key *sk)
1981 {
1982 	struct btrfs_key test;
1983 	int ret;
1984 
1985 	test.objectid = sk->min_objectid;
1986 	test.type = sk->min_type;
1987 	test.offset = sk->min_offset;
1988 
1989 	ret = btrfs_comp_cpu_keys(key, &test);
1990 	if (ret < 0)
1991 		return 0;
1992 
1993 	test.objectid = sk->max_objectid;
1994 	test.type = sk->max_type;
1995 	test.offset = sk->max_offset;
1996 
1997 	ret = btrfs_comp_cpu_keys(key, &test);
1998 	if (ret > 0)
1999 		return 0;
2000 	return 1;
2001 }
2002 
2003 static noinline int copy_to_sk(struct btrfs_path *path,
2004 			       struct btrfs_key *key,
2005 			       struct btrfs_ioctl_search_key *sk,
2006 			       size_t *buf_size,
2007 			       char __user *ubuf,
2008 			       unsigned long *sk_offset,
2009 			       int *num_found)
2010 {
2011 	u64 found_transid;
2012 	struct extent_buffer *leaf;
2013 	struct btrfs_ioctl_search_header sh;
2014 	struct btrfs_key test;
2015 	unsigned long item_off;
2016 	unsigned long item_len;
2017 	int nritems;
2018 	int i;
2019 	int slot;
2020 	int ret = 0;
2021 
2022 	leaf = path->nodes[0];
2023 	slot = path->slots[0];
2024 	nritems = btrfs_header_nritems(leaf);
2025 
2026 	if (btrfs_header_generation(leaf) > sk->max_transid) {
2027 		i = nritems;
2028 		goto advance_key;
2029 	}
2030 	found_transid = btrfs_header_generation(leaf);
2031 
2032 	for (i = slot; i < nritems; i++) {
2033 		item_off = btrfs_item_ptr_offset(leaf, i);
2034 		item_len = btrfs_item_size_nr(leaf, i);
2035 
2036 		btrfs_item_key_to_cpu(leaf, key, i);
2037 		if (!key_in_sk(key, sk))
2038 			continue;
2039 
2040 		if (sizeof(sh) + item_len > *buf_size) {
2041 			if (*num_found) {
2042 				ret = 1;
2043 				goto out;
2044 			}
2045 
2046 			/*
2047 			 * return one empty item back for v1, which does not
2048 			 * handle -EOVERFLOW
2049 			 */
2050 
2051 			*buf_size = sizeof(sh) + item_len;
2052 			item_len = 0;
2053 			ret = -EOVERFLOW;
2054 		}
2055 
2056 		if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2057 			ret = 1;
2058 			goto out;
2059 		}
2060 
2061 		sh.objectid = key->objectid;
2062 		sh.offset = key->offset;
2063 		sh.type = key->type;
2064 		sh.len = item_len;
2065 		sh.transid = found_transid;
2066 
2067 		/* copy search result header */
2068 		if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
2069 			ret = -EFAULT;
2070 			goto out;
2071 		}
2072 
2073 		*sk_offset += sizeof(sh);
2074 
2075 		if (item_len) {
2076 			char __user *up = ubuf + *sk_offset;
2077 			/* copy the item */
2078 			if (read_extent_buffer_to_user(leaf, up,
2079 						       item_off, item_len)) {
2080 				ret = -EFAULT;
2081 				goto out;
2082 			}
2083 
2084 			*sk_offset += item_len;
2085 		}
2086 		(*num_found)++;
2087 
2088 		if (ret) /* -EOVERFLOW from above */
2089 			goto out;
2090 
2091 		if (*num_found >= sk->nr_items) {
2092 			ret = 1;
2093 			goto out;
2094 		}
2095 	}
2096 advance_key:
2097 	ret = 0;
2098 	test.objectid = sk->max_objectid;
2099 	test.type = sk->max_type;
2100 	test.offset = sk->max_offset;
2101 	if (btrfs_comp_cpu_keys(key, &test) >= 0)
2102 		ret = 1;
2103 	else if (key->offset < (u64)-1)
2104 		key->offset++;
2105 	else if (key->type < (u8)-1) {
2106 		key->offset = 0;
2107 		key->type++;
2108 	} else if (key->objectid < (u64)-1) {
2109 		key->offset = 0;
2110 		key->type = 0;
2111 		key->objectid++;
2112 	} else
2113 		ret = 1;
2114 out:
2115 	/*
2116 	 *  0: all items from this leaf copied, continue with next
2117 	 *  1: * more items can be copied, but unused buffer is too small
2118 	 *     * all items were found
2119 	 *     Either way, it will stops the loop which iterates to the next
2120 	 *     leaf
2121 	 *  -EOVERFLOW: item was to large for buffer
2122 	 *  -EFAULT: could not copy extent buffer back to userspace
2123 	 */
2124 	return ret;
2125 }
2126 
2127 static noinline int search_ioctl(struct inode *inode,
2128 				 struct btrfs_ioctl_search_key *sk,
2129 				 size_t *buf_size,
2130 				 char __user *ubuf)
2131 {
2132 	struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2133 	struct btrfs_root *root;
2134 	struct btrfs_key key;
2135 	struct btrfs_path *path;
2136 	int ret;
2137 	int num_found = 0;
2138 	unsigned long sk_offset = 0;
2139 
2140 	if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2141 		*buf_size = sizeof(struct btrfs_ioctl_search_header);
2142 		return -EOVERFLOW;
2143 	}
2144 
2145 	path = btrfs_alloc_path();
2146 	if (!path)
2147 		return -ENOMEM;
2148 
2149 	if (sk->tree_id == 0) {
2150 		/* search the root of the inode that was passed */
2151 		root = BTRFS_I(inode)->root;
2152 	} else {
2153 		key.objectid = sk->tree_id;
2154 		key.type = BTRFS_ROOT_ITEM_KEY;
2155 		key.offset = (u64)-1;
2156 		root = btrfs_read_fs_root_no_name(info, &key);
2157 		if (IS_ERR(root)) {
2158 			btrfs_free_path(path);
2159 			return PTR_ERR(root);
2160 		}
2161 	}
2162 
2163 	key.objectid = sk->min_objectid;
2164 	key.type = sk->min_type;
2165 	key.offset = sk->min_offset;
2166 
2167 	while (1) {
2168 		ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2169 		if (ret != 0) {
2170 			if (ret > 0)
2171 				ret = 0;
2172 			goto err;
2173 		}
2174 		ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2175 				 &sk_offset, &num_found);
2176 		btrfs_release_path(path);
2177 		if (ret)
2178 			break;
2179 
2180 	}
2181 	if (ret > 0)
2182 		ret = 0;
2183 err:
2184 	sk->nr_items = num_found;
2185 	btrfs_free_path(path);
2186 	return ret;
2187 }
2188 
2189 static noinline int btrfs_ioctl_tree_search(struct file *file,
2190 					   void __user *argp)
2191 {
2192 	struct btrfs_ioctl_search_args __user *uargs;
2193 	struct btrfs_ioctl_search_key sk;
2194 	struct inode *inode;
2195 	int ret;
2196 	size_t buf_size;
2197 
2198 	if (!capable(CAP_SYS_ADMIN))
2199 		return -EPERM;
2200 
2201 	uargs = (struct btrfs_ioctl_search_args __user *)argp;
2202 
2203 	if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2204 		return -EFAULT;
2205 
2206 	buf_size = sizeof(uargs->buf);
2207 
2208 	inode = file_inode(file);
2209 	ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2210 
2211 	/*
2212 	 * In the origin implementation an overflow is handled by returning a
2213 	 * search header with a len of zero, so reset ret.
2214 	 */
2215 	if (ret == -EOVERFLOW)
2216 		ret = 0;
2217 
2218 	if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2219 		ret = -EFAULT;
2220 	return ret;
2221 }
2222 
2223 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2224 					       void __user *argp)
2225 {
2226 	struct btrfs_ioctl_search_args_v2 __user *uarg;
2227 	struct btrfs_ioctl_search_args_v2 args;
2228 	struct inode *inode;
2229 	int ret;
2230 	size_t buf_size;
2231 	const size_t buf_limit = SZ_16M;
2232 
2233 	if (!capable(CAP_SYS_ADMIN))
2234 		return -EPERM;
2235 
2236 	/* copy search header and buffer size */
2237 	uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2238 	if (copy_from_user(&args, uarg, sizeof(args)))
2239 		return -EFAULT;
2240 
2241 	buf_size = args.buf_size;
2242 
2243 	/* limit result size to 16MB */
2244 	if (buf_size > buf_limit)
2245 		buf_size = buf_limit;
2246 
2247 	inode = file_inode(file);
2248 	ret = search_ioctl(inode, &args.key, &buf_size,
2249 			   (char __user *)(&uarg->buf[0]));
2250 	if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2251 		ret = -EFAULT;
2252 	else if (ret == -EOVERFLOW &&
2253 		copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2254 		ret = -EFAULT;
2255 
2256 	return ret;
2257 }
2258 
2259 /*
2260  * Search INODE_REFs to identify path name of 'dirid' directory
2261  * in a 'tree_id' tree. and sets path name to 'name'.
2262  */
2263 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2264 				u64 tree_id, u64 dirid, char *name)
2265 {
2266 	struct btrfs_root *root;
2267 	struct btrfs_key key;
2268 	char *ptr;
2269 	int ret = -1;
2270 	int slot;
2271 	int len;
2272 	int total_len = 0;
2273 	struct btrfs_inode_ref *iref;
2274 	struct extent_buffer *l;
2275 	struct btrfs_path *path;
2276 
2277 	if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2278 		name[0]='\0';
2279 		return 0;
2280 	}
2281 
2282 	path = btrfs_alloc_path();
2283 	if (!path)
2284 		return -ENOMEM;
2285 
2286 	ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2287 
2288 	key.objectid = tree_id;
2289 	key.type = BTRFS_ROOT_ITEM_KEY;
2290 	key.offset = (u64)-1;
2291 	root = btrfs_read_fs_root_no_name(info, &key);
2292 	if (IS_ERR(root)) {
2293 		ret = PTR_ERR(root);
2294 		goto out;
2295 	}
2296 
2297 	key.objectid = dirid;
2298 	key.type = BTRFS_INODE_REF_KEY;
2299 	key.offset = (u64)-1;
2300 
2301 	while (1) {
2302 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2303 		if (ret < 0)
2304 			goto out;
2305 		else if (ret > 0) {
2306 			ret = btrfs_previous_item(root, path, dirid,
2307 						  BTRFS_INODE_REF_KEY);
2308 			if (ret < 0)
2309 				goto out;
2310 			else if (ret > 0) {
2311 				ret = -ENOENT;
2312 				goto out;
2313 			}
2314 		}
2315 
2316 		l = path->nodes[0];
2317 		slot = path->slots[0];
2318 		btrfs_item_key_to_cpu(l, &key, slot);
2319 
2320 		iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2321 		len = btrfs_inode_ref_name_len(l, iref);
2322 		ptr -= len + 1;
2323 		total_len += len + 1;
2324 		if (ptr < name) {
2325 			ret = -ENAMETOOLONG;
2326 			goto out;
2327 		}
2328 
2329 		*(ptr + len) = '/';
2330 		read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2331 
2332 		if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2333 			break;
2334 
2335 		btrfs_release_path(path);
2336 		key.objectid = key.offset;
2337 		key.offset = (u64)-1;
2338 		dirid = key.objectid;
2339 	}
2340 	memmove(name, ptr, total_len);
2341 	name[total_len] = '\0';
2342 	ret = 0;
2343 out:
2344 	btrfs_free_path(path);
2345 	return ret;
2346 }
2347 
2348 static int btrfs_search_path_in_tree_user(struct inode *inode,
2349 				struct btrfs_ioctl_ino_lookup_user_args *args)
2350 {
2351 	struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2352 	struct super_block *sb = inode->i_sb;
2353 	struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2354 	u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2355 	u64 dirid = args->dirid;
2356 	unsigned long item_off;
2357 	unsigned long item_len;
2358 	struct btrfs_inode_ref *iref;
2359 	struct btrfs_root_ref *rref;
2360 	struct btrfs_root *root;
2361 	struct btrfs_path *path;
2362 	struct btrfs_key key, key2;
2363 	struct extent_buffer *leaf;
2364 	struct inode *temp_inode;
2365 	char *ptr;
2366 	int slot;
2367 	int len;
2368 	int total_len = 0;
2369 	int ret;
2370 
2371 	path = btrfs_alloc_path();
2372 	if (!path)
2373 		return -ENOMEM;
2374 
2375 	/*
2376 	 * If the bottom subvolume does not exist directly under upper_limit,
2377 	 * construct the path in from the bottom up.
2378 	 */
2379 	if (dirid != upper_limit.objectid) {
2380 		ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2381 
2382 		key.objectid = treeid;
2383 		key.type = BTRFS_ROOT_ITEM_KEY;
2384 		key.offset = (u64)-1;
2385 		root = btrfs_read_fs_root_no_name(fs_info, &key);
2386 		if (IS_ERR(root)) {
2387 			ret = PTR_ERR(root);
2388 			goto out;
2389 		}
2390 
2391 		key.objectid = dirid;
2392 		key.type = BTRFS_INODE_REF_KEY;
2393 		key.offset = (u64)-1;
2394 		while (1) {
2395 			ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2396 			if (ret < 0) {
2397 				goto out;
2398 			} else if (ret > 0) {
2399 				ret = btrfs_previous_item(root, path, dirid,
2400 							  BTRFS_INODE_REF_KEY);
2401 				if (ret < 0) {
2402 					goto out;
2403 				} else if (ret > 0) {
2404 					ret = -ENOENT;
2405 					goto out;
2406 				}
2407 			}
2408 
2409 			leaf = path->nodes[0];
2410 			slot = path->slots[0];
2411 			btrfs_item_key_to_cpu(leaf, &key, slot);
2412 
2413 			iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2414 			len = btrfs_inode_ref_name_len(leaf, iref);
2415 			ptr -= len + 1;
2416 			total_len += len + 1;
2417 			if (ptr < args->path) {
2418 				ret = -ENAMETOOLONG;
2419 				goto out;
2420 			}
2421 
2422 			*(ptr + len) = '/';
2423 			read_extent_buffer(leaf, ptr,
2424 					(unsigned long)(iref + 1), len);
2425 
2426 			/* Check the read+exec permission of this directory */
2427 			ret = btrfs_previous_item(root, path, dirid,
2428 						  BTRFS_INODE_ITEM_KEY);
2429 			if (ret < 0) {
2430 				goto out;
2431 			} else if (ret > 0) {
2432 				ret = -ENOENT;
2433 				goto out;
2434 			}
2435 
2436 			leaf = path->nodes[0];
2437 			slot = path->slots[0];
2438 			btrfs_item_key_to_cpu(leaf, &key2, slot);
2439 			if (key2.objectid != dirid) {
2440 				ret = -ENOENT;
2441 				goto out;
2442 			}
2443 
2444 			temp_inode = btrfs_iget(sb, &key2, root, NULL);
2445 			if (IS_ERR(temp_inode)) {
2446 				ret = PTR_ERR(temp_inode);
2447 				goto out;
2448 			}
2449 			ret = inode_permission(temp_inode, MAY_READ | MAY_EXEC);
2450 			iput(temp_inode);
2451 			if (ret) {
2452 				ret = -EACCES;
2453 				goto out;
2454 			}
2455 
2456 			if (key.offset == upper_limit.objectid)
2457 				break;
2458 			if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2459 				ret = -EACCES;
2460 				goto out;
2461 			}
2462 
2463 			btrfs_release_path(path);
2464 			key.objectid = key.offset;
2465 			key.offset = (u64)-1;
2466 			dirid = key.objectid;
2467 		}
2468 
2469 		memmove(args->path, ptr, total_len);
2470 		args->path[total_len] = '\0';
2471 		btrfs_release_path(path);
2472 	}
2473 
2474 	/* Get the bottom subvolume's name from ROOT_REF */
2475 	root = fs_info->tree_root;
2476 	key.objectid = treeid;
2477 	key.type = BTRFS_ROOT_REF_KEY;
2478 	key.offset = args->treeid;
2479 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2480 	if (ret < 0) {
2481 		goto out;
2482 	} else if (ret > 0) {
2483 		ret = -ENOENT;
2484 		goto out;
2485 	}
2486 
2487 	leaf = path->nodes[0];
2488 	slot = path->slots[0];
2489 	btrfs_item_key_to_cpu(leaf, &key, slot);
2490 
2491 	item_off = btrfs_item_ptr_offset(leaf, slot);
2492 	item_len = btrfs_item_size_nr(leaf, slot);
2493 	/* Check if dirid in ROOT_REF corresponds to passed dirid */
2494 	rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2495 	if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2496 		ret = -EINVAL;
2497 		goto out;
2498 	}
2499 
2500 	/* Copy subvolume's name */
2501 	item_off += sizeof(struct btrfs_root_ref);
2502 	item_len -= sizeof(struct btrfs_root_ref);
2503 	read_extent_buffer(leaf, args->name, item_off, item_len);
2504 	args->name[item_len] = 0;
2505 
2506 out:
2507 	btrfs_free_path(path);
2508 	return ret;
2509 }
2510 
2511 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2512 					   void __user *argp)
2513 {
2514 	struct btrfs_ioctl_ino_lookup_args *args;
2515 	struct inode *inode;
2516 	int ret = 0;
2517 
2518 	args = memdup_user(argp, sizeof(*args));
2519 	if (IS_ERR(args))
2520 		return PTR_ERR(args);
2521 
2522 	inode = file_inode(file);
2523 
2524 	/*
2525 	 * Unprivileged query to obtain the containing subvolume root id. The
2526 	 * path is reset so it's consistent with btrfs_search_path_in_tree.
2527 	 */
2528 	if (args->treeid == 0)
2529 		args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2530 
2531 	if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2532 		args->name[0] = 0;
2533 		goto out;
2534 	}
2535 
2536 	if (!capable(CAP_SYS_ADMIN)) {
2537 		ret = -EPERM;
2538 		goto out;
2539 	}
2540 
2541 	ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2542 					args->treeid, args->objectid,
2543 					args->name);
2544 
2545 out:
2546 	if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2547 		ret = -EFAULT;
2548 
2549 	kfree(args);
2550 	return ret;
2551 }
2552 
2553 /*
2554  * Version of ino_lookup ioctl (unprivileged)
2555  *
2556  * The main differences from ino_lookup ioctl are:
2557  *
2558  *   1. Read + Exec permission will be checked using inode_permission() during
2559  *      path construction. -EACCES will be returned in case of failure.
2560  *   2. Path construction will be stopped at the inode number which corresponds
2561  *      to the fd with which this ioctl is called. If constructed path does not
2562  *      exist under fd's inode, -EACCES will be returned.
2563  *   3. The name of bottom subvolume is also searched and filled.
2564  */
2565 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2566 {
2567 	struct btrfs_ioctl_ino_lookup_user_args *args;
2568 	struct inode *inode;
2569 	int ret;
2570 
2571 	args = memdup_user(argp, sizeof(*args));
2572 	if (IS_ERR(args))
2573 		return PTR_ERR(args);
2574 
2575 	inode = file_inode(file);
2576 
2577 	if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2578 	    BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2579 		/*
2580 		 * The subvolume does not exist under fd with which this is
2581 		 * called
2582 		 */
2583 		kfree(args);
2584 		return -EACCES;
2585 	}
2586 
2587 	ret = btrfs_search_path_in_tree_user(inode, args);
2588 
2589 	if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2590 		ret = -EFAULT;
2591 
2592 	kfree(args);
2593 	return ret;
2594 }
2595 
2596 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2597 static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp)
2598 {
2599 	struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2600 	struct btrfs_fs_info *fs_info;
2601 	struct btrfs_root *root;
2602 	struct btrfs_path *path;
2603 	struct btrfs_key key;
2604 	struct btrfs_root_item *root_item;
2605 	struct btrfs_root_ref *rref;
2606 	struct extent_buffer *leaf;
2607 	unsigned long item_off;
2608 	unsigned long item_len;
2609 	struct inode *inode;
2610 	int slot;
2611 	int ret = 0;
2612 
2613 	path = btrfs_alloc_path();
2614 	if (!path)
2615 		return -ENOMEM;
2616 
2617 	subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2618 	if (!subvol_info) {
2619 		btrfs_free_path(path);
2620 		return -ENOMEM;
2621 	}
2622 
2623 	inode = file_inode(file);
2624 	fs_info = BTRFS_I(inode)->root->fs_info;
2625 
2626 	/* Get root_item of inode's subvolume */
2627 	key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2628 	key.type = BTRFS_ROOT_ITEM_KEY;
2629 	key.offset = (u64)-1;
2630 	root = btrfs_read_fs_root_no_name(fs_info, &key);
2631 	if (IS_ERR(root)) {
2632 		ret = PTR_ERR(root);
2633 		goto out;
2634 	}
2635 	root_item = &root->root_item;
2636 
2637 	subvol_info->treeid = key.objectid;
2638 
2639 	subvol_info->generation = btrfs_root_generation(root_item);
2640 	subvol_info->flags = btrfs_root_flags(root_item);
2641 
2642 	memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2643 	memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2644 						    BTRFS_UUID_SIZE);
2645 	memcpy(subvol_info->received_uuid, root_item->received_uuid,
2646 						    BTRFS_UUID_SIZE);
2647 
2648 	subvol_info->ctransid = btrfs_root_ctransid(root_item);
2649 	subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2650 	subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2651 
2652 	subvol_info->otransid = btrfs_root_otransid(root_item);
2653 	subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2654 	subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2655 
2656 	subvol_info->stransid = btrfs_root_stransid(root_item);
2657 	subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2658 	subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2659 
2660 	subvol_info->rtransid = btrfs_root_rtransid(root_item);
2661 	subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2662 	subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2663 
2664 	if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2665 		/* Search root tree for ROOT_BACKREF of this subvolume */
2666 		root = fs_info->tree_root;
2667 
2668 		key.type = BTRFS_ROOT_BACKREF_KEY;
2669 		key.offset = 0;
2670 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2671 		if (ret < 0) {
2672 			goto out;
2673 		} else if (path->slots[0] >=
2674 			   btrfs_header_nritems(path->nodes[0])) {
2675 			ret = btrfs_next_leaf(root, path);
2676 			if (ret < 0) {
2677 				goto out;
2678 			} else if (ret > 0) {
2679 				ret = -EUCLEAN;
2680 				goto out;
2681 			}
2682 		}
2683 
2684 		leaf = path->nodes[0];
2685 		slot = path->slots[0];
2686 		btrfs_item_key_to_cpu(leaf, &key, slot);
2687 		if (key.objectid == subvol_info->treeid &&
2688 		    key.type == BTRFS_ROOT_BACKREF_KEY) {
2689 			subvol_info->parent_id = key.offset;
2690 
2691 			rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2692 			subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2693 
2694 			item_off = btrfs_item_ptr_offset(leaf, slot)
2695 					+ sizeof(struct btrfs_root_ref);
2696 			item_len = btrfs_item_size_nr(leaf, slot)
2697 					- sizeof(struct btrfs_root_ref);
2698 			read_extent_buffer(leaf, subvol_info->name,
2699 					   item_off, item_len);
2700 		} else {
2701 			ret = -ENOENT;
2702 			goto out;
2703 		}
2704 	}
2705 
2706 	if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2707 		ret = -EFAULT;
2708 
2709 out:
2710 	btrfs_free_path(path);
2711 	kzfree(subvol_info);
2712 	return ret;
2713 }
2714 
2715 /*
2716  * Return ROOT_REF information of the subvolume containing this inode
2717  * except the subvolume name.
2718  */
2719 static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp)
2720 {
2721 	struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2722 	struct btrfs_root_ref *rref;
2723 	struct btrfs_root *root;
2724 	struct btrfs_path *path;
2725 	struct btrfs_key key;
2726 	struct extent_buffer *leaf;
2727 	struct inode *inode;
2728 	u64 objectid;
2729 	int slot;
2730 	int ret;
2731 	u8 found;
2732 
2733 	path = btrfs_alloc_path();
2734 	if (!path)
2735 		return -ENOMEM;
2736 
2737 	rootrefs = memdup_user(argp, sizeof(*rootrefs));
2738 	if (IS_ERR(rootrefs)) {
2739 		btrfs_free_path(path);
2740 		return PTR_ERR(rootrefs);
2741 	}
2742 
2743 	inode = file_inode(file);
2744 	root = BTRFS_I(inode)->root->fs_info->tree_root;
2745 	objectid = BTRFS_I(inode)->root->root_key.objectid;
2746 
2747 	key.objectid = objectid;
2748 	key.type = BTRFS_ROOT_REF_KEY;
2749 	key.offset = rootrefs->min_treeid;
2750 	found = 0;
2751 
2752 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2753 	if (ret < 0) {
2754 		goto out;
2755 	} else if (path->slots[0] >=
2756 		   btrfs_header_nritems(path->nodes[0])) {
2757 		ret = btrfs_next_leaf(root, path);
2758 		if (ret < 0) {
2759 			goto out;
2760 		} else if (ret > 0) {
2761 			ret = -EUCLEAN;
2762 			goto out;
2763 		}
2764 	}
2765 	while (1) {
2766 		leaf = path->nodes[0];
2767 		slot = path->slots[0];
2768 
2769 		btrfs_item_key_to_cpu(leaf, &key, slot);
2770 		if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2771 			ret = 0;
2772 			goto out;
2773 		}
2774 
2775 		if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2776 			ret = -EOVERFLOW;
2777 			goto out;
2778 		}
2779 
2780 		rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2781 		rootrefs->rootref[found].treeid = key.offset;
2782 		rootrefs->rootref[found].dirid =
2783 				  btrfs_root_ref_dirid(leaf, rref);
2784 		found++;
2785 
2786 		ret = btrfs_next_item(root, path);
2787 		if (ret < 0) {
2788 			goto out;
2789 		} else if (ret > 0) {
2790 			ret = -EUCLEAN;
2791 			goto out;
2792 		}
2793 	}
2794 
2795 out:
2796 	if (!ret || ret == -EOVERFLOW) {
2797 		rootrefs->num_items = found;
2798 		/* update min_treeid for next search */
2799 		if (found)
2800 			rootrefs->min_treeid =
2801 				rootrefs->rootref[found - 1].treeid + 1;
2802 		if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2803 			ret = -EFAULT;
2804 	}
2805 
2806 	kfree(rootrefs);
2807 	btrfs_free_path(path);
2808 
2809 	return ret;
2810 }
2811 
2812 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2813 					     void __user *arg)
2814 {
2815 	struct dentry *parent = file->f_path.dentry;
2816 	struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2817 	struct dentry *dentry;
2818 	struct inode *dir = d_inode(parent);
2819 	struct inode *inode;
2820 	struct btrfs_root *root = BTRFS_I(dir)->root;
2821 	struct btrfs_root *dest = NULL;
2822 	struct btrfs_ioctl_vol_args *vol_args;
2823 	int namelen;
2824 	int err = 0;
2825 
2826 	if (!S_ISDIR(dir->i_mode))
2827 		return -ENOTDIR;
2828 
2829 	vol_args = memdup_user(arg, sizeof(*vol_args));
2830 	if (IS_ERR(vol_args))
2831 		return PTR_ERR(vol_args);
2832 
2833 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2834 	namelen = strlen(vol_args->name);
2835 	if (strchr(vol_args->name, '/') ||
2836 	    strncmp(vol_args->name, "..", namelen) == 0) {
2837 		err = -EINVAL;
2838 		goto out;
2839 	}
2840 
2841 	err = mnt_want_write_file(file);
2842 	if (err)
2843 		goto out;
2844 
2845 
2846 	err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2847 	if (err == -EINTR)
2848 		goto out_drop_write;
2849 	dentry = lookup_one_len(vol_args->name, parent, namelen);
2850 	if (IS_ERR(dentry)) {
2851 		err = PTR_ERR(dentry);
2852 		goto out_unlock_dir;
2853 	}
2854 
2855 	if (d_really_is_negative(dentry)) {
2856 		err = -ENOENT;
2857 		goto out_dput;
2858 	}
2859 
2860 	inode = d_inode(dentry);
2861 	dest = BTRFS_I(inode)->root;
2862 	if (!capable(CAP_SYS_ADMIN)) {
2863 		/*
2864 		 * Regular user.  Only allow this with a special mount
2865 		 * option, when the user has write+exec access to the
2866 		 * subvol root, and when rmdir(2) would have been
2867 		 * allowed.
2868 		 *
2869 		 * Note that this is _not_ check that the subvol is
2870 		 * empty or doesn't contain data that we wouldn't
2871 		 * otherwise be able to delete.
2872 		 *
2873 		 * Users who want to delete empty subvols should try
2874 		 * rmdir(2).
2875 		 */
2876 		err = -EPERM;
2877 		if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2878 			goto out_dput;
2879 
2880 		/*
2881 		 * Do not allow deletion if the parent dir is the same
2882 		 * as the dir to be deleted.  That means the ioctl
2883 		 * must be called on the dentry referencing the root
2884 		 * of the subvol, not a random directory contained
2885 		 * within it.
2886 		 */
2887 		err = -EINVAL;
2888 		if (root == dest)
2889 			goto out_dput;
2890 
2891 		err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2892 		if (err)
2893 			goto out_dput;
2894 	}
2895 
2896 	/* check if subvolume may be deleted by a user */
2897 	err = btrfs_may_delete(dir, dentry, 1);
2898 	if (err)
2899 		goto out_dput;
2900 
2901 	if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2902 		err = -EINVAL;
2903 		goto out_dput;
2904 	}
2905 
2906 	inode_lock(inode);
2907 	err = btrfs_delete_subvolume(dir, dentry);
2908 	inode_unlock(inode);
2909 	if (!err)
2910 		d_delete(dentry);
2911 
2912 out_dput:
2913 	dput(dentry);
2914 out_unlock_dir:
2915 	inode_unlock(dir);
2916 out_drop_write:
2917 	mnt_drop_write_file(file);
2918 out:
2919 	kfree(vol_args);
2920 	return err;
2921 }
2922 
2923 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2924 {
2925 	struct inode *inode = file_inode(file);
2926 	struct btrfs_root *root = BTRFS_I(inode)->root;
2927 	struct btrfs_ioctl_defrag_range_args *range;
2928 	int ret;
2929 
2930 	ret = mnt_want_write_file(file);
2931 	if (ret)
2932 		return ret;
2933 
2934 	if (btrfs_root_readonly(root)) {
2935 		ret = -EROFS;
2936 		goto out;
2937 	}
2938 
2939 	switch (inode->i_mode & S_IFMT) {
2940 	case S_IFDIR:
2941 		if (!capable(CAP_SYS_ADMIN)) {
2942 			ret = -EPERM;
2943 			goto out;
2944 		}
2945 		ret = btrfs_defrag_root(root);
2946 		break;
2947 	case S_IFREG:
2948 		/*
2949 		 * Note that this does not check the file descriptor for write
2950 		 * access. This prevents defragmenting executables that are
2951 		 * running and allows defrag on files open in read-only mode.
2952 		 */
2953 		if (!capable(CAP_SYS_ADMIN) &&
2954 		    inode_permission(inode, MAY_WRITE)) {
2955 			ret = -EPERM;
2956 			goto out;
2957 		}
2958 
2959 		range = kzalloc(sizeof(*range), GFP_KERNEL);
2960 		if (!range) {
2961 			ret = -ENOMEM;
2962 			goto out;
2963 		}
2964 
2965 		if (argp) {
2966 			if (copy_from_user(range, argp,
2967 					   sizeof(*range))) {
2968 				ret = -EFAULT;
2969 				kfree(range);
2970 				goto out;
2971 			}
2972 			/* compression requires us to start the IO */
2973 			if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2974 				range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2975 				range->extent_thresh = (u32)-1;
2976 			}
2977 		} else {
2978 			/* the rest are all set to zero by kzalloc */
2979 			range->len = (u64)-1;
2980 		}
2981 		ret = btrfs_defrag_file(file_inode(file), file,
2982 					range, BTRFS_OLDEST_GENERATION, 0);
2983 		if (ret > 0)
2984 			ret = 0;
2985 		kfree(range);
2986 		break;
2987 	default:
2988 		ret = -EINVAL;
2989 	}
2990 out:
2991 	mnt_drop_write_file(file);
2992 	return ret;
2993 }
2994 
2995 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
2996 {
2997 	struct btrfs_ioctl_vol_args *vol_args;
2998 	int ret;
2999 
3000 	if (!capable(CAP_SYS_ADMIN))
3001 		return -EPERM;
3002 
3003 	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
3004 		return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3005 
3006 	vol_args = memdup_user(arg, sizeof(*vol_args));
3007 	if (IS_ERR(vol_args)) {
3008 		ret = PTR_ERR(vol_args);
3009 		goto out;
3010 	}
3011 
3012 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3013 	ret = btrfs_init_new_device(fs_info, vol_args->name);
3014 
3015 	if (!ret)
3016 		btrfs_info(fs_info, "disk added %s", vol_args->name);
3017 
3018 	kfree(vol_args);
3019 out:
3020 	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3021 	return ret;
3022 }
3023 
3024 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3025 {
3026 	struct inode *inode = file_inode(file);
3027 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3028 	struct btrfs_ioctl_vol_args_v2 *vol_args;
3029 	int ret;
3030 
3031 	if (!capable(CAP_SYS_ADMIN))
3032 		return -EPERM;
3033 
3034 	ret = mnt_want_write_file(file);
3035 	if (ret)
3036 		return ret;
3037 
3038 	vol_args = memdup_user(arg, sizeof(*vol_args));
3039 	if (IS_ERR(vol_args)) {
3040 		ret = PTR_ERR(vol_args);
3041 		goto err_drop;
3042 	}
3043 
3044 	/* Check for compatibility reject unknown flags */
3045 	if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED) {
3046 		ret = -EOPNOTSUPP;
3047 		goto out;
3048 	}
3049 
3050 	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3051 		ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3052 		goto out;
3053 	}
3054 
3055 	if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3056 		ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
3057 	} else {
3058 		vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3059 		ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3060 	}
3061 	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3062 
3063 	if (!ret) {
3064 		if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3065 			btrfs_info(fs_info, "device deleted: id %llu",
3066 					vol_args->devid);
3067 		else
3068 			btrfs_info(fs_info, "device deleted: %s",
3069 					vol_args->name);
3070 	}
3071 out:
3072 	kfree(vol_args);
3073 err_drop:
3074 	mnt_drop_write_file(file);
3075 	return ret;
3076 }
3077 
3078 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3079 {
3080 	struct inode *inode = file_inode(file);
3081 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3082 	struct btrfs_ioctl_vol_args *vol_args;
3083 	int ret;
3084 
3085 	if (!capable(CAP_SYS_ADMIN))
3086 		return -EPERM;
3087 
3088 	ret = mnt_want_write_file(file);
3089 	if (ret)
3090 		return ret;
3091 
3092 	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3093 		ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3094 		goto out_drop_write;
3095 	}
3096 
3097 	vol_args = memdup_user(arg, sizeof(*vol_args));
3098 	if (IS_ERR(vol_args)) {
3099 		ret = PTR_ERR(vol_args);
3100 		goto out;
3101 	}
3102 
3103 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3104 	ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3105 
3106 	if (!ret)
3107 		btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3108 	kfree(vol_args);
3109 out:
3110 	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3111 out_drop_write:
3112 	mnt_drop_write_file(file);
3113 
3114 	return ret;
3115 }
3116 
3117 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3118 				void __user *arg)
3119 {
3120 	struct btrfs_ioctl_fs_info_args *fi_args;
3121 	struct btrfs_device *device;
3122 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3123 	int ret = 0;
3124 
3125 	fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
3126 	if (!fi_args)
3127 		return -ENOMEM;
3128 
3129 	rcu_read_lock();
3130 	fi_args->num_devices = fs_devices->num_devices;
3131 
3132 	list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3133 		if (device->devid > fi_args->max_id)
3134 			fi_args->max_id = device->devid;
3135 	}
3136 	rcu_read_unlock();
3137 
3138 	memcpy(&fi_args->fsid, fs_info->fsid, sizeof(fi_args->fsid));
3139 	fi_args->nodesize = fs_info->nodesize;
3140 	fi_args->sectorsize = fs_info->sectorsize;
3141 	fi_args->clone_alignment = fs_info->sectorsize;
3142 
3143 	if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3144 		ret = -EFAULT;
3145 
3146 	kfree(fi_args);
3147 	return ret;
3148 }
3149 
3150 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3151 				 void __user *arg)
3152 {
3153 	struct btrfs_ioctl_dev_info_args *di_args;
3154 	struct btrfs_device *dev;
3155 	int ret = 0;
3156 	char *s_uuid = NULL;
3157 
3158 	di_args = memdup_user(arg, sizeof(*di_args));
3159 	if (IS_ERR(di_args))
3160 		return PTR_ERR(di_args);
3161 
3162 	if (!btrfs_is_empty_uuid(di_args->uuid))
3163 		s_uuid = di_args->uuid;
3164 
3165 	rcu_read_lock();
3166 	dev = btrfs_find_device(fs_info, di_args->devid, s_uuid, NULL);
3167 
3168 	if (!dev) {
3169 		ret = -ENODEV;
3170 		goto out;
3171 	}
3172 
3173 	di_args->devid = dev->devid;
3174 	di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3175 	di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3176 	memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3177 	if (dev->name) {
3178 		strncpy(di_args->path, rcu_str_deref(dev->name),
3179 				sizeof(di_args->path) - 1);
3180 		di_args->path[sizeof(di_args->path) - 1] = 0;
3181 	} else {
3182 		di_args->path[0] = '\0';
3183 	}
3184 
3185 out:
3186 	rcu_read_unlock();
3187 	if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3188 		ret = -EFAULT;
3189 
3190 	kfree(di_args);
3191 	return ret;
3192 }
3193 
3194 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
3195 {
3196 	struct page *page;
3197 
3198 	page = grab_cache_page(inode->i_mapping, index);
3199 	if (!page)
3200 		return ERR_PTR(-ENOMEM);
3201 
3202 	if (!PageUptodate(page)) {
3203 		int ret;
3204 
3205 		ret = btrfs_readpage(NULL, page);
3206 		if (ret)
3207 			return ERR_PTR(ret);
3208 		lock_page(page);
3209 		if (!PageUptodate(page)) {
3210 			unlock_page(page);
3211 			put_page(page);
3212 			return ERR_PTR(-EIO);
3213 		}
3214 		if (page->mapping != inode->i_mapping) {
3215 			unlock_page(page);
3216 			put_page(page);
3217 			return ERR_PTR(-EAGAIN);
3218 		}
3219 	}
3220 
3221 	return page;
3222 }
3223 
3224 static int gather_extent_pages(struct inode *inode, struct page **pages,
3225 			       int num_pages, u64 off)
3226 {
3227 	int i;
3228 	pgoff_t index = off >> PAGE_SHIFT;
3229 
3230 	for (i = 0; i < num_pages; i++) {
3231 again:
3232 		pages[i] = extent_same_get_page(inode, index + i);
3233 		if (IS_ERR(pages[i])) {
3234 			int err = PTR_ERR(pages[i]);
3235 
3236 			if (err == -EAGAIN)
3237 				goto again;
3238 			pages[i] = NULL;
3239 			return err;
3240 		}
3241 	}
3242 	return 0;
3243 }
3244 
3245 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
3246 			     bool retry_range_locking)
3247 {
3248 	/*
3249 	 * Do any pending delalloc/csum calculations on inode, one way or
3250 	 * another, and lock file content.
3251 	 * The locking order is:
3252 	 *
3253 	 *   1) pages
3254 	 *   2) range in the inode's io tree
3255 	 */
3256 	while (1) {
3257 		struct btrfs_ordered_extent *ordered;
3258 		lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
3259 		ordered = btrfs_lookup_first_ordered_extent(inode,
3260 							    off + len - 1);
3261 		if ((!ordered ||
3262 		     ordered->file_offset + ordered->len <= off ||
3263 		     ordered->file_offset >= off + len) &&
3264 		    !test_range_bit(&BTRFS_I(inode)->io_tree, off,
3265 				    off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
3266 			if (ordered)
3267 				btrfs_put_ordered_extent(ordered);
3268 			break;
3269 		}
3270 		unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
3271 		if (ordered)
3272 			btrfs_put_ordered_extent(ordered);
3273 		if (!retry_range_locking)
3274 			return -EAGAIN;
3275 		btrfs_wait_ordered_range(inode, off, len);
3276 	}
3277 	return 0;
3278 }
3279 
3280 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
3281 {
3282 	inode_unlock(inode1);
3283 	inode_unlock(inode2);
3284 }
3285 
3286 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
3287 {
3288 	if (inode1 < inode2)
3289 		swap(inode1, inode2);
3290 
3291 	inode_lock_nested(inode1, I_MUTEX_PARENT);
3292 	inode_lock_nested(inode2, I_MUTEX_CHILD);
3293 }
3294 
3295 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
3296 				      struct inode *inode2, u64 loff2, u64 len)
3297 {
3298 	unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
3299 	unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
3300 }
3301 
3302 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
3303 				    struct inode *inode2, u64 loff2, u64 len,
3304 				    bool retry_range_locking)
3305 {
3306 	int ret;
3307 
3308 	if (inode1 < inode2) {
3309 		swap(inode1, inode2);
3310 		swap(loff1, loff2);
3311 	}
3312 	ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
3313 	if (ret)
3314 		return ret;
3315 	ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
3316 	if (ret)
3317 		unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
3318 			      loff1 + len - 1);
3319 	return ret;
3320 }
3321 
3322 struct cmp_pages {
3323 	int		num_pages;
3324 	struct page	**src_pages;
3325 	struct page	**dst_pages;
3326 };
3327 
3328 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
3329 {
3330 	int i;
3331 	struct page *pg;
3332 
3333 	for (i = 0; i < cmp->num_pages; i++) {
3334 		pg = cmp->src_pages[i];
3335 		if (pg) {
3336 			unlock_page(pg);
3337 			put_page(pg);
3338 			cmp->src_pages[i] = NULL;
3339 		}
3340 		pg = cmp->dst_pages[i];
3341 		if (pg) {
3342 			unlock_page(pg);
3343 			put_page(pg);
3344 			cmp->dst_pages[i] = NULL;
3345 		}
3346 	}
3347 }
3348 
3349 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
3350 				  struct inode *dst, u64 dst_loff,
3351 				  u64 len, struct cmp_pages *cmp)
3352 {
3353 	int ret;
3354 	int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT;
3355 
3356 	cmp->num_pages = num_pages;
3357 
3358 	ret = gather_extent_pages(src, cmp->src_pages, num_pages, loff);
3359 	if (ret)
3360 		goto out;
3361 
3362 	ret = gather_extent_pages(dst, cmp->dst_pages, num_pages, dst_loff);
3363 
3364 out:
3365 	if (ret)
3366 		btrfs_cmp_data_free(cmp);
3367 	return ret;
3368 }
3369 
3370 static int btrfs_cmp_data(u64 len, struct cmp_pages *cmp)
3371 {
3372 	int ret = 0;
3373 	int i;
3374 	struct page *src_page, *dst_page;
3375 	unsigned int cmp_len = PAGE_SIZE;
3376 	void *addr, *dst_addr;
3377 
3378 	i = 0;
3379 	while (len) {
3380 		if (len < PAGE_SIZE)
3381 			cmp_len = len;
3382 
3383 		BUG_ON(i >= cmp->num_pages);
3384 
3385 		src_page = cmp->src_pages[i];
3386 		dst_page = cmp->dst_pages[i];
3387 		ASSERT(PageLocked(src_page));
3388 		ASSERT(PageLocked(dst_page));
3389 
3390 		addr = kmap_atomic(src_page);
3391 		dst_addr = kmap_atomic(dst_page);
3392 
3393 		flush_dcache_page(src_page);
3394 		flush_dcache_page(dst_page);
3395 
3396 		if (memcmp(addr, dst_addr, cmp_len))
3397 			ret = -EBADE;
3398 
3399 		kunmap_atomic(addr);
3400 		kunmap_atomic(dst_addr);
3401 
3402 		if (ret)
3403 			break;
3404 
3405 		len -= cmp_len;
3406 		i++;
3407 	}
3408 
3409 	return ret;
3410 }
3411 
3412 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3413 				     u64 olen)
3414 {
3415 	u64 len = *plen;
3416 	u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3417 
3418 	if (off + olen > inode->i_size || off + olen < off)
3419 		return -EINVAL;
3420 
3421 	/* if we extend to eof, continue to block boundary */
3422 	if (off + len == inode->i_size)
3423 		*plen = len = ALIGN(inode->i_size, bs) - off;
3424 
3425 	/* Check that we are block aligned - btrfs_clone() requires this */
3426 	if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3427 		return -EINVAL;
3428 
3429 	return 0;
3430 }
3431 
3432 static int btrfs_extent_same_range(struct inode *src, u64 loff, u64 olen,
3433 				   struct inode *dst, u64 dst_loff,
3434 				   struct cmp_pages *cmp)
3435 {
3436 	int ret;
3437 	u64 len = olen;
3438 	bool same_inode = (src == dst);
3439 	u64 same_lock_start = 0;
3440 	u64 same_lock_len = 0;
3441 
3442 	ret = extent_same_check_offsets(src, loff, &len, olen);
3443 	if (ret)
3444 		return ret;
3445 
3446 	ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3447 	if (ret)
3448 		return ret;
3449 
3450 	if (same_inode) {
3451 		/*
3452 		 * Single inode case wants the same checks, except we
3453 		 * don't want our length pushed out past i_size as
3454 		 * comparing that data range makes no sense.
3455 		 *
3456 		 * extent_same_check_offsets() will do this for an
3457 		 * unaligned length at i_size, so catch it here and
3458 		 * reject the request.
3459 		 *
3460 		 * This effectively means we require aligned extents
3461 		 * for the single-inode case, whereas the other cases
3462 		 * allow an unaligned length so long as it ends at
3463 		 * i_size.
3464 		 */
3465 		if (len != olen)
3466 			return -EINVAL;
3467 
3468 		/* Check for overlapping ranges */
3469 		if (dst_loff + len > loff && dst_loff < loff + len)
3470 			return -EINVAL;
3471 
3472 		same_lock_start = min_t(u64, loff, dst_loff);
3473 		same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3474 	} else {
3475 		/*
3476 		 * If the source and destination inodes are different, the
3477 		 * source's range end offset matches the source's i_size, that
3478 		 * i_size is not a multiple of the sector size, and the
3479 		 * destination range does not go past the destination's i_size,
3480 		 * we must round down the length to the nearest sector size
3481 		 * multiple. If we don't do this adjustment we end replacing
3482 		 * with zeroes the bytes in the range that starts at the
3483 		 * deduplication range's end offset and ends at the next sector
3484 		 * size multiple.
3485 		 */
3486 		if (loff + olen == i_size_read(src) &&
3487 		    dst_loff + len < i_size_read(dst)) {
3488 			const u64 sz = BTRFS_I(src)->root->fs_info->sectorsize;
3489 
3490 			len = round_down(i_size_read(src), sz) - loff;
3491 			if (len == 0)
3492 				return 0;
3493 			olen = len;
3494 		}
3495 	}
3496 
3497 again:
3498 	ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, cmp);
3499 	if (ret)
3500 		return ret;
3501 
3502 	if (same_inode)
3503 		ret = lock_extent_range(src, same_lock_start, same_lock_len,
3504 					false);
3505 	else
3506 		ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3507 					       false);
3508 	/*
3509 	 * If one of the inodes has dirty pages in the respective range or
3510 	 * ordered extents, we need to flush dellaloc and wait for all ordered
3511 	 * extents in the range. We must unlock the pages and the ranges in the
3512 	 * io trees to avoid deadlocks when flushing delalloc (requires locking
3513 	 * pages) and when waiting for ordered extents to complete (they require
3514 	 * range locking).
3515 	 */
3516 	if (ret == -EAGAIN) {
3517 		/*
3518 		 * Ranges in the io trees already unlocked. Now unlock all
3519 		 * pages before waiting for all IO to complete.
3520 		 */
3521 		btrfs_cmp_data_free(cmp);
3522 		if (same_inode) {
3523 			btrfs_wait_ordered_range(src, same_lock_start,
3524 						 same_lock_len);
3525 		} else {
3526 			btrfs_wait_ordered_range(src, loff, len);
3527 			btrfs_wait_ordered_range(dst, dst_loff, len);
3528 		}
3529 		goto again;
3530 	}
3531 	ASSERT(ret == 0);
3532 	if (WARN_ON(ret)) {
3533 		/* ranges in the io trees already unlocked */
3534 		btrfs_cmp_data_free(cmp);
3535 		return ret;
3536 	}
3537 
3538 	/* pass original length for comparison so we stay within i_size */
3539 	ret = btrfs_cmp_data(olen, cmp);
3540 	if (ret == 0)
3541 		ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3542 
3543 	if (same_inode)
3544 		unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3545 			      same_lock_start + same_lock_len - 1);
3546 	else
3547 		btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3548 
3549 	btrfs_cmp_data_free(cmp);
3550 
3551 	return ret;
3552 }
3553 
3554 #define BTRFS_MAX_DEDUPE_LEN	SZ_16M
3555 
3556 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3557 			     struct inode *dst, u64 dst_loff)
3558 {
3559 	int ret;
3560 	struct cmp_pages cmp;
3561 	int num_pages = PAGE_ALIGN(BTRFS_MAX_DEDUPE_LEN) >> PAGE_SHIFT;
3562 	bool same_inode = (src == dst);
3563 	u64 i, tail_len, chunk_count;
3564 
3565 	if (olen == 0)
3566 		return 0;
3567 
3568 	if (same_inode)
3569 		inode_lock(src);
3570 	else
3571 		btrfs_double_inode_lock(src, dst);
3572 
3573 	/* don't make the dst file partly checksummed */
3574 	if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3575 	    (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3576 		ret = -EINVAL;
3577 		goto out_unlock;
3578 	}
3579 
3580 	tail_len = olen % BTRFS_MAX_DEDUPE_LEN;
3581 	chunk_count = div_u64(olen, BTRFS_MAX_DEDUPE_LEN);
3582 	if (chunk_count == 0)
3583 		num_pages = PAGE_ALIGN(tail_len) >> PAGE_SHIFT;
3584 
3585 	/*
3586 	 * If deduping ranges in the same inode, locking rules make it
3587 	 * mandatory to always lock pages in ascending order to avoid deadlocks
3588 	 * with concurrent tasks (such as starting writeback/delalloc).
3589 	 */
3590 	if (same_inode && dst_loff < loff)
3591 		swap(loff, dst_loff);
3592 
3593 	/*
3594 	 * We must gather up all the pages before we initiate our extent
3595 	 * locking. We use an array for the page pointers. Size of the array is
3596 	 * bounded by len, which is in turn bounded by BTRFS_MAX_DEDUPE_LEN.
3597 	 */
3598 	cmp.src_pages = kvmalloc_array(num_pages, sizeof(struct page *),
3599 				       GFP_KERNEL | __GFP_ZERO);
3600 	cmp.dst_pages = kvmalloc_array(num_pages, sizeof(struct page *),
3601 				       GFP_KERNEL | __GFP_ZERO);
3602 	if (!cmp.src_pages || !cmp.dst_pages) {
3603 		ret = -ENOMEM;
3604 		goto out_free;
3605 	}
3606 
3607 	for (i = 0; i < chunk_count; i++) {
3608 		ret = btrfs_extent_same_range(src, loff, BTRFS_MAX_DEDUPE_LEN,
3609 					      dst, dst_loff, &cmp);
3610 		if (ret)
3611 			goto out_free;
3612 
3613 		loff += BTRFS_MAX_DEDUPE_LEN;
3614 		dst_loff += BTRFS_MAX_DEDUPE_LEN;
3615 	}
3616 
3617 	if (tail_len > 0)
3618 		ret = btrfs_extent_same_range(src, loff, tail_len, dst,
3619 					      dst_loff, &cmp);
3620 
3621 out_free:
3622 	kvfree(cmp.src_pages);
3623 	kvfree(cmp.dst_pages);
3624 
3625 out_unlock:
3626 	if (same_inode)
3627 		inode_unlock(src);
3628 	else
3629 		btrfs_double_inode_unlock(src, dst);
3630 
3631 	return ret;
3632 }
3633 
3634 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3635 				     struct inode *inode,
3636 				     u64 endoff,
3637 				     const u64 destoff,
3638 				     const u64 olen,
3639 				     int no_time_update)
3640 {
3641 	struct btrfs_root *root = BTRFS_I(inode)->root;
3642 	int ret;
3643 
3644 	inode_inc_iversion(inode);
3645 	if (!no_time_update)
3646 		inode->i_mtime = inode->i_ctime = current_time(inode);
3647 	/*
3648 	 * We round up to the block size at eof when determining which
3649 	 * extents to clone above, but shouldn't round up the file size.
3650 	 */
3651 	if (endoff > destoff + olen)
3652 		endoff = destoff + olen;
3653 	if (endoff > inode->i_size)
3654 		btrfs_i_size_write(BTRFS_I(inode), endoff);
3655 
3656 	ret = btrfs_update_inode(trans, root, inode);
3657 	if (ret) {
3658 		btrfs_abort_transaction(trans, ret);
3659 		btrfs_end_transaction(trans);
3660 		goto out;
3661 	}
3662 	ret = btrfs_end_transaction(trans);
3663 out:
3664 	return ret;
3665 }
3666 
3667 static void clone_update_extent_map(struct btrfs_inode *inode,
3668 				    const struct btrfs_trans_handle *trans,
3669 				    const struct btrfs_path *path,
3670 				    const u64 hole_offset,
3671 				    const u64 hole_len)
3672 {
3673 	struct extent_map_tree *em_tree = &inode->extent_tree;
3674 	struct extent_map *em;
3675 	int ret;
3676 
3677 	em = alloc_extent_map();
3678 	if (!em) {
3679 		set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3680 		return;
3681 	}
3682 
3683 	if (path) {
3684 		struct btrfs_file_extent_item *fi;
3685 
3686 		fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3687 				    struct btrfs_file_extent_item);
3688 		btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3689 		em->generation = -1;
3690 		if (btrfs_file_extent_type(path->nodes[0], fi) ==
3691 		    BTRFS_FILE_EXTENT_INLINE)
3692 			set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3693 					&inode->runtime_flags);
3694 	} else {
3695 		em->start = hole_offset;
3696 		em->len = hole_len;
3697 		em->ram_bytes = em->len;
3698 		em->orig_start = hole_offset;
3699 		em->block_start = EXTENT_MAP_HOLE;
3700 		em->block_len = 0;
3701 		em->orig_block_len = 0;
3702 		em->compress_type = BTRFS_COMPRESS_NONE;
3703 		em->generation = trans->transid;
3704 	}
3705 
3706 	while (1) {
3707 		write_lock(&em_tree->lock);
3708 		ret = add_extent_mapping(em_tree, em, 1);
3709 		write_unlock(&em_tree->lock);
3710 		if (ret != -EEXIST) {
3711 			free_extent_map(em);
3712 			break;
3713 		}
3714 		btrfs_drop_extent_cache(inode, em->start,
3715 					em->start + em->len - 1, 0);
3716 	}
3717 
3718 	if (ret)
3719 		set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3720 }
3721 
3722 /*
3723  * Make sure we do not end up inserting an inline extent into a file that has
3724  * already other (non-inline) extents. If a file has an inline extent it can
3725  * not have any other extents and the (single) inline extent must start at the
3726  * file offset 0. Failing to respect these rules will lead to file corruption,
3727  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3728  *
3729  * We can have extents that have been already written to disk or we can have
3730  * dirty ranges still in delalloc, in which case the extent maps and items are
3731  * created only when we run delalloc, and the delalloc ranges might fall outside
3732  * the range we are currently locking in the inode's io tree. So we check the
3733  * inode's i_size because of that (i_size updates are done while holding the
3734  * i_mutex, which we are holding here).
3735  * We also check to see if the inode has a size not greater than "datal" but has
3736  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3737  * protected against such concurrent fallocate calls by the i_mutex).
3738  *
3739  * If the file has no extents but a size greater than datal, do not allow the
3740  * copy because we would need turn the inline extent into a non-inline one (even
3741  * with NO_HOLES enabled). If we find our destination inode only has one inline
3742  * extent, just overwrite it with the source inline extent if its size is less
3743  * than the source extent's size, or we could copy the source inline extent's
3744  * data into the destination inode's inline extent if the later is greater then
3745  * the former.
3746  */
3747 static int clone_copy_inline_extent(struct inode *dst,
3748 				    struct btrfs_trans_handle *trans,
3749 				    struct btrfs_path *path,
3750 				    struct btrfs_key *new_key,
3751 				    const u64 drop_start,
3752 				    const u64 datal,
3753 				    const u64 skip,
3754 				    const u64 size,
3755 				    char *inline_data)
3756 {
3757 	struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb);
3758 	struct btrfs_root *root = BTRFS_I(dst)->root;
3759 	const u64 aligned_end = ALIGN(new_key->offset + datal,
3760 				      fs_info->sectorsize);
3761 	int ret;
3762 	struct btrfs_key key;
3763 
3764 	if (new_key->offset > 0)
3765 		return -EOPNOTSUPP;
3766 
3767 	key.objectid = btrfs_ino(BTRFS_I(dst));
3768 	key.type = BTRFS_EXTENT_DATA_KEY;
3769 	key.offset = 0;
3770 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3771 	if (ret < 0) {
3772 		return ret;
3773 	} else if (ret > 0) {
3774 		if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3775 			ret = btrfs_next_leaf(root, path);
3776 			if (ret < 0)
3777 				return ret;
3778 			else if (ret > 0)
3779 				goto copy_inline_extent;
3780 		}
3781 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3782 		if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3783 		    key.type == BTRFS_EXTENT_DATA_KEY) {
3784 			ASSERT(key.offset > 0);
3785 			return -EOPNOTSUPP;
3786 		}
3787 	} else if (i_size_read(dst) <= datal) {
3788 		struct btrfs_file_extent_item *ei;
3789 		u64 ext_len;
3790 
3791 		/*
3792 		 * If the file size is <= datal, make sure there are no other
3793 		 * extents following (can happen do to an fallocate call with
3794 		 * the flag FALLOC_FL_KEEP_SIZE).
3795 		 */
3796 		ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3797 				    struct btrfs_file_extent_item);
3798 		/*
3799 		 * If it's an inline extent, it can not have other extents
3800 		 * following it.
3801 		 */
3802 		if (btrfs_file_extent_type(path->nodes[0], ei) ==
3803 		    BTRFS_FILE_EXTENT_INLINE)
3804 			goto copy_inline_extent;
3805 
3806 		ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3807 		if (ext_len > aligned_end)
3808 			return -EOPNOTSUPP;
3809 
3810 		ret = btrfs_next_item(root, path);
3811 		if (ret < 0) {
3812 			return ret;
3813 		} else if (ret == 0) {
3814 			btrfs_item_key_to_cpu(path->nodes[0], &key,
3815 					      path->slots[0]);
3816 			if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3817 			    key.type == BTRFS_EXTENT_DATA_KEY)
3818 				return -EOPNOTSUPP;
3819 		}
3820 	}
3821 
3822 copy_inline_extent:
3823 	/*
3824 	 * We have no extent items, or we have an extent at offset 0 which may
3825 	 * or may not be inlined. All these cases are dealt the same way.
3826 	 */
3827 	if (i_size_read(dst) > datal) {
3828 		/*
3829 		 * If the destination inode has an inline extent...
3830 		 * This would require copying the data from the source inline
3831 		 * extent into the beginning of the destination's inline extent.
3832 		 * But this is really complex, both extents can be compressed
3833 		 * or just one of them, which would require decompressing and
3834 		 * re-compressing data (which could increase the new compressed
3835 		 * size, not allowing the compressed data to fit anymore in an
3836 		 * inline extent).
3837 		 * So just don't support this case for now (it should be rare,
3838 		 * we are not really saving space when cloning inline extents).
3839 		 */
3840 		return -EOPNOTSUPP;
3841 	}
3842 
3843 	btrfs_release_path(path);
3844 	ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3845 	if (ret)
3846 		return ret;
3847 	ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3848 	if (ret)
3849 		return ret;
3850 
3851 	if (skip) {
3852 		const u32 start = btrfs_file_extent_calc_inline_size(0);
3853 
3854 		memmove(inline_data + start, inline_data + start + skip, datal);
3855 	}
3856 
3857 	write_extent_buffer(path->nodes[0], inline_data,
3858 			    btrfs_item_ptr_offset(path->nodes[0],
3859 						  path->slots[0]),
3860 			    size);
3861 	inode_add_bytes(dst, datal);
3862 
3863 	return 0;
3864 }
3865 
3866 /**
3867  * btrfs_clone() - clone a range from inode file to another
3868  *
3869  * @src: Inode to clone from
3870  * @inode: Inode to clone to
3871  * @off: Offset within source to start clone from
3872  * @olen: Original length, passed by user, of range to clone
3873  * @olen_aligned: Block-aligned value of olen
3874  * @destoff: Offset within @inode to start clone
3875  * @no_time_update: Whether to update mtime/ctime on the target inode
3876  */
3877 static int btrfs_clone(struct inode *src, struct inode *inode,
3878 		       const u64 off, const u64 olen, const u64 olen_aligned,
3879 		       const u64 destoff, int no_time_update)
3880 {
3881 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3882 	struct btrfs_root *root = BTRFS_I(inode)->root;
3883 	struct btrfs_path *path = NULL;
3884 	struct extent_buffer *leaf;
3885 	struct btrfs_trans_handle *trans;
3886 	char *buf = NULL;
3887 	struct btrfs_key key;
3888 	u32 nritems;
3889 	int slot;
3890 	int ret;
3891 	const u64 len = olen_aligned;
3892 	u64 last_dest_end = destoff;
3893 
3894 	ret = -ENOMEM;
3895 	buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
3896 	if (!buf)
3897 		return ret;
3898 
3899 	path = btrfs_alloc_path();
3900 	if (!path) {
3901 		kvfree(buf);
3902 		return ret;
3903 	}
3904 
3905 	path->reada = READA_FORWARD;
3906 	/* clone data */
3907 	key.objectid = btrfs_ino(BTRFS_I(src));
3908 	key.type = BTRFS_EXTENT_DATA_KEY;
3909 	key.offset = off;
3910 
3911 	while (1) {
3912 		u64 next_key_min_offset = key.offset + 1;
3913 
3914 		/*
3915 		 * note the key will change type as we walk through the
3916 		 * tree.
3917 		 */
3918 		path->leave_spinning = 1;
3919 		ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3920 				0, 0);
3921 		if (ret < 0)
3922 			goto out;
3923 		/*
3924 		 * First search, if no extent item that starts at offset off was
3925 		 * found but the previous item is an extent item, it's possible
3926 		 * it might overlap our target range, therefore process it.
3927 		 */
3928 		if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3929 			btrfs_item_key_to_cpu(path->nodes[0], &key,
3930 					      path->slots[0] - 1);
3931 			if (key.type == BTRFS_EXTENT_DATA_KEY)
3932 				path->slots[0]--;
3933 		}
3934 
3935 		nritems = btrfs_header_nritems(path->nodes[0]);
3936 process_slot:
3937 		if (path->slots[0] >= nritems) {
3938 			ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3939 			if (ret < 0)
3940 				goto out;
3941 			if (ret > 0)
3942 				break;
3943 			nritems = btrfs_header_nritems(path->nodes[0]);
3944 		}
3945 		leaf = path->nodes[0];
3946 		slot = path->slots[0];
3947 
3948 		btrfs_item_key_to_cpu(leaf, &key, slot);
3949 		if (key.type > BTRFS_EXTENT_DATA_KEY ||
3950 		    key.objectid != btrfs_ino(BTRFS_I(src)))
3951 			break;
3952 
3953 		if (key.type == BTRFS_EXTENT_DATA_KEY) {
3954 			struct btrfs_file_extent_item *extent;
3955 			int type;
3956 			u32 size;
3957 			struct btrfs_key new_key;
3958 			u64 disko = 0, diskl = 0;
3959 			u64 datao = 0, datal = 0;
3960 			u8 comp;
3961 			u64 drop_start;
3962 
3963 			extent = btrfs_item_ptr(leaf, slot,
3964 						struct btrfs_file_extent_item);
3965 			comp = btrfs_file_extent_compression(leaf, extent);
3966 			type = btrfs_file_extent_type(leaf, extent);
3967 			if (type == BTRFS_FILE_EXTENT_REG ||
3968 			    type == BTRFS_FILE_EXTENT_PREALLOC) {
3969 				disko = btrfs_file_extent_disk_bytenr(leaf,
3970 								      extent);
3971 				diskl = btrfs_file_extent_disk_num_bytes(leaf,
3972 								 extent);
3973 				datao = btrfs_file_extent_offset(leaf, extent);
3974 				datal = btrfs_file_extent_num_bytes(leaf,
3975 								    extent);
3976 			} else if (type == BTRFS_FILE_EXTENT_INLINE) {
3977 				/* take upper bound, may be compressed */
3978 				datal = btrfs_file_extent_ram_bytes(leaf,
3979 								    extent);
3980 			}
3981 
3982 			/*
3983 			 * The first search might have left us at an extent
3984 			 * item that ends before our target range's start, can
3985 			 * happen if we have holes and NO_HOLES feature enabled.
3986 			 */
3987 			if (key.offset + datal <= off) {
3988 				path->slots[0]++;
3989 				goto process_slot;
3990 			} else if (key.offset >= off + len) {
3991 				break;
3992 			}
3993 			next_key_min_offset = key.offset + datal;
3994 			size = btrfs_item_size_nr(leaf, slot);
3995 			read_extent_buffer(leaf, buf,
3996 					   btrfs_item_ptr_offset(leaf, slot),
3997 					   size);
3998 
3999 			btrfs_release_path(path);
4000 			path->leave_spinning = 0;
4001 
4002 			memcpy(&new_key, &key, sizeof(new_key));
4003 			new_key.objectid = btrfs_ino(BTRFS_I(inode));
4004 			if (off <= key.offset)
4005 				new_key.offset = key.offset + destoff - off;
4006 			else
4007 				new_key.offset = destoff;
4008 
4009 			/*
4010 			 * Deal with a hole that doesn't have an extent item
4011 			 * that represents it (NO_HOLES feature enabled).
4012 			 * This hole is either in the middle of the cloning
4013 			 * range or at the beginning (fully overlaps it or
4014 			 * partially overlaps it).
4015 			 */
4016 			if (new_key.offset != last_dest_end)
4017 				drop_start = last_dest_end;
4018 			else
4019 				drop_start = new_key.offset;
4020 
4021 			/*
4022 			 * 1 - adjusting old extent (we may have to split it)
4023 			 * 1 - add new extent
4024 			 * 1 - inode update
4025 			 */
4026 			trans = btrfs_start_transaction(root, 3);
4027 			if (IS_ERR(trans)) {
4028 				ret = PTR_ERR(trans);
4029 				goto out;
4030 			}
4031 
4032 			if (type == BTRFS_FILE_EXTENT_REG ||
4033 			    type == BTRFS_FILE_EXTENT_PREALLOC) {
4034 				/*
4035 				 *    a  | --- range to clone ---|  b
4036 				 * | ------------- extent ------------- |
4037 				 */
4038 
4039 				/* subtract range b */
4040 				if (key.offset + datal > off + len)
4041 					datal = off + len - key.offset;
4042 
4043 				/* subtract range a */
4044 				if (off > key.offset) {
4045 					datao += off - key.offset;
4046 					datal -= off - key.offset;
4047 				}
4048 
4049 				ret = btrfs_drop_extents(trans, root, inode,
4050 							 drop_start,
4051 							 new_key.offset + datal,
4052 							 1);
4053 				if (ret) {
4054 					if (ret != -EOPNOTSUPP)
4055 						btrfs_abort_transaction(trans,
4056 									ret);
4057 					btrfs_end_transaction(trans);
4058 					goto out;
4059 				}
4060 
4061 				ret = btrfs_insert_empty_item(trans, root, path,
4062 							      &new_key, size);
4063 				if (ret) {
4064 					btrfs_abort_transaction(trans, ret);
4065 					btrfs_end_transaction(trans);
4066 					goto out;
4067 				}
4068 
4069 				leaf = path->nodes[0];
4070 				slot = path->slots[0];
4071 				write_extent_buffer(leaf, buf,
4072 					    btrfs_item_ptr_offset(leaf, slot),
4073 					    size);
4074 
4075 				extent = btrfs_item_ptr(leaf, slot,
4076 						struct btrfs_file_extent_item);
4077 
4078 				/* disko == 0 means it's a hole */
4079 				if (!disko)
4080 					datao = 0;
4081 
4082 				btrfs_set_file_extent_offset(leaf, extent,
4083 							     datao);
4084 				btrfs_set_file_extent_num_bytes(leaf, extent,
4085 								datal);
4086 
4087 				if (disko) {
4088 					inode_add_bytes(inode, datal);
4089 					ret = btrfs_inc_extent_ref(trans,
4090 							root,
4091 							disko, diskl, 0,
4092 							root->root_key.objectid,
4093 							btrfs_ino(BTRFS_I(inode)),
4094 							new_key.offset - datao);
4095 					if (ret) {
4096 						btrfs_abort_transaction(trans,
4097 									ret);
4098 						btrfs_end_transaction(trans);
4099 						goto out;
4100 
4101 					}
4102 				}
4103 			} else if (type == BTRFS_FILE_EXTENT_INLINE) {
4104 				u64 skip = 0;
4105 				u64 trim = 0;
4106 
4107 				if (off > key.offset) {
4108 					skip = off - key.offset;
4109 					new_key.offset += skip;
4110 				}
4111 
4112 				if (key.offset + datal > off + len)
4113 					trim = key.offset + datal - (off + len);
4114 
4115 				if (comp && (skip || trim)) {
4116 					ret = -EINVAL;
4117 					btrfs_end_transaction(trans);
4118 					goto out;
4119 				}
4120 				size -= skip + trim;
4121 				datal -= skip + trim;
4122 
4123 				ret = clone_copy_inline_extent(inode,
4124 							       trans, path,
4125 							       &new_key,
4126 							       drop_start,
4127 							       datal,
4128 							       skip, size, buf);
4129 				if (ret) {
4130 					if (ret != -EOPNOTSUPP)
4131 						btrfs_abort_transaction(trans,
4132 									ret);
4133 					btrfs_end_transaction(trans);
4134 					goto out;
4135 				}
4136 				leaf = path->nodes[0];
4137 				slot = path->slots[0];
4138 			}
4139 
4140 			/* If we have an implicit hole (NO_HOLES feature). */
4141 			if (drop_start < new_key.offset)
4142 				clone_update_extent_map(BTRFS_I(inode), trans,
4143 						NULL, drop_start,
4144 						new_key.offset - drop_start);
4145 
4146 			clone_update_extent_map(BTRFS_I(inode), trans,
4147 					path, 0, 0);
4148 
4149 			btrfs_mark_buffer_dirty(leaf);
4150 			btrfs_release_path(path);
4151 
4152 			last_dest_end = ALIGN(new_key.offset + datal,
4153 					      fs_info->sectorsize);
4154 			ret = clone_finish_inode_update(trans, inode,
4155 							last_dest_end,
4156 							destoff, olen,
4157 							no_time_update);
4158 			if (ret)
4159 				goto out;
4160 			if (new_key.offset + datal >= destoff + len)
4161 				break;
4162 		}
4163 		btrfs_release_path(path);
4164 		key.offset = next_key_min_offset;
4165 
4166 		if (fatal_signal_pending(current)) {
4167 			ret = -EINTR;
4168 			goto out;
4169 		}
4170 	}
4171 	ret = 0;
4172 
4173 	if (last_dest_end < destoff + len) {
4174 		/*
4175 		 * We have an implicit hole (NO_HOLES feature is enabled) that
4176 		 * fully or partially overlaps our cloning range at its end.
4177 		 */
4178 		btrfs_release_path(path);
4179 
4180 		/*
4181 		 * 1 - remove extent(s)
4182 		 * 1 - inode update
4183 		 */
4184 		trans = btrfs_start_transaction(root, 2);
4185 		if (IS_ERR(trans)) {
4186 			ret = PTR_ERR(trans);
4187 			goto out;
4188 		}
4189 		ret = btrfs_drop_extents(trans, root, inode,
4190 					 last_dest_end, destoff + len, 1);
4191 		if (ret) {
4192 			if (ret != -EOPNOTSUPP)
4193 				btrfs_abort_transaction(trans, ret);
4194 			btrfs_end_transaction(trans);
4195 			goto out;
4196 		}
4197 		clone_update_extent_map(BTRFS_I(inode), trans, NULL,
4198 				last_dest_end,
4199 				destoff + len - last_dest_end);
4200 		ret = clone_finish_inode_update(trans, inode, destoff + len,
4201 						destoff, olen, no_time_update);
4202 	}
4203 
4204 out:
4205 	btrfs_free_path(path);
4206 	kvfree(buf);
4207 	return ret;
4208 }
4209 
4210 static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
4211 					u64 off, u64 olen, u64 destoff)
4212 {
4213 	struct inode *inode = file_inode(file);
4214 	struct inode *src = file_inode(file_src);
4215 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4216 	struct btrfs_root *root = BTRFS_I(inode)->root;
4217 	int ret;
4218 	u64 len = olen;
4219 	u64 bs = fs_info->sb->s_blocksize;
4220 	int same_inode = src == inode;
4221 
4222 	/*
4223 	 * TODO:
4224 	 * - split compressed inline extents.  annoying: we need to
4225 	 *   decompress into destination's address_space (the file offset
4226 	 *   may change, so source mapping won't do), then recompress (or
4227 	 *   otherwise reinsert) a subrange.
4228 	 *
4229 	 * - split destination inode's inline extents.  The inline extents can
4230 	 *   be either compressed or non-compressed.
4231 	 */
4232 
4233 	if (btrfs_root_readonly(root))
4234 		return -EROFS;
4235 
4236 	if (file_src->f_path.mnt != file->f_path.mnt ||
4237 	    src->i_sb != inode->i_sb)
4238 		return -EXDEV;
4239 
4240 	if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
4241 		return -EISDIR;
4242 
4243 	if (!same_inode) {
4244 		btrfs_double_inode_lock(src, inode);
4245 	} else {
4246 		inode_lock(src);
4247 	}
4248 
4249 	/* don't make the dst file partly checksummed */
4250 	if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
4251 	    (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
4252 		ret = -EINVAL;
4253 		goto out_unlock;
4254 	}
4255 
4256 	/* determine range to clone */
4257 	ret = -EINVAL;
4258 	if (off + len > src->i_size || off + len < off)
4259 		goto out_unlock;
4260 	if (len == 0)
4261 		olen = len = src->i_size - off;
4262 	/*
4263 	 * If we extend to eof, continue to block boundary if and only if the
4264 	 * destination end offset matches the destination file's size, otherwise
4265 	 * we would be corrupting data by placing the eof block into the middle
4266 	 * of a file.
4267 	 */
4268 	if (off + len == src->i_size) {
4269 		if (!IS_ALIGNED(len, bs) && destoff + len < inode->i_size)
4270 			goto out_unlock;
4271 		len = ALIGN(src->i_size, bs) - off;
4272 	}
4273 
4274 	if (len == 0) {
4275 		ret = 0;
4276 		goto out_unlock;
4277 	}
4278 
4279 	/* verify the end result is block aligned */
4280 	if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
4281 	    !IS_ALIGNED(destoff, bs))
4282 		goto out_unlock;
4283 
4284 	/* verify if ranges are overlapped within the same file */
4285 	if (same_inode) {
4286 		if (destoff + len > off && destoff < off + len)
4287 			goto out_unlock;
4288 	}
4289 
4290 	if (destoff > inode->i_size) {
4291 		ret = btrfs_cont_expand(inode, inode->i_size, destoff);
4292 		if (ret)
4293 			goto out_unlock;
4294 	}
4295 
4296 	/*
4297 	 * Lock the target range too. Right after we replace the file extent
4298 	 * items in the fs tree (which now point to the cloned data), we might
4299 	 * have a worker replace them with extent items relative to a write
4300 	 * operation that was issued before this clone operation (i.e. confront
4301 	 * with inode.c:btrfs_finish_ordered_io).
4302 	 */
4303 	if (same_inode) {
4304 		u64 lock_start = min_t(u64, off, destoff);
4305 		u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
4306 
4307 		ret = lock_extent_range(src, lock_start, lock_len, true);
4308 	} else {
4309 		ret = btrfs_double_extent_lock(src, off, inode, destoff, len,
4310 					       true);
4311 	}
4312 	ASSERT(ret == 0);
4313 	if (WARN_ON(ret)) {
4314 		/* ranges in the io trees already unlocked */
4315 		goto out_unlock;
4316 	}
4317 
4318 	ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
4319 
4320 	if (same_inode) {
4321 		u64 lock_start = min_t(u64, off, destoff);
4322 		u64 lock_end = max_t(u64, off, destoff) + len - 1;
4323 
4324 		unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
4325 	} else {
4326 		btrfs_double_extent_unlock(src, off, inode, destoff, len);
4327 	}
4328 	/*
4329 	 * Truncate page cache pages so that future reads will see the cloned
4330 	 * data immediately and not the previous data.
4331 	 */
4332 	truncate_inode_pages_range(&inode->i_data,
4333 				round_down(destoff, PAGE_SIZE),
4334 				round_up(destoff + len, PAGE_SIZE) - 1);
4335 out_unlock:
4336 	if (!same_inode)
4337 		btrfs_double_inode_unlock(src, inode);
4338 	else
4339 		inode_unlock(src);
4340 	return ret;
4341 }
4342 
4343 loff_t btrfs_remap_file_range(struct file *src_file, loff_t off,
4344 		struct file *dst_file, loff_t destoff, loff_t len,
4345 		unsigned int remap_flags)
4346 {
4347 	int ret;
4348 
4349 	if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
4350 		return -EINVAL;
4351 
4352 	if (remap_flags & REMAP_FILE_DEDUP) {
4353 		struct inode *src = file_inode(src_file);
4354 		struct inode *dst = file_inode(dst_file);
4355 		u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
4356 
4357 		if (WARN_ON_ONCE(bs < PAGE_SIZE)) {
4358 			/*
4359 			 * Btrfs does not support blocksize < page_size. As a
4360 			 * result, btrfs_cmp_data() won't correctly handle
4361 			 * this situation without an update.
4362 			 */
4363 			return -EINVAL;
4364 		}
4365 
4366 		ret = btrfs_extent_same(src, off, len, dst, destoff);
4367 	} else {
4368 		ret = btrfs_clone_files(dst_file, src_file, off, len, destoff);
4369 	}
4370 	return ret < 0 ? ret : len;
4371 }
4372 
4373 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
4374 {
4375 	struct inode *inode = file_inode(file);
4376 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4377 	struct btrfs_root *root = BTRFS_I(inode)->root;
4378 	struct btrfs_root *new_root;
4379 	struct btrfs_dir_item *di;
4380 	struct btrfs_trans_handle *trans;
4381 	struct btrfs_path *path;
4382 	struct btrfs_key location;
4383 	struct btrfs_disk_key disk_key;
4384 	u64 objectid = 0;
4385 	u64 dir_id;
4386 	int ret;
4387 
4388 	if (!capable(CAP_SYS_ADMIN))
4389 		return -EPERM;
4390 
4391 	ret = mnt_want_write_file(file);
4392 	if (ret)
4393 		return ret;
4394 
4395 	if (copy_from_user(&objectid, argp, sizeof(objectid))) {
4396 		ret = -EFAULT;
4397 		goto out;
4398 	}
4399 
4400 	if (!objectid)
4401 		objectid = BTRFS_FS_TREE_OBJECTID;
4402 
4403 	location.objectid = objectid;
4404 	location.type = BTRFS_ROOT_ITEM_KEY;
4405 	location.offset = (u64)-1;
4406 
4407 	new_root = btrfs_read_fs_root_no_name(fs_info, &location);
4408 	if (IS_ERR(new_root)) {
4409 		ret = PTR_ERR(new_root);
4410 		goto out;
4411 	}
4412 	if (!is_fstree(new_root->root_key.objectid)) {
4413 		ret = -ENOENT;
4414 		goto out;
4415 	}
4416 
4417 	path = btrfs_alloc_path();
4418 	if (!path) {
4419 		ret = -ENOMEM;
4420 		goto out;
4421 	}
4422 	path->leave_spinning = 1;
4423 
4424 	trans = btrfs_start_transaction(root, 1);
4425 	if (IS_ERR(trans)) {
4426 		btrfs_free_path(path);
4427 		ret = PTR_ERR(trans);
4428 		goto out;
4429 	}
4430 
4431 	dir_id = btrfs_super_root_dir(fs_info->super_copy);
4432 	di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
4433 				   dir_id, "default", 7, 1);
4434 	if (IS_ERR_OR_NULL(di)) {
4435 		btrfs_free_path(path);
4436 		btrfs_end_transaction(trans);
4437 		btrfs_err(fs_info,
4438 			  "Umm, you don't have the default diritem, this isn't going to work");
4439 		ret = -ENOENT;
4440 		goto out;
4441 	}
4442 
4443 	btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
4444 	btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
4445 	btrfs_mark_buffer_dirty(path->nodes[0]);
4446 	btrfs_free_path(path);
4447 
4448 	btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
4449 	btrfs_end_transaction(trans);
4450 out:
4451 	mnt_drop_write_file(file);
4452 	return ret;
4453 }
4454 
4455 static void get_block_group_info(struct list_head *groups_list,
4456 				 struct btrfs_ioctl_space_info *space)
4457 {
4458 	struct btrfs_block_group_cache *block_group;
4459 
4460 	space->total_bytes = 0;
4461 	space->used_bytes = 0;
4462 	space->flags = 0;
4463 	list_for_each_entry(block_group, groups_list, list) {
4464 		space->flags = block_group->flags;
4465 		space->total_bytes += block_group->key.offset;
4466 		space->used_bytes +=
4467 			btrfs_block_group_used(&block_group->item);
4468 	}
4469 }
4470 
4471 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
4472 				   void __user *arg)
4473 {
4474 	struct btrfs_ioctl_space_args space_args;
4475 	struct btrfs_ioctl_space_info space;
4476 	struct btrfs_ioctl_space_info *dest;
4477 	struct btrfs_ioctl_space_info *dest_orig;
4478 	struct btrfs_ioctl_space_info __user *user_dest;
4479 	struct btrfs_space_info *info;
4480 	static const u64 types[] = {
4481 		BTRFS_BLOCK_GROUP_DATA,
4482 		BTRFS_BLOCK_GROUP_SYSTEM,
4483 		BTRFS_BLOCK_GROUP_METADATA,
4484 		BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
4485 	};
4486 	int num_types = 4;
4487 	int alloc_size;
4488 	int ret = 0;
4489 	u64 slot_count = 0;
4490 	int i, c;
4491 
4492 	if (copy_from_user(&space_args,
4493 			   (struct btrfs_ioctl_space_args __user *)arg,
4494 			   sizeof(space_args)))
4495 		return -EFAULT;
4496 
4497 	for (i = 0; i < num_types; i++) {
4498 		struct btrfs_space_info *tmp;
4499 
4500 		info = NULL;
4501 		rcu_read_lock();
4502 		list_for_each_entry_rcu(tmp, &fs_info->space_info,
4503 					list) {
4504 			if (tmp->flags == types[i]) {
4505 				info = tmp;
4506 				break;
4507 			}
4508 		}
4509 		rcu_read_unlock();
4510 
4511 		if (!info)
4512 			continue;
4513 
4514 		down_read(&info->groups_sem);
4515 		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4516 			if (!list_empty(&info->block_groups[c]))
4517 				slot_count++;
4518 		}
4519 		up_read(&info->groups_sem);
4520 	}
4521 
4522 	/*
4523 	 * Global block reserve, exported as a space_info
4524 	 */
4525 	slot_count++;
4526 
4527 	/* space_slots == 0 means they are asking for a count */
4528 	if (space_args.space_slots == 0) {
4529 		space_args.total_spaces = slot_count;
4530 		goto out;
4531 	}
4532 
4533 	slot_count = min_t(u64, space_args.space_slots, slot_count);
4534 
4535 	alloc_size = sizeof(*dest) * slot_count;
4536 
4537 	/* we generally have at most 6 or so space infos, one for each raid
4538 	 * level.  So, a whole page should be more than enough for everyone
4539 	 */
4540 	if (alloc_size > PAGE_SIZE)
4541 		return -ENOMEM;
4542 
4543 	space_args.total_spaces = 0;
4544 	dest = kmalloc(alloc_size, GFP_KERNEL);
4545 	if (!dest)
4546 		return -ENOMEM;
4547 	dest_orig = dest;
4548 
4549 	/* now we have a buffer to copy into */
4550 	for (i = 0; i < num_types; i++) {
4551 		struct btrfs_space_info *tmp;
4552 
4553 		if (!slot_count)
4554 			break;
4555 
4556 		info = NULL;
4557 		rcu_read_lock();
4558 		list_for_each_entry_rcu(tmp, &fs_info->space_info,
4559 					list) {
4560 			if (tmp->flags == types[i]) {
4561 				info = tmp;
4562 				break;
4563 			}
4564 		}
4565 		rcu_read_unlock();
4566 
4567 		if (!info)
4568 			continue;
4569 		down_read(&info->groups_sem);
4570 		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4571 			if (!list_empty(&info->block_groups[c])) {
4572 				get_block_group_info(&info->block_groups[c],
4573 						     &space);
4574 				memcpy(dest, &space, sizeof(space));
4575 				dest++;
4576 				space_args.total_spaces++;
4577 				slot_count--;
4578 			}
4579 			if (!slot_count)
4580 				break;
4581 		}
4582 		up_read(&info->groups_sem);
4583 	}
4584 
4585 	/*
4586 	 * Add global block reserve
4587 	 */
4588 	if (slot_count) {
4589 		struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4590 
4591 		spin_lock(&block_rsv->lock);
4592 		space.total_bytes = block_rsv->size;
4593 		space.used_bytes = block_rsv->size - block_rsv->reserved;
4594 		spin_unlock(&block_rsv->lock);
4595 		space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4596 		memcpy(dest, &space, sizeof(space));
4597 		space_args.total_spaces++;
4598 	}
4599 
4600 	user_dest = (struct btrfs_ioctl_space_info __user *)
4601 		(arg + sizeof(struct btrfs_ioctl_space_args));
4602 
4603 	if (copy_to_user(user_dest, dest_orig, alloc_size))
4604 		ret = -EFAULT;
4605 
4606 	kfree(dest_orig);
4607 out:
4608 	if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4609 		ret = -EFAULT;
4610 
4611 	return ret;
4612 }
4613 
4614 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4615 					    void __user *argp)
4616 {
4617 	struct btrfs_trans_handle *trans;
4618 	u64 transid;
4619 	int ret;
4620 
4621 	trans = btrfs_attach_transaction_barrier(root);
4622 	if (IS_ERR(trans)) {
4623 		if (PTR_ERR(trans) != -ENOENT)
4624 			return PTR_ERR(trans);
4625 
4626 		/* No running transaction, don't bother */
4627 		transid = root->fs_info->last_trans_committed;
4628 		goto out;
4629 	}
4630 	transid = trans->transid;
4631 	ret = btrfs_commit_transaction_async(trans, 0);
4632 	if (ret) {
4633 		btrfs_end_transaction(trans);
4634 		return ret;
4635 	}
4636 out:
4637 	if (argp)
4638 		if (copy_to_user(argp, &transid, sizeof(transid)))
4639 			return -EFAULT;
4640 	return 0;
4641 }
4642 
4643 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
4644 					   void __user *argp)
4645 {
4646 	u64 transid;
4647 
4648 	if (argp) {
4649 		if (copy_from_user(&transid, argp, sizeof(transid)))
4650 			return -EFAULT;
4651 	} else {
4652 		transid = 0;  /* current trans */
4653 	}
4654 	return btrfs_wait_for_commit(fs_info, transid);
4655 }
4656 
4657 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4658 {
4659 	struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
4660 	struct btrfs_ioctl_scrub_args *sa;
4661 	int ret;
4662 
4663 	if (!capable(CAP_SYS_ADMIN))
4664 		return -EPERM;
4665 
4666 	sa = memdup_user(arg, sizeof(*sa));
4667 	if (IS_ERR(sa))
4668 		return PTR_ERR(sa);
4669 
4670 	if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4671 		ret = mnt_want_write_file(file);
4672 		if (ret)
4673 			goto out;
4674 	}
4675 
4676 	ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
4677 			      &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4678 			      0);
4679 
4680 	if (copy_to_user(arg, sa, sizeof(*sa)))
4681 		ret = -EFAULT;
4682 
4683 	if (!(sa->flags & BTRFS_SCRUB_READONLY))
4684 		mnt_drop_write_file(file);
4685 out:
4686 	kfree(sa);
4687 	return ret;
4688 }
4689 
4690 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
4691 {
4692 	if (!capable(CAP_SYS_ADMIN))
4693 		return -EPERM;
4694 
4695 	return btrfs_scrub_cancel(fs_info);
4696 }
4697 
4698 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
4699 				       void __user *arg)
4700 {
4701 	struct btrfs_ioctl_scrub_args *sa;
4702 	int ret;
4703 
4704 	if (!capable(CAP_SYS_ADMIN))
4705 		return -EPERM;
4706 
4707 	sa = memdup_user(arg, sizeof(*sa));
4708 	if (IS_ERR(sa))
4709 		return PTR_ERR(sa);
4710 
4711 	ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
4712 
4713 	if (copy_to_user(arg, sa, sizeof(*sa)))
4714 		ret = -EFAULT;
4715 
4716 	kfree(sa);
4717 	return ret;
4718 }
4719 
4720 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
4721 				      void __user *arg)
4722 {
4723 	struct btrfs_ioctl_get_dev_stats *sa;
4724 	int ret;
4725 
4726 	sa = memdup_user(arg, sizeof(*sa));
4727 	if (IS_ERR(sa))
4728 		return PTR_ERR(sa);
4729 
4730 	if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4731 		kfree(sa);
4732 		return -EPERM;
4733 	}
4734 
4735 	ret = btrfs_get_dev_stats(fs_info, sa);
4736 
4737 	if (copy_to_user(arg, sa, sizeof(*sa)))
4738 		ret = -EFAULT;
4739 
4740 	kfree(sa);
4741 	return ret;
4742 }
4743 
4744 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
4745 				    void __user *arg)
4746 {
4747 	struct btrfs_ioctl_dev_replace_args *p;
4748 	int ret;
4749 
4750 	if (!capable(CAP_SYS_ADMIN))
4751 		return -EPERM;
4752 
4753 	p = memdup_user(arg, sizeof(*p));
4754 	if (IS_ERR(p))
4755 		return PTR_ERR(p);
4756 
4757 	switch (p->cmd) {
4758 	case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4759 		if (sb_rdonly(fs_info->sb)) {
4760 			ret = -EROFS;
4761 			goto out;
4762 		}
4763 		if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4764 			ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4765 		} else {
4766 			ret = btrfs_dev_replace_by_ioctl(fs_info, p);
4767 			clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4768 		}
4769 		break;
4770 	case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4771 		btrfs_dev_replace_status(fs_info, p);
4772 		ret = 0;
4773 		break;
4774 	case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4775 		p->result = btrfs_dev_replace_cancel(fs_info);
4776 		ret = 0;
4777 		break;
4778 	default:
4779 		ret = -EINVAL;
4780 		break;
4781 	}
4782 
4783 	if (copy_to_user(arg, p, sizeof(*p)))
4784 		ret = -EFAULT;
4785 out:
4786 	kfree(p);
4787 	return ret;
4788 }
4789 
4790 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4791 {
4792 	int ret = 0;
4793 	int i;
4794 	u64 rel_ptr;
4795 	int size;
4796 	struct btrfs_ioctl_ino_path_args *ipa = NULL;
4797 	struct inode_fs_paths *ipath = NULL;
4798 	struct btrfs_path *path;
4799 
4800 	if (!capable(CAP_DAC_READ_SEARCH))
4801 		return -EPERM;
4802 
4803 	path = btrfs_alloc_path();
4804 	if (!path) {
4805 		ret = -ENOMEM;
4806 		goto out;
4807 	}
4808 
4809 	ipa = memdup_user(arg, sizeof(*ipa));
4810 	if (IS_ERR(ipa)) {
4811 		ret = PTR_ERR(ipa);
4812 		ipa = NULL;
4813 		goto out;
4814 	}
4815 
4816 	size = min_t(u32, ipa->size, 4096);
4817 	ipath = init_ipath(size, root, path);
4818 	if (IS_ERR(ipath)) {
4819 		ret = PTR_ERR(ipath);
4820 		ipath = NULL;
4821 		goto out;
4822 	}
4823 
4824 	ret = paths_from_inode(ipa->inum, ipath);
4825 	if (ret < 0)
4826 		goto out;
4827 
4828 	for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4829 		rel_ptr = ipath->fspath->val[i] -
4830 			  (u64)(unsigned long)ipath->fspath->val;
4831 		ipath->fspath->val[i] = rel_ptr;
4832 	}
4833 
4834 	ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
4835 			   ipath->fspath, size);
4836 	if (ret) {
4837 		ret = -EFAULT;
4838 		goto out;
4839 	}
4840 
4841 out:
4842 	btrfs_free_path(path);
4843 	free_ipath(ipath);
4844 	kfree(ipa);
4845 
4846 	return ret;
4847 }
4848 
4849 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4850 {
4851 	struct btrfs_data_container *inodes = ctx;
4852 	const size_t c = 3 * sizeof(u64);
4853 
4854 	if (inodes->bytes_left >= c) {
4855 		inodes->bytes_left -= c;
4856 		inodes->val[inodes->elem_cnt] = inum;
4857 		inodes->val[inodes->elem_cnt + 1] = offset;
4858 		inodes->val[inodes->elem_cnt + 2] = root;
4859 		inodes->elem_cnt += 3;
4860 	} else {
4861 		inodes->bytes_missing += c - inodes->bytes_left;
4862 		inodes->bytes_left = 0;
4863 		inodes->elem_missed += 3;
4864 	}
4865 
4866 	return 0;
4867 }
4868 
4869 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
4870 					void __user *arg, int version)
4871 {
4872 	int ret = 0;
4873 	int size;
4874 	struct btrfs_ioctl_logical_ino_args *loi;
4875 	struct btrfs_data_container *inodes = NULL;
4876 	struct btrfs_path *path = NULL;
4877 	bool ignore_offset;
4878 
4879 	if (!capable(CAP_SYS_ADMIN))
4880 		return -EPERM;
4881 
4882 	loi = memdup_user(arg, sizeof(*loi));
4883 	if (IS_ERR(loi))
4884 		return PTR_ERR(loi);
4885 
4886 	if (version == 1) {
4887 		ignore_offset = false;
4888 		size = min_t(u32, loi->size, SZ_64K);
4889 	} else {
4890 		/* All reserved bits must be 0 for now */
4891 		if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
4892 			ret = -EINVAL;
4893 			goto out_loi;
4894 		}
4895 		/* Only accept flags we have defined so far */
4896 		if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
4897 			ret = -EINVAL;
4898 			goto out_loi;
4899 		}
4900 		ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
4901 		size = min_t(u32, loi->size, SZ_16M);
4902 	}
4903 
4904 	path = btrfs_alloc_path();
4905 	if (!path) {
4906 		ret = -ENOMEM;
4907 		goto out;
4908 	}
4909 
4910 	inodes = init_data_container(size);
4911 	if (IS_ERR(inodes)) {
4912 		ret = PTR_ERR(inodes);
4913 		inodes = NULL;
4914 		goto out;
4915 	}
4916 
4917 	ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
4918 					  build_ino_list, inodes, ignore_offset);
4919 	if (ret == -EINVAL)
4920 		ret = -ENOENT;
4921 	if (ret < 0)
4922 		goto out;
4923 
4924 	ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
4925 			   size);
4926 	if (ret)
4927 		ret = -EFAULT;
4928 
4929 out:
4930 	btrfs_free_path(path);
4931 	kvfree(inodes);
4932 out_loi:
4933 	kfree(loi);
4934 
4935 	return ret;
4936 }
4937 
4938 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
4939 			       struct btrfs_ioctl_balance_args *bargs)
4940 {
4941 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4942 
4943 	bargs->flags = bctl->flags;
4944 
4945 	if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
4946 		bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4947 	if (atomic_read(&fs_info->balance_pause_req))
4948 		bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4949 	if (atomic_read(&fs_info->balance_cancel_req))
4950 		bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4951 
4952 	memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4953 	memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4954 	memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4955 
4956 	spin_lock(&fs_info->balance_lock);
4957 	memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4958 	spin_unlock(&fs_info->balance_lock);
4959 }
4960 
4961 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4962 {
4963 	struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4964 	struct btrfs_fs_info *fs_info = root->fs_info;
4965 	struct btrfs_ioctl_balance_args *bargs;
4966 	struct btrfs_balance_control *bctl;
4967 	bool need_unlock; /* for mut. excl. ops lock */
4968 	int ret;
4969 
4970 	if (!capable(CAP_SYS_ADMIN))
4971 		return -EPERM;
4972 
4973 	ret = mnt_want_write_file(file);
4974 	if (ret)
4975 		return ret;
4976 
4977 again:
4978 	if (!test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4979 		mutex_lock(&fs_info->balance_mutex);
4980 		need_unlock = true;
4981 		goto locked;
4982 	}
4983 
4984 	/*
4985 	 * mut. excl. ops lock is locked.  Three possibilities:
4986 	 *   (1) some other op is running
4987 	 *   (2) balance is running
4988 	 *   (3) balance is paused -- special case (think resume)
4989 	 */
4990 	mutex_lock(&fs_info->balance_mutex);
4991 	if (fs_info->balance_ctl) {
4992 		/* this is either (2) or (3) */
4993 		if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4994 			mutex_unlock(&fs_info->balance_mutex);
4995 			/*
4996 			 * Lock released to allow other waiters to continue,
4997 			 * we'll reexamine the status again.
4998 			 */
4999 			mutex_lock(&fs_info->balance_mutex);
5000 
5001 			if (fs_info->balance_ctl &&
5002 			    !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
5003 				/* this is (3) */
5004 				need_unlock = false;
5005 				goto locked;
5006 			}
5007 
5008 			mutex_unlock(&fs_info->balance_mutex);
5009 			goto again;
5010 		} else {
5011 			/* this is (2) */
5012 			mutex_unlock(&fs_info->balance_mutex);
5013 			ret = -EINPROGRESS;
5014 			goto out;
5015 		}
5016 	} else {
5017 		/* this is (1) */
5018 		mutex_unlock(&fs_info->balance_mutex);
5019 		ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
5020 		goto out;
5021 	}
5022 
5023 locked:
5024 	BUG_ON(!test_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
5025 
5026 	if (arg) {
5027 		bargs = memdup_user(arg, sizeof(*bargs));
5028 		if (IS_ERR(bargs)) {
5029 			ret = PTR_ERR(bargs);
5030 			goto out_unlock;
5031 		}
5032 
5033 		if (bargs->flags & BTRFS_BALANCE_RESUME) {
5034 			if (!fs_info->balance_ctl) {
5035 				ret = -ENOTCONN;
5036 				goto out_bargs;
5037 			}
5038 
5039 			bctl = fs_info->balance_ctl;
5040 			spin_lock(&fs_info->balance_lock);
5041 			bctl->flags |= BTRFS_BALANCE_RESUME;
5042 			spin_unlock(&fs_info->balance_lock);
5043 
5044 			goto do_balance;
5045 		}
5046 	} else {
5047 		bargs = NULL;
5048 	}
5049 
5050 	if (fs_info->balance_ctl) {
5051 		ret = -EINPROGRESS;
5052 		goto out_bargs;
5053 	}
5054 
5055 	bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
5056 	if (!bctl) {
5057 		ret = -ENOMEM;
5058 		goto out_bargs;
5059 	}
5060 
5061 	if (arg) {
5062 		memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
5063 		memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
5064 		memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
5065 
5066 		bctl->flags = bargs->flags;
5067 	} else {
5068 		/* balance everything - no filters */
5069 		bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
5070 	}
5071 
5072 	if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
5073 		ret = -EINVAL;
5074 		goto out_bctl;
5075 	}
5076 
5077 do_balance:
5078 	/*
5079 	 * Ownership of bctl and filesystem flag BTRFS_FS_EXCL_OP goes to
5080 	 * btrfs_balance.  bctl is freed in reset_balance_state, or, if
5081 	 * restriper was paused all the way until unmount, in free_fs_info.
5082 	 * The flag should be cleared after reset_balance_state.
5083 	 */
5084 	need_unlock = false;
5085 
5086 	ret = btrfs_balance(fs_info, bctl, bargs);
5087 	bctl = NULL;
5088 
5089 	if (arg) {
5090 		if (copy_to_user(arg, bargs, sizeof(*bargs)))
5091 			ret = -EFAULT;
5092 	}
5093 
5094 out_bctl:
5095 	kfree(bctl);
5096 out_bargs:
5097 	kfree(bargs);
5098 out_unlock:
5099 	mutex_unlock(&fs_info->balance_mutex);
5100 	if (need_unlock)
5101 		clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
5102 out:
5103 	mnt_drop_write_file(file);
5104 	return ret;
5105 }
5106 
5107 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
5108 {
5109 	if (!capable(CAP_SYS_ADMIN))
5110 		return -EPERM;
5111 
5112 	switch (cmd) {
5113 	case BTRFS_BALANCE_CTL_PAUSE:
5114 		return btrfs_pause_balance(fs_info);
5115 	case BTRFS_BALANCE_CTL_CANCEL:
5116 		return btrfs_cancel_balance(fs_info);
5117 	}
5118 
5119 	return -EINVAL;
5120 }
5121 
5122 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
5123 					 void __user *arg)
5124 {
5125 	struct btrfs_ioctl_balance_args *bargs;
5126 	int ret = 0;
5127 
5128 	if (!capable(CAP_SYS_ADMIN))
5129 		return -EPERM;
5130 
5131 	mutex_lock(&fs_info->balance_mutex);
5132 	if (!fs_info->balance_ctl) {
5133 		ret = -ENOTCONN;
5134 		goto out;
5135 	}
5136 
5137 	bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
5138 	if (!bargs) {
5139 		ret = -ENOMEM;
5140 		goto out;
5141 	}
5142 
5143 	btrfs_update_ioctl_balance_args(fs_info, bargs);
5144 
5145 	if (copy_to_user(arg, bargs, sizeof(*bargs)))
5146 		ret = -EFAULT;
5147 
5148 	kfree(bargs);
5149 out:
5150 	mutex_unlock(&fs_info->balance_mutex);
5151 	return ret;
5152 }
5153 
5154 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
5155 {
5156 	struct inode *inode = file_inode(file);
5157 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5158 	struct btrfs_ioctl_quota_ctl_args *sa;
5159 	int ret;
5160 
5161 	if (!capable(CAP_SYS_ADMIN))
5162 		return -EPERM;
5163 
5164 	ret = mnt_want_write_file(file);
5165 	if (ret)
5166 		return ret;
5167 
5168 	sa = memdup_user(arg, sizeof(*sa));
5169 	if (IS_ERR(sa)) {
5170 		ret = PTR_ERR(sa);
5171 		goto drop_write;
5172 	}
5173 
5174 	down_write(&fs_info->subvol_sem);
5175 
5176 	switch (sa->cmd) {
5177 	case BTRFS_QUOTA_CTL_ENABLE:
5178 		ret = btrfs_quota_enable(fs_info);
5179 		break;
5180 	case BTRFS_QUOTA_CTL_DISABLE:
5181 		ret = btrfs_quota_disable(fs_info);
5182 		break;
5183 	default:
5184 		ret = -EINVAL;
5185 		break;
5186 	}
5187 
5188 	kfree(sa);
5189 	up_write(&fs_info->subvol_sem);
5190 drop_write:
5191 	mnt_drop_write_file(file);
5192 	return ret;
5193 }
5194 
5195 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
5196 {
5197 	struct inode *inode = file_inode(file);
5198 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5199 	struct btrfs_root *root = BTRFS_I(inode)->root;
5200 	struct btrfs_ioctl_qgroup_assign_args *sa;
5201 	struct btrfs_trans_handle *trans;
5202 	int ret;
5203 	int err;
5204 
5205 	if (!capable(CAP_SYS_ADMIN))
5206 		return -EPERM;
5207 
5208 	ret = mnt_want_write_file(file);
5209 	if (ret)
5210 		return ret;
5211 
5212 	sa = memdup_user(arg, sizeof(*sa));
5213 	if (IS_ERR(sa)) {
5214 		ret = PTR_ERR(sa);
5215 		goto drop_write;
5216 	}
5217 
5218 	trans = btrfs_join_transaction(root);
5219 	if (IS_ERR(trans)) {
5220 		ret = PTR_ERR(trans);
5221 		goto out;
5222 	}
5223 
5224 	if (sa->assign) {
5225 		ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
5226 	} else {
5227 		ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
5228 	}
5229 
5230 	/* update qgroup status and info */
5231 	err = btrfs_run_qgroups(trans);
5232 	if (err < 0)
5233 		btrfs_handle_fs_error(fs_info, err,
5234 				      "failed to update qgroup status and info");
5235 	err = btrfs_end_transaction(trans);
5236 	if (err && !ret)
5237 		ret = err;
5238 
5239 out:
5240 	kfree(sa);
5241 drop_write:
5242 	mnt_drop_write_file(file);
5243 	return ret;
5244 }
5245 
5246 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
5247 {
5248 	struct inode *inode = file_inode(file);
5249 	struct btrfs_root *root = BTRFS_I(inode)->root;
5250 	struct btrfs_ioctl_qgroup_create_args *sa;
5251 	struct btrfs_trans_handle *trans;
5252 	int ret;
5253 	int err;
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 	if (!sa->qgroupid) {
5269 		ret = -EINVAL;
5270 		goto out;
5271 	}
5272 
5273 	trans = btrfs_join_transaction(root);
5274 	if (IS_ERR(trans)) {
5275 		ret = PTR_ERR(trans);
5276 		goto out;
5277 	}
5278 
5279 	if (sa->create) {
5280 		ret = btrfs_create_qgroup(trans, sa->qgroupid);
5281 	} else {
5282 		ret = btrfs_remove_qgroup(trans, sa->qgroupid);
5283 	}
5284 
5285 	err = btrfs_end_transaction(trans);
5286 	if (err && !ret)
5287 		ret = err;
5288 
5289 out:
5290 	kfree(sa);
5291 drop_write:
5292 	mnt_drop_write_file(file);
5293 	return ret;
5294 }
5295 
5296 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
5297 {
5298 	struct inode *inode = file_inode(file);
5299 	struct btrfs_root *root = BTRFS_I(inode)->root;
5300 	struct btrfs_ioctl_qgroup_limit_args *sa;
5301 	struct btrfs_trans_handle *trans;
5302 	int ret;
5303 	int err;
5304 	u64 qgroupid;
5305 
5306 	if (!capable(CAP_SYS_ADMIN))
5307 		return -EPERM;
5308 
5309 	ret = mnt_want_write_file(file);
5310 	if (ret)
5311 		return ret;
5312 
5313 	sa = memdup_user(arg, sizeof(*sa));
5314 	if (IS_ERR(sa)) {
5315 		ret = PTR_ERR(sa);
5316 		goto drop_write;
5317 	}
5318 
5319 	trans = btrfs_join_transaction(root);
5320 	if (IS_ERR(trans)) {
5321 		ret = PTR_ERR(trans);
5322 		goto out;
5323 	}
5324 
5325 	qgroupid = sa->qgroupid;
5326 	if (!qgroupid) {
5327 		/* take the current subvol as qgroup */
5328 		qgroupid = root->root_key.objectid;
5329 	}
5330 
5331 	ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
5332 
5333 	err = btrfs_end_transaction(trans);
5334 	if (err && !ret)
5335 		ret = err;
5336 
5337 out:
5338 	kfree(sa);
5339 drop_write:
5340 	mnt_drop_write_file(file);
5341 	return ret;
5342 }
5343 
5344 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
5345 {
5346 	struct inode *inode = file_inode(file);
5347 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5348 	struct btrfs_ioctl_quota_rescan_args *qsa;
5349 	int ret;
5350 
5351 	if (!capable(CAP_SYS_ADMIN))
5352 		return -EPERM;
5353 
5354 	ret = mnt_want_write_file(file);
5355 	if (ret)
5356 		return ret;
5357 
5358 	qsa = memdup_user(arg, sizeof(*qsa));
5359 	if (IS_ERR(qsa)) {
5360 		ret = PTR_ERR(qsa);
5361 		goto drop_write;
5362 	}
5363 
5364 	if (qsa->flags) {
5365 		ret = -EINVAL;
5366 		goto out;
5367 	}
5368 
5369 	ret = btrfs_qgroup_rescan(fs_info);
5370 
5371 out:
5372 	kfree(qsa);
5373 drop_write:
5374 	mnt_drop_write_file(file);
5375 	return ret;
5376 }
5377 
5378 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
5379 {
5380 	struct inode *inode = file_inode(file);
5381 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5382 	struct btrfs_ioctl_quota_rescan_args *qsa;
5383 	int ret = 0;
5384 
5385 	if (!capable(CAP_SYS_ADMIN))
5386 		return -EPERM;
5387 
5388 	qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
5389 	if (!qsa)
5390 		return -ENOMEM;
5391 
5392 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
5393 		qsa->flags = 1;
5394 		qsa->progress = fs_info->qgroup_rescan_progress.objectid;
5395 	}
5396 
5397 	if (copy_to_user(arg, qsa, sizeof(*qsa)))
5398 		ret = -EFAULT;
5399 
5400 	kfree(qsa);
5401 	return ret;
5402 }
5403 
5404 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
5405 {
5406 	struct inode *inode = file_inode(file);
5407 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5408 
5409 	if (!capable(CAP_SYS_ADMIN))
5410 		return -EPERM;
5411 
5412 	return btrfs_qgroup_wait_for_completion(fs_info, true);
5413 }
5414 
5415 static long _btrfs_ioctl_set_received_subvol(struct file *file,
5416 					    struct btrfs_ioctl_received_subvol_args *sa)
5417 {
5418 	struct inode *inode = file_inode(file);
5419 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5420 	struct btrfs_root *root = BTRFS_I(inode)->root;
5421 	struct btrfs_root_item *root_item = &root->root_item;
5422 	struct btrfs_trans_handle *trans;
5423 	struct timespec64 ct = current_time(inode);
5424 	int ret = 0;
5425 	int received_uuid_changed;
5426 
5427 	if (!inode_owner_or_capable(inode))
5428 		return -EPERM;
5429 
5430 	ret = mnt_want_write_file(file);
5431 	if (ret < 0)
5432 		return ret;
5433 
5434 	down_write(&fs_info->subvol_sem);
5435 
5436 	if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
5437 		ret = -EINVAL;
5438 		goto out;
5439 	}
5440 
5441 	if (btrfs_root_readonly(root)) {
5442 		ret = -EROFS;
5443 		goto out;
5444 	}
5445 
5446 	/*
5447 	 * 1 - root item
5448 	 * 2 - uuid items (received uuid + subvol uuid)
5449 	 */
5450 	trans = btrfs_start_transaction(root, 3);
5451 	if (IS_ERR(trans)) {
5452 		ret = PTR_ERR(trans);
5453 		trans = NULL;
5454 		goto out;
5455 	}
5456 
5457 	sa->rtransid = trans->transid;
5458 	sa->rtime.sec = ct.tv_sec;
5459 	sa->rtime.nsec = ct.tv_nsec;
5460 
5461 	received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5462 				       BTRFS_UUID_SIZE);
5463 	if (received_uuid_changed &&
5464 	    !btrfs_is_empty_uuid(root_item->received_uuid)) {
5465 		ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
5466 					  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5467 					  root->root_key.objectid);
5468 		if (ret && ret != -ENOENT) {
5469 		        btrfs_abort_transaction(trans, ret);
5470 		        btrfs_end_transaction(trans);
5471 		        goto out;
5472 		}
5473 	}
5474 	memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5475 	btrfs_set_root_stransid(root_item, sa->stransid);
5476 	btrfs_set_root_rtransid(root_item, sa->rtransid);
5477 	btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5478 	btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5479 	btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5480 	btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5481 
5482 	ret = btrfs_update_root(trans, fs_info->tree_root,
5483 				&root->root_key, &root->root_item);
5484 	if (ret < 0) {
5485 		btrfs_end_transaction(trans);
5486 		goto out;
5487 	}
5488 	if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5489 		ret = btrfs_uuid_tree_add(trans, sa->uuid,
5490 					  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5491 					  root->root_key.objectid);
5492 		if (ret < 0 && ret != -EEXIST) {
5493 			btrfs_abort_transaction(trans, ret);
5494 			btrfs_end_transaction(trans);
5495 			goto out;
5496 		}
5497 	}
5498 	ret = btrfs_commit_transaction(trans);
5499 out:
5500 	up_write(&fs_info->subvol_sem);
5501 	mnt_drop_write_file(file);
5502 	return ret;
5503 }
5504 
5505 #ifdef CONFIG_64BIT
5506 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5507 						void __user *arg)
5508 {
5509 	struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5510 	struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5511 	int ret = 0;
5512 
5513 	args32 = memdup_user(arg, sizeof(*args32));
5514 	if (IS_ERR(args32))
5515 		return PTR_ERR(args32);
5516 
5517 	args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
5518 	if (!args64) {
5519 		ret = -ENOMEM;
5520 		goto out;
5521 	}
5522 
5523 	memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5524 	args64->stransid = args32->stransid;
5525 	args64->rtransid = args32->rtransid;
5526 	args64->stime.sec = args32->stime.sec;
5527 	args64->stime.nsec = args32->stime.nsec;
5528 	args64->rtime.sec = args32->rtime.sec;
5529 	args64->rtime.nsec = args32->rtime.nsec;
5530 	args64->flags = args32->flags;
5531 
5532 	ret = _btrfs_ioctl_set_received_subvol(file, args64);
5533 	if (ret)
5534 		goto out;
5535 
5536 	memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5537 	args32->stransid = args64->stransid;
5538 	args32->rtransid = args64->rtransid;
5539 	args32->stime.sec = args64->stime.sec;
5540 	args32->stime.nsec = args64->stime.nsec;
5541 	args32->rtime.sec = args64->rtime.sec;
5542 	args32->rtime.nsec = args64->rtime.nsec;
5543 	args32->flags = args64->flags;
5544 
5545 	ret = copy_to_user(arg, args32, sizeof(*args32));
5546 	if (ret)
5547 		ret = -EFAULT;
5548 
5549 out:
5550 	kfree(args32);
5551 	kfree(args64);
5552 	return ret;
5553 }
5554 #endif
5555 
5556 static long btrfs_ioctl_set_received_subvol(struct file *file,
5557 					    void __user *arg)
5558 {
5559 	struct btrfs_ioctl_received_subvol_args *sa = NULL;
5560 	int ret = 0;
5561 
5562 	sa = memdup_user(arg, sizeof(*sa));
5563 	if (IS_ERR(sa))
5564 		return PTR_ERR(sa);
5565 
5566 	ret = _btrfs_ioctl_set_received_subvol(file, sa);
5567 
5568 	if (ret)
5569 		goto out;
5570 
5571 	ret = copy_to_user(arg, sa, sizeof(*sa));
5572 	if (ret)
5573 		ret = -EFAULT;
5574 
5575 out:
5576 	kfree(sa);
5577 	return ret;
5578 }
5579 
5580 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5581 {
5582 	struct inode *inode = file_inode(file);
5583 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5584 	size_t len;
5585 	int ret;
5586 	char label[BTRFS_LABEL_SIZE];
5587 
5588 	spin_lock(&fs_info->super_lock);
5589 	memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5590 	spin_unlock(&fs_info->super_lock);
5591 
5592 	len = strnlen(label, BTRFS_LABEL_SIZE);
5593 
5594 	if (len == BTRFS_LABEL_SIZE) {
5595 		btrfs_warn(fs_info,
5596 			   "label is too long, return the first %zu bytes",
5597 			   --len);
5598 	}
5599 
5600 	ret = copy_to_user(arg, label, len);
5601 
5602 	return ret ? -EFAULT : 0;
5603 }
5604 
5605 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5606 {
5607 	struct inode *inode = file_inode(file);
5608 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5609 	struct btrfs_root *root = BTRFS_I(inode)->root;
5610 	struct btrfs_super_block *super_block = fs_info->super_copy;
5611 	struct btrfs_trans_handle *trans;
5612 	char label[BTRFS_LABEL_SIZE];
5613 	int ret;
5614 
5615 	if (!capable(CAP_SYS_ADMIN))
5616 		return -EPERM;
5617 
5618 	if (copy_from_user(label, arg, sizeof(label)))
5619 		return -EFAULT;
5620 
5621 	if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5622 		btrfs_err(fs_info,
5623 			  "unable to set label with more than %d bytes",
5624 			  BTRFS_LABEL_SIZE - 1);
5625 		return -EINVAL;
5626 	}
5627 
5628 	ret = mnt_want_write_file(file);
5629 	if (ret)
5630 		return ret;
5631 
5632 	trans = btrfs_start_transaction(root, 0);
5633 	if (IS_ERR(trans)) {
5634 		ret = PTR_ERR(trans);
5635 		goto out_unlock;
5636 	}
5637 
5638 	spin_lock(&fs_info->super_lock);
5639 	strcpy(super_block->label, label);
5640 	spin_unlock(&fs_info->super_lock);
5641 	ret = btrfs_commit_transaction(trans);
5642 
5643 out_unlock:
5644 	mnt_drop_write_file(file);
5645 	return ret;
5646 }
5647 
5648 #define INIT_FEATURE_FLAGS(suffix) \
5649 	{ .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5650 	  .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5651 	  .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5652 
5653 int btrfs_ioctl_get_supported_features(void __user *arg)
5654 {
5655 	static const struct btrfs_ioctl_feature_flags features[3] = {
5656 		INIT_FEATURE_FLAGS(SUPP),
5657 		INIT_FEATURE_FLAGS(SAFE_SET),
5658 		INIT_FEATURE_FLAGS(SAFE_CLEAR)
5659 	};
5660 
5661 	if (copy_to_user(arg, &features, sizeof(features)))
5662 		return -EFAULT;
5663 
5664 	return 0;
5665 }
5666 
5667 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5668 {
5669 	struct inode *inode = file_inode(file);
5670 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5671 	struct btrfs_super_block *super_block = fs_info->super_copy;
5672 	struct btrfs_ioctl_feature_flags features;
5673 
5674 	features.compat_flags = btrfs_super_compat_flags(super_block);
5675 	features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5676 	features.incompat_flags = btrfs_super_incompat_flags(super_block);
5677 
5678 	if (copy_to_user(arg, &features, sizeof(features)))
5679 		return -EFAULT;
5680 
5681 	return 0;
5682 }
5683 
5684 static int check_feature_bits(struct btrfs_fs_info *fs_info,
5685 			      enum btrfs_feature_set set,
5686 			      u64 change_mask, u64 flags, u64 supported_flags,
5687 			      u64 safe_set, u64 safe_clear)
5688 {
5689 	const char *type = btrfs_feature_set_names[set];
5690 	char *names;
5691 	u64 disallowed, unsupported;
5692 	u64 set_mask = flags & change_mask;
5693 	u64 clear_mask = ~flags & change_mask;
5694 
5695 	unsupported = set_mask & ~supported_flags;
5696 	if (unsupported) {
5697 		names = btrfs_printable_features(set, unsupported);
5698 		if (names) {
5699 			btrfs_warn(fs_info,
5700 				   "this kernel does not support the %s feature bit%s",
5701 				   names, strchr(names, ',') ? "s" : "");
5702 			kfree(names);
5703 		} else
5704 			btrfs_warn(fs_info,
5705 				   "this kernel does not support %s bits 0x%llx",
5706 				   type, unsupported);
5707 		return -EOPNOTSUPP;
5708 	}
5709 
5710 	disallowed = set_mask & ~safe_set;
5711 	if (disallowed) {
5712 		names = btrfs_printable_features(set, disallowed);
5713 		if (names) {
5714 			btrfs_warn(fs_info,
5715 				   "can't set the %s feature bit%s while mounted",
5716 				   names, strchr(names, ',') ? "s" : "");
5717 			kfree(names);
5718 		} else
5719 			btrfs_warn(fs_info,
5720 				   "can't set %s bits 0x%llx while mounted",
5721 				   type, disallowed);
5722 		return -EPERM;
5723 	}
5724 
5725 	disallowed = clear_mask & ~safe_clear;
5726 	if (disallowed) {
5727 		names = btrfs_printable_features(set, disallowed);
5728 		if (names) {
5729 			btrfs_warn(fs_info,
5730 				   "can't clear the %s feature bit%s while mounted",
5731 				   names, strchr(names, ',') ? "s" : "");
5732 			kfree(names);
5733 		} else
5734 			btrfs_warn(fs_info,
5735 				   "can't clear %s bits 0x%llx while mounted",
5736 				   type, disallowed);
5737 		return -EPERM;
5738 	}
5739 
5740 	return 0;
5741 }
5742 
5743 #define check_feature(fs_info, change_mask, flags, mask_base)	\
5744 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,	\
5745 		   BTRFS_FEATURE_ ## mask_base ## _SUPP,	\
5746 		   BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,	\
5747 		   BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5748 
5749 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5750 {
5751 	struct inode *inode = file_inode(file);
5752 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5753 	struct btrfs_root *root = BTRFS_I(inode)->root;
5754 	struct btrfs_super_block *super_block = fs_info->super_copy;
5755 	struct btrfs_ioctl_feature_flags flags[2];
5756 	struct btrfs_trans_handle *trans;
5757 	u64 newflags;
5758 	int ret;
5759 
5760 	if (!capable(CAP_SYS_ADMIN))
5761 		return -EPERM;
5762 
5763 	if (copy_from_user(flags, arg, sizeof(flags)))
5764 		return -EFAULT;
5765 
5766 	/* Nothing to do */
5767 	if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5768 	    !flags[0].incompat_flags)
5769 		return 0;
5770 
5771 	ret = check_feature(fs_info, flags[0].compat_flags,
5772 			    flags[1].compat_flags, COMPAT);
5773 	if (ret)
5774 		return ret;
5775 
5776 	ret = check_feature(fs_info, flags[0].compat_ro_flags,
5777 			    flags[1].compat_ro_flags, COMPAT_RO);
5778 	if (ret)
5779 		return ret;
5780 
5781 	ret = check_feature(fs_info, flags[0].incompat_flags,
5782 			    flags[1].incompat_flags, INCOMPAT);
5783 	if (ret)
5784 		return ret;
5785 
5786 	ret = mnt_want_write_file(file);
5787 	if (ret)
5788 		return ret;
5789 
5790 	trans = btrfs_start_transaction(root, 0);
5791 	if (IS_ERR(trans)) {
5792 		ret = PTR_ERR(trans);
5793 		goto out_drop_write;
5794 	}
5795 
5796 	spin_lock(&fs_info->super_lock);
5797 	newflags = btrfs_super_compat_flags(super_block);
5798 	newflags |= flags[0].compat_flags & flags[1].compat_flags;
5799 	newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5800 	btrfs_set_super_compat_flags(super_block, newflags);
5801 
5802 	newflags = btrfs_super_compat_ro_flags(super_block);
5803 	newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5804 	newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5805 	btrfs_set_super_compat_ro_flags(super_block, newflags);
5806 
5807 	newflags = btrfs_super_incompat_flags(super_block);
5808 	newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5809 	newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5810 	btrfs_set_super_incompat_flags(super_block, newflags);
5811 	spin_unlock(&fs_info->super_lock);
5812 
5813 	ret = btrfs_commit_transaction(trans);
5814 out_drop_write:
5815 	mnt_drop_write_file(file);
5816 
5817 	return ret;
5818 }
5819 
5820 static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
5821 {
5822 	struct btrfs_ioctl_send_args *arg;
5823 	int ret;
5824 
5825 	if (compat) {
5826 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5827 		struct btrfs_ioctl_send_args_32 args32;
5828 
5829 		ret = copy_from_user(&args32, argp, sizeof(args32));
5830 		if (ret)
5831 			return -EFAULT;
5832 		arg = kzalloc(sizeof(*arg), GFP_KERNEL);
5833 		if (!arg)
5834 			return -ENOMEM;
5835 		arg->send_fd = args32.send_fd;
5836 		arg->clone_sources_count = args32.clone_sources_count;
5837 		arg->clone_sources = compat_ptr(args32.clone_sources);
5838 		arg->parent_root = args32.parent_root;
5839 		arg->flags = args32.flags;
5840 		memcpy(arg->reserved, args32.reserved,
5841 		       sizeof(args32.reserved));
5842 #else
5843 		return -ENOTTY;
5844 #endif
5845 	} else {
5846 		arg = memdup_user(argp, sizeof(*arg));
5847 		if (IS_ERR(arg))
5848 			return PTR_ERR(arg);
5849 	}
5850 	ret = btrfs_ioctl_send(file, arg);
5851 	kfree(arg);
5852 	return ret;
5853 }
5854 
5855 long btrfs_ioctl(struct file *file, unsigned int
5856 		cmd, unsigned long arg)
5857 {
5858 	struct inode *inode = file_inode(file);
5859 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5860 	struct btrfs_root *root = BTRFS_I(inode)->root;
5861 	void __user *argp = (void __user *)arg;
5862 
5863 	switch (cmd) {
5864 	case FS_IOC_GETFLAGS:
5865 		return btrfs_ioctl_getflags(file, argp);
5866 	case FS_IOC_SETFLAGS:
5867 		return btrfs_ioctl_setflags(file, argp);
5868 	case FS_IOC_GETVERSION:
5869 		return btrfs_ioctl_getversion(file, argp);
5870 	case FITRIM:
5871 		return btrfs_ioctl_fitrim(file, argp);
5872 	case BTRFS_IOC_SNAP_CREATE:
5873 		return btrfs_ioctl_snap_create(file, argp, 0);
5874 	case BTRFS_IOC_SNAP_CREATE_V2:
5875 		return btrfs_ioctl_snap_create_v2(file, argp, 0);
5876 	case BTRFS_IOC_SUBVOL_CREATE:
5877 		return btrfs_ioctl_snap_create(file, argp, 1);
5878 	case BTRFS_IOC_SUBVOL_CREATE_V2:
5879 		return btrfs_ioctl_snap_create_v2(file, argp, 1);
5880 	case BTRFS_IOC_SNAP_DESTROY:
5881 		return btrfs_ioctl_snap_destroy(file, argp);
5882 	case BTRFS_IOC_SUBVOL_GETFLAGS:
5883 		return btrfs_ioctl_subvol_getflags(file, argp);
5884 	case BTRFS_IOC_SUBVOL_SETFLAGS:
5885 		return btrfs_ioctl_subvol_setflags(file, argp);
5886 	case BTRFS_IOC_DEFAULT_SUBVOL:
5887 		return btrfs_ioctl_default_subvol(file, argp);
5888 	case BTRFS_IOC_DEFRAG:
5889 		return btrfs_ioctl_defrag(file, NULL);
5890 	case BTRFS_IOC_DEFRAG_RANGE:
5891 		return btrfs_ioctl_defrag(file, argp);
5892 	case BTRFS_IOC_RESIZE:
5893 		return btrfs_ioctl_resize(file, argp);
5894 	case BTRFS_IOC_ADD_DEV:
5895 		return btrfs_ioctl_add_dev(fs_info, argp);
5896 	case BTRFS_IOC_RM_DEV:
5897 		return btrfs_ioctl_rm_dev(file, argp);
5898 	case BTRFS_IOC_RM_DEV_V2:
5899 		return btrfs_ioctl_rm_dev_v2(file, argp);
5900 	case BTRFS_IOC_FS_INFO:
5901 		return btrfs_ioctl_fs_info(fs_info, argp);
5902 	case BTRFS_IOC_DEV_INFO:
5903 		return btrfs_ioctl_dev_info(fs_info, argp);
5904 	case BTRFS_IOC_BALANCE:
5905 		return btrfs_ioctl_balance(file, NULL);
5906 	case BTRFS_IOC_TREE_SEARCH:
5907 		return btrfs_ioctl_tree_search(file, argp);
5908 	case BTRFS_IOC_TREE_SEARCH_V2:
5909 		return btrfs_ioctl_tree_search_v2(file, argp);
5910 	case BTRFS_IOC_INO_LOOKUP:
5911 		return btrfs_ioctl_ino_lookup(file, argp);
5912 	case BTRFS_IOC_INO_PATHS:
5913 		return btrfs_ioctl_ino_to_path(root, argp);
5914 	case BTRFS_IOC_LOGICAL_INO:
5915 		return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
5916 	case BTRFS_IOC_LOGICAL_INO_V2:
5917 		return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
5918 	case BTRFS_IOC_SPACE_INFO:
5919 		return btrfs_ioctl_space_info(fs_info, argp);
5920 	case BTRFS_IOC_SYNC: {
5921 		int ret;
5922 
5923 		ret = btrfs_start_delalloc_roots(fs_info, -1);
5924 		if (ret)
5925 			return ret;
5926 		ret = btrfs_sync_fs(inode->i_sb, 1);
5927 		/*
5928 		 * The transaction thread may want to do more work,
5929 		 * namely it pokes the cleaner kthread that will start
5930 		 * processing uncleaned subvols.
5931 		 */
5932 		wake_up_process(fs_info->transaction_kthread);
5933 		return ret;
5934 	}
5935 	case BTRFS_IOC_START_SYNC:
5936 		return btrfs_ioctl_start_sync(root, argp);
5937 	case BTRFS_IOC_WAIT_SYNC:
5938 		return btrfs_ioctl_wait_sync(fs_info, argp);
5939 	case BTRFS_IOC_SCRUB:
5940 		return btrfs_ioctl_scrub(file, argp);
5941 	case BTRFS_IOC_SCRUB_CANCEL:
5942 		return btrfs_ioctl_scrub_cancel(fs_info);
5943 	case BTRFS_IOC_SCRUB_PROGRESS:
5944 		return btrfs_ioctl_scrub_progress(fs_info, argp);
5945 	case BTRFS_IOC_BALANCE_V2:
5946 		return btrfs_ioctl_balance(file, argp);
5947 	case BTRFS_IOC_BALANCE_CTL:
5948 		return btrfs_ioctl_balance_ctl(fs_info, arg);
5949 	case BTRFS_IOC_BALANCE_PROGRESS:
5950 		return btrfs_ioctl_balance_progress(fs_info, argp);
5951 	case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5952 		return btrfs_ioctl_set_received_subvol(file, argp);
5953 #ifdef CONFIG_64BIT
5954 	case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5955 		return btrfs_ioctl_set_received_subvol_32(file, argp);
5956 #endif
5957 	case BTRFS_IOC_SEND:
5958 		return _btrfs_ioctl_send(file, argp, false);
5959 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5960 	case BTRFS_IOC_SEND_32:
5961 		return _btrfs_ioctl_send(file, argp, true);
5962 #endif
5963 	case BTRFS_IOC_GET_DEV_STATS:
5964 		return btrfs_ioctl_get_dev_stats(fs_info, argp);
5965 	case BTRFS_IOC_QUOTA_CTL:
5966 		return btrfs_ioctl_quota_ctl(file, argp);
5967 	case BTRFS_IOC_QGROUP_ASSIGN:
5968 		return btrfs_ioctl_qgroup_assign(file, argp);
5969 	case BTRFS_IOC_QGROUP_CREATE:
5970 		return btrfs_ioctl_qgroup_create(file, argp);
5971 	case BTRFS_IOC_QGROUP_LIMIT:
5972 		return btrfs_ioctl_qgroup_limit(file, argp);
5973 	case BTRFS_IOC_QUOTA_RESCAN:
5974 		return btrfs_ioctl_quota_rescan(file, argp);
5975 	case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5976 		return btrfs_ioctl_quota_rescan_status(file, argp);
5977 	case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5978 		return btrfs_ioctl_quota_rescan_wait(file, argp);
5979 	case BTRFS_IOC_DEV_REPLACE:
5980 		return btrfs_ioctl_dev_replace(fs_info, argp);
5981 	case BTRFS_IOC_GET_FSLABEL:
5982 		return btrfs_ioctl_get_fslabel(file, argp);
5983 	case BTRFS_IOC_SET_FSLABEL:
5984 		return btrfs_ioctl_set_fslabel(file, argp);
5985 	case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5986 		return btrfs_ioctl_get_supported_features(argp);
5987 	case BTRFS_IOC_GET_FEATURES:
5988 		return btrfs_ioctl_get_features(file, argp);
5989 	case BTRFS_IOC_SET_FEATURES:
5990 		return btrfs_ioctl_set_features(file, argp);
5991 	case FS_IOC_FSGETXATTR:
5992 		return btrfs_ioctl_fsgetxattr(file, argp);
5993 	case FS_IOC_FSSETXATTR:
5994 		return btrfs_ioctl_fssetxattr(file, argp);
5995 	case BTRFS_IOC_GET_SUBVOL_INFO:
5996 		return btrfs_ioctl_get_subvol_info(file, argp);
5997 	case BTRFS_IOC_GET_SUBVOL_ROOTREF:
5998 		return btrfs_ioctl_get_subvol_rootref(file, argp);
5999 	case BTRFS_IOC_INO_LOOKUP_USER:
6000 		return btrfs_ioctl_ino_lookup_user(file, argp);
6001 	}
6002 
6003 	return -ENOTTY;
6004 }
6005 
6006 #ifdef CONFIG_COMPAT
6007 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
6008 {
6009 	/*
6010 	 * These all access 32-bit values anyway so no further
6011 	 * handling is necessary.
6012 	 */
6013 	switch (cmd) {
6014 	case FS_IOC32_GETFLAGS:
6015 		cmd = FS_IOC_GETFLAGS;
6016 		break;
6017 	case FS_IOC32_SETFLAGS:
6018 		cmd = FS_IOC_SETFLAGS;
6019 		break;
6020 	case FS_IOC32_GETVERSION:
6021 		cmd = FS_IOC_GETVERSION;
6022 		break;
6023 	}
6024 
6025 	return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
6026 }
6027 #endif
6028