xref: /openbmc/linux/fs/btrfs/ioctl.c (revision ae3473231e77a3f1909d48cd144cebe5e1d049b3)
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/fsnotify.h>
25 #include <linux/pagemap.h>
26 #include <linux/highmem.h>
27 #include <linux/time.h>
28 #include <linux/init.h>
29 #include <linux/string.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mount.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/swap.h>
35 #include <linux/writeback.h>
36 #include <linux/compat.h>
37 #include <linux/bit_spinlock.h>
38 #include <linux/security.h>
39 #include <linux/xattr.h>
40 #include <linux/vmalloc.h>
41 #include <linux/slab.h>
42 #include <linux/blkdev.h>
43 #include <linux/uuid.h>
44 #include <linux/btrfs.h>
45 #include <linux/uaccess.h>
46 #include "ctree.h"
47 #include "disk-io.h"
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "volumes.h"
52 #include "locking.h"
53 #include "inode-map.h"
54 #include "backref.h"
55 #include "rcu-string.h"
56 #include "send.h"
57 #include "dev-replace.h"
58 #include "props.h"
59 #include "sysfs.h"
60 #include "qgroup.h"
61 #include "tree-log.h"
62 #include "compression.h"
63 
64 #ifdef CONFIG_64BIT
65 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
66  * structures are incorrect, as the timespec structure from userspace
67  * is 4 bytes too small. We define these alternatives here to teach
68  * the kernel about the 32-bit struct packing.
69  */
70 struct btrfs_ioctl_timespec_32 {
71 	__u64 sec;
72 	__u32 nsec;
73 } __attribute__ ((__packed__));
74 
75 struct btrfs_ioctl_received_subvol_args_32 {
76 	char	uuid[BTRFS_UUID_SIZE];	/* in */
77 	__u64	stransid;		/* in */
78 	__u64	rtransid;		/* out */
79 	struct btrfs_ioctl_timespec_32 stime; /* in */
80 	struct btrfs_ioctl_timespec_32 rtime; /* out */
81 	__u64	flags;			/* in */
82 	__u64	reserved[16];		/* in */
83 } __attribute__ ((__packed__));
84 
85 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
86 				struct btrfs_ioctl_received_subvol_args_32)
87 #endif
88 
89 
90 static int btrfs_clone(struct inode *src, struct inode *inode,
91 		       u64 off, u64 olen, u64 olen_aligned, u64 destoff,
92 		       int no_time_update);
93 
94 /* Mask out flags that are inappropriate for the given type of inode. */
95 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
96 {
97 	if (S_ISDIR(mode))
98 		return flags;
99 	else if (S_ISREG(mode))
100 		return flags & ~FS_DIRSYNC_FL;
101 	else
102 		return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
103 }
104 
105 /*
106  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
107  */
108 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
109 {
110 	unsigned int iflags = 0;
111 
112 	if (flags & BTRFS_INODE_SYNC)
113 		iflags |= FS_SYNC_FL;
114 	if (flags & BTRFS_INODE_IMMUTABLE)
115 		iflags |= FS_IMMUTABLE_FL;
116 	if (flags & BTRFS_INODE_APPEND)
117 		iflags |= FS_APPEND_FL;
118 	if (flags & BTRFS_INODE_NODUMP)
119 		iflags |= FS_NODUMP_FL;
120 	if (flags & BTRFS_INODE_NOATIME)
121 		iflags |= FS_NOATIME_FL;
122 	if (flags & BTRFS_INODE_DIRSYNC)
123 		iflags |= FS_DIRSYNC_FL;
124 	if (flags & BTRFS_INODE_NODATACOW)
125 		iflags |= FS_NOCOW_FL;
126 
127 	if (flags & BTRFS_INODE_NOCOMPRESS)
128 		iflags |= FS_NOCOMP_FL;
129 	else if (flags & BTRFS_INODE_COMPRESS)
130 		iflags |= FS_COMPR_FL;
131 
132 	return iflags;
133 }
134 
135 /*
136  * Update inode->i_flags based on the btrfs internal flags.
137  */
138 void btrfs_update_iflags(struct inode *inode)
139 {
140 	struct btrfs_inode *ip = BTRFS_I(inode);
141 	unsigned int new_fl = 0;
142 
143 	if (ip->flags & BTRFS_INODE_SYNC)
144 		new_fl |= S_SYNC;
145 	if (ip->flags & BTRFS_INODE_IMMUTABLE)
146 		new_fl |= S_IMMUTABLE;
147 	if (ip->flags & BTRFS_INODE_APPEND)
148 		new_fl |= S_APPEND;
149 	if (ip->flags & BTRFS_INODE_NOATIME)
150 		new_fl |= S_NOATIME;
151 	if (ip->flags & BTRFS_INODE_DIRSYNC)
152 		new_fl |= S_DIRSYNC;
153 
154 	set_mask_bits(&inode->i_flags,
155 		      S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
156 		      new_fl);
157 }
158 
159 /*
160  * Inherit flags from the parent inode.
161  *
162  * Currently only the compression flags and the cow flags are inherited.
163  */
164 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
165 {
166 	unsigned int flags;
167 
168 	if (!dir)
169 		return;
170 
171 	flags = BTRFS_I(dir)->flags;
172 
173 	if (flags & BTRFS_INODE_NOCOMPRESS) {
174 		BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
175 		BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
176 	} else if (flags & BTRFS_INODE_COMPRESS) {
177 		BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
178 		BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
179 	}
180 
181 	if (flags & BTRFS_INODE_NODATACOW) {
182 		BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
183 		if (S_ISREG(inode->i_mode))
184 			BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
185 	}
186 
187 	btrfs_update_iflags(inode);
188 }
189 
190 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
191 {
192 	struct btrfs_inode *ip = BTRFS_I(file_inode(file));
193 	unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
194 
195 	if (copy_to_user(arg, &flags, sizeof(flags)))
196 		return -EFAULT;
197 	return 0;
198 }
199 
200 static int check_flags(unsigned int flags)
201 {
202 	if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
203 		      FS_NOATIME_FL | FS_NODUMP_FL | \
204 		      FS_SYNC_FL | FS_DIRSYNC_FL | \
205 		      FS_NOCOMP_FL | FS_COMPR_FL |
206 		      FS_NOCOW_FL))
207 		return -EOPNOTSUPP;
208 
209 	if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
210 		return -EINVAL;
211 
212 	return 0;
213 }
214 
215 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
216 {
217 	struct inode *inode = file_inode(file);
218 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
219 	struct btrfs_inode *ip = BTRFS_I(inode);
220 	struct btrfs_root *root = ip->root;
221 	struct btrfs_trans_handle *trans;
222 	unsigned int flags, oldflags;
223 	int ret;
224 	u64 ip_oldflags;
225 	unsigned int i_oldflags;
226 	umode_t mode;
227 
228 	if (!inode_owner_or_capable(inode))
229 		return -EPERM;
230 
231 	if (btrfs_root_readonly(root))
232 		return -EROFS;
233 
234 	if (copy_from_user(&flags, arg, sizeof(flags)))
235 		return -EFAULT;
236 
237 	ret = check_flags(flags);
238 	if (ret)
239 		return ret;
240 
241 	ret = mnt_want_write_file(file);
242 	if (ret)
243 		return ret;
244 
245 	inode_lock(inode);
246 
247 	ip_oldflags = ip->flags;
248 	i_oldflags = inode->i_flags;
249 	mode = inode->i_mode;
250 
251 	flags = btrfs_mask_flags(inode->i_mode, flags);
252 	oldflags = btrfs_flags_to_ioctl(ip->flags);
253 	if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
254 		if (!capable(CAP_LINUX_IMMUTABLE)) {
255 			ret = -EPERM;
256 			goto out_unlock;
257 		}
258 	}
259 
260 	if (flags & FS_SYNC_FL)
261 		ip->flags |= BTRFS_INODE_SYNC;
262 	else
263 		ip->flags &= ~BTRFS_INODE_SYNC;
264 	if (flags & FS_IMMUTABLE_FL)
265 		ip->flags |= BTRFS_INODE_IMMUTABLE;
266 	else
267 		ip->flags &= ~BTRFS_INODE_IMMUTABLE;
268 	if (flags & FS_APPEND_FL)
269 		ip->flags |= BTRFS_INODE_APPEND;
270 	else
271 		ip->flags &= ~BTRFS_INODE_APPEND;
272 	if (flags & FS_NODUMP_FL)
273 		ip->flags |= BTRFS_INODE_NODUMP;
274 	else
275 		ip->flags &= ~BTRFS_INODE_NODUMP;
276 	if (flags & FS_NOATIME_FL)
277 		ip->flags |= BTRFS_INODE_NOATIME;
278 	else
279 		ip->flags &= ~BTRFS_INODE_NOATIME;
280 	if (flags & FS_DIRSYNC_FL)
281 		ip->flags |= BTRFS_INODE_DIRSYNC;
282 	else
283 		ip->flags &= ~BTRFS_INODE_DIRSYNC;
284 	if (flags & FS_NOCOW_FL) {
285 		if (S_ISREG(mode)) {
286 			/*
287 			 * It's safe to turn csums off here, no extents exist.
288 			 * Otherwise we want the flag to reflect the real COW
289 			 * status of the file and will not set it.
290 			 */
291 			if (inode->i_size == 0)
292 				ip->flags |= BTRFS_INODE_NODATACOW
293 					   | BTRFS_INODE_NODATASUM;
294 		} else {
295 			ip->flags |= BTRFS_INODE_NODATACOW;
296 		}
297 	} else {
298 		/*
299 		 * Revert back under same assumptions as above
300 		 */
301 		if (S_ISREG(mode)) {
302 			if (inode->i_size == 0)
303 				ip->flags &= ~(BTRFS_INODE_NODATACOW
304 				             | BTRFS_INODE_NODATASUM);
305 		} else {
306 			ip->flags &= ~BTRFS_INODE_NODATACOW;
307 		}
308 	}
309 
310 	/*
311 	 * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
312 	 * flag may be changed automatically if compression code won't make
313 	 * things smaller.
314 	 */
315 	if (flags & FS_NOCOMP_FL) {
316 		ip->flags &= ~BTRFS_INODE_COMPRESS;
317 		ip->flags |= BTRFS_INODE_NOCOMPRESS;
318 
319 		ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
320 		if (ret && ret != -ENODATA)
321 			goto out_drop;
322 	} else if (flags & FS_COMPR_FL) {
323 		const char *comp;
324 
325 		ip->flags |= BTRFS_INODE_COMPRESS;
326 		ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
327 
328 		if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
329 			comp = "lzo";
330 		else
331 			comp = "zlib";
332 		ret = btrfs_set_prop(inode, "btrfs.compression",
333 				     comp, strlen(comp), 0);
334 		if (ret)
335 			goto out_drop;
336 
337 	} else {
338 		ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
339 		if (ret && ret != -ENODATA)
340 			goto out_drop;
341 		ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
342 	}
343 
344 	trans = btrfs_start_transaction(root, 1);
345 	if (IS_ERR(trans)) {
346 		ret = PTR_ERR(trans);
347 		goto out_drop;
348 	}
349 
350 	btrfs_update_iflags(inode);
351 	inode_inc_iversion(inode);
352 	inode->i_ctime = current_time(inode);
353 	ret = btrfs_update_inode(trans, root, inode);
354 
355 	btrfs_end_transaction(trans);
356  out_drop:
357 	if (ret) {
358 		ip->flags = ip_oldflags;
359 		inode->i_flags = i_oldflags;
360 	}
361 
362  out_unlock:
363 	inode_unlock(inode);
364 	mnt_drop_write_file(file);
365 	return ret;
366 }
367 
368 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
369 {
370 	struct inode *inode = file_inode(file);
371 
372 	return put_user(inode->i_generation, arg);
373 }
374 
375 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
376 {
377 	struct inode *inode = file_inode(file);
378 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
379 	struct btrfs_device *device;
380 	struct request_queue *q;
381 	struct fstrim_range range;
382 	u64 minlen = ULLONG_MAX;
383 	u64 num_devices = 0;
384 	u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
385 	int ret;
386 
387 	if (!capable(CAP_SYS_ADMIN))
388 		return -EPERM;
389 
390 	rcu_read_lock();
391 	list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
392 				dev_list) {
393 		if (!device->bdev)
394 			continue;
395 		q = bdev_get_queue(device->bdev);
396 		if (blk_queue_discard(q)) {
397 			num_devices++;
398 			minlen = min((u64)q->limits.discard_granularity,
399 				     minlen);
400 		}
401 	}
402 	rcu_read_unlock();
403 
404 	if (!num_devices)
405 		return -EOPNOTSUPP;
406 	if (copy_from_user(&range, arg, sizeof(range)))
407 		return -EFAULT;
408 	if (range.start > total_bytes ||
409 	    range.len < fs_info->sb->s_blocksize)
410 		return -EINVAL;
411 
412 	range.len = min(range.len, total_bytes - range.start);
413 	range.minlen = max(range.minlen, minlen);
414 	ret = btrfs_trim_fs(fs_info, &range);
415 	if (ret < 0)
416 		return ret;
417 
418 	if (copy_to_user(arg, &range, sizeof(range)))
419 		return -EFAULT;
420 
421 	return 0;
422 }
423 
424 int btrfs_is_empty_uuid(u8 *uuid)
425 {
426 	int i;
427 
428 	for (i = 0; i < BTRFS_UUID_SIZE; i++) {
429 		if (uuid[i])
430 			return 0;
431 	}
432 	return 1;
433 }
434 
435 static noinline int create_subvol(struct inode *dir,
436 				  struct dentry *dentry,
437 				  char *name, int namelen,
438 				  u64 *async_transid,
439 				  struct btrfs_qgroup_inherit *inherit)
440 {
441 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
442 	struct btrfs_trans_handle *trans;
443 	struct btrfs_key key;
444 	struct btrfs_root_item *root_item;
445 	struct btrfs_inode_item *inode_item;
446 	struct extent_buffer *leaf;
447 	struct btrfs_root *root = BTRFS_I(dir)->root;
448 	struct btrfs_root *new_root;
449 	struct btrfs_block_rsv block_rsv;
450 	struct timespec cur_time = current_time(dir);
451 	struct inode *inode;
452 	int ret;
453 	int err;
454 	u64 objectid;
455 	u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
456 	u64 index = 0;
457 	u64 qgroup_reserved;
458 	uuid_le new_uuid;
459 
460 	root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
461 	if (!root_item)
462 		return -ENOMEM;
463 
464 	ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid);
465 	if (ret)
466 		goto fail_free;
467 
468 	/*
469 	 * Don't create subvolume whose level is not zero. Or qgroup will be
470 	 * screwed up since it assumes subvolume qgroup's level to be 0.
471 	 */
472 	if (btrfs_qgroup_level(objectid)) {
473 		ret = -ENOSPC;
474 		goto fail_free;
475 	}
476 
477 	btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
478 	/*
479 	 * The same as the snapshot creation, please see the comment
480 	 * of create_snapshot().
481 	 */
482 	ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
483 					       8, &qgroup_reserved, false);
484 	if (ret)
485 		goto fail_free;
486 
487 	trans = btrfs_start_transaction(root, 0);
488 	if (IS_ERR(trans)) {
489 		ret = PTR_ERR(trans);
490 		btrfs_subvolume_release_metadata(fs_info, &block_rsv,
491 						 qgroup_reserved);
492 		goto fail_free;
493 	}
494 	trans->block_rsv = &block_rsv;
495 	trans->bytes_reserved = block_rsv.size;
496 
497 	ret = btrfs_qgroup_inherit(trans, fs_info, 0, objectid, inherit);
498 	if (ret)
499 		goto fail;
500 
501 	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
502 	if (IS_ERR(leaf)) {
503 		ret = PTR_ERR(leaf);
504 		goto fail;
505 	}
506 
507 	memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
508 	btrfs_set_header_bytenr(leaf, leaf->start);
509 	btrfs_set_header_generation(leaf, trans->transid);
510 	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
511 	btrfs_set_header_owner(leaf, objectid);
512 
513 	write_extent_buffer_fsid(leaf, fs_info->fsid);
514 	write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
515 	btrfs_mark_buffer_dirty(leaf);
516 
517 	inode_item = &root_item->inode;
518 	btrfs_set_stack_inode_generation(inode_item, 1);
519 	btrfs_set_stack_inode_size(inode_item, 3);
520 	btrfs_set_stack_inode_nlink(inode_item, 1);
521 	btrfs_set_stack_inode_nbytes(inode_item,
522 				     fs_info->nodesize);
523 	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
524 
525 	btrfs_set_root_flags(root_item, 0);
526 	btrfs_set_root_limit(root_item, 0);
527 	btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
528 
529 	btrfs_set_root_bytenr(root_item, leaf->start);
530 	btrfs_set_root_generation(root_item, trans->transid);
531 	btrfs_set_root_level(root_item, 0);
532 	btrfs_set_root_refs(root_item, 1);
533 	btrfs_set_root_used(root_item, leaf->len);
534 	btrfs_set_root_last_snapshot(root_item, 0);
535 
536 	btrfs_set_root_generation_v2(root_item,
537 			btrfs_root_generation(root_item));
538 	uuid_le_gen(&new_uuid);
539 	memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
540 	btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
541 	btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
542 	root_item->ctime = root_item->otime;
543 	btrfs_set_root_ctransid(root_item, trans->transid);
544 	btrfs_set_root_otransid(root_item, trans->transid);
545 
546 	btrfs_tree_unlock(leaf);
547 	free_extent_buffer(leaf);
548 	leaf = NULL;
549 
550 	btrfs_set_root_dirid(root_item, new_dirid);
551 
552 	key.objectid = objectid;
553 	key.offset = 0;
554 	key.type = BTRFS_ROOT_ITEM_KEY;
555 	ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
556 				root_item);
557 	if (ret)
558 		goto fail;
559 
560 	key.offset = (u64)-1;
561 	new_root = btrfs_read_fs_root_no_name(fs_info, &key);
562 	if (IS_ERR(new_root)) {
563 		ret = PTR_ERR(new_root);
564 		btrfs_abort_transaction(trans, ret);
565 		goto fail;
566 	}
567 
568 	btrfs_record_root_in_trans(trans, new_root);
569 
570 	ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
571 	if (ret) {
572 		/* We potentially lose an unused inode item here */
573 		btrfs_abort_transaction(trans, ret);
574 		goto fail;
575 	}
576 
577 	mutex_lock(&new_root->objectid_mutex);
578 	new_root->highest_objectid = new_dirid;
579 	mutex_unlock(&new_root->objectid_mutex);
580 
581 	/*
582 	 * insert the directory item
583 	 */
584 	ret = btrfs_set_inode_index(dir, &index);
585 	if (ret) {
586 		btrfs_abort_transaction(trans, ret);
587 		goto fail;
588 	}
589 
590 	ret = btrfs_insert_dir_item(trans, root,
591 				    name, namelen, dir, &key,
592 				    BTRFS_FT_DIR, index);
593 	if (ret) {
594 		btrfs_abort_transaction(trans, ret);
595 		goto fail;
596 	}
597 
598 	btrfs_i_size_write(dir, dir->i_size + namelen * 2);
599 	ret = btrfs_update_inode(trans, root, dir);
600 	BUG_ON(ret);
601 
602 	ret = btrfs_add_root_ref(trans, fs_info,
603 				 objectid, root->root_key.objectid,
604 				 btrfs_ino(dir), index, name, namelen);
605 	BUG_ON(ret);
606 
607 	ret = btrfs_uuid_tree_add(trans, fs_info, root_item->uuid,
608 				  BTRFS_UUID_KEY_SUBVOL, objectid);
609 	if (ret)
610 		btrfs_abort_transaction(trans, ret);
611 
612 fail:
613 	kfree(root_item);
614 	trans->block_rsv = NULL;
615 	trans->bytes_reserved = 0;
616 	btrfs_subvolume_release_metadata(fs_info, &block_rsv, qgroup_reserved);
617 
618 	if (async_transid) {
619 		*async_transid = trans->transid;
620 		err = btrfs_commit_transaction_async(trans, 1);
621 		if (err)
622 			err = btrfs_commit_transaction(trans);
623 	} else {
624 		err = btrfs_commit_transaction(trans);
625 	}
626 	if (err && !ret)
627 		ret = err;
628 
629 	if (!ret) {
630 		inode = btrfs_lookup_dentry(dir, dentry);
631 		if (IS_ERR(inode))
632 			return PTR_ERR(inode);
633 		d_instantiate(dentry, inode);
634 	}
635 	return ret;
636 
637 fail_free:
638 	kfree(root_item);
639 	return ret;
640 }
641 
642 static void btrfs_wait_for_no_snapshoting_writes(struct btrfs_root *root)
643 {
644 	s64 writers;
645 	DEFINE_WAIT(wait);
646 
647 	do {
648 		prepare_to_wait(&root->subv_writers->wait, &wait,
649 				TASK_UNINTERRUPTIBLE);
650 
651 		writers = percpu_counter_sum(&root->subv_writers->counter);
652 		if (writers)
653 			schedule();
654 
655 		finish_wait(&root->subv_writers->wait, &wait);
656 	} while (writers);
657 }
658 
659 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
660 			   struct dentry *dentry, char *name, int namelen,
661 			   u64 *async_transid, bool readonly,
662 			   struct btrfs_qgroup_inherit *inherit)
663 {
664 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
665 	struct inode *inode;
666 	struct btrfs_pending_snapshot *pending_snapshot;
667 	struct btrfs_trans_handle *trans;
668 	int ret;
669 
670 	if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
671 		return -EINVAL;
672 
673 	pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
674 	if (!pending_snapshot)
675 		return -ENOMEM;
676 
677 	pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
678 			GFP_NOFS);
679 	pending_snapshot->path = btrfs_alloc_path();
680 	if (!pending_snapshot->root_item || !pending_snapshot->path) {
681 		ret = -ENOMEM;
682 		goto free_pending;
683 	}
684 
685 	atomic_inc(&root->will_be_snapshoted);
686 	smp_mb__after_atomic();
687 	btrfs_wait_for_no_snapshoting_writes(root);
688 
689 	ret = btrfs_start_delalloc_inodes(root, 0);
690 	if (ret)
691 		goto dec_and_free;
692 
693 	btrfs_wait_ordered_extents(root, -1, 0, (u64)-1);
694 
695 	btrfs_init_block_rsv(&pending_snapshot->block_rsv,
696 			     BTRFS_BLOCK_RSV_TEMP);
697 	/*
698 	 * 1 - parent dir inode
699 	 * 2 - dir entries
700 	 * 1 - root item
701 	 * 2 - root ref/backref
702 	 * 1 - root of snapshot
703 	 * 1 - UUID item
704 	 */
705 	ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
706 					&pending_snapshot->block_rsv, 8,
707 					&pending_snapshot->qgroup_reserved,
708 					false);
709 	if (ret)
710 		goto dec_and_free;
711 
712 	pending_snapshot->dentry = dentry;
713 	pending_snapshot->root = root;
714 	pending_snapshot->readonly = readonly;
715 	pending_snapshot->dir = dir;
716 	pending_snapshot->inherit = inherit;
717 
718 	trans = btrfs_start_transaction(root, 0);
719 	if (IS_ERR(trans)) {
720 		ret = PTR_ERR(trans);
721 		goto fail;
722 	}
723 
724 	spin_lock(&fs_info->trans_lock);
725 	list_add(&pending_snapshot->list,
726 		 &trans->transaction->pending_snapshots);
727 	spin_unlock(&fs_info->trans_lock);
728 	if (async_transid) {
729 		*async_transid = trans->transid;
730 		ret = btrfs_commit_transaction_async(trans, 1);
731 		if (ret)
732 			ret = btrfs_commit_transaction(trans);
733 	} else {
734 		ret = btrfs_commit_transaction(trans);
735 	}
736 	if (ret)
737 		goto fail;
738 
739 	ret = pending_snapshot->error;
740 	if (ret)
741 		goto fail;
742 
743 	ret = btrfs_orphan_cleanup(pending_snapshot->snap);
744 	if (ret)
745 		goto fail;
746 
747 	inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
748 	if (IS_ERR(inode)) {
749 		ret = PTR_ERR(inode);
750 		goto fail;
751 	}
752 
753 	d_instantiate(dentry, inode);
754 	ret = 0;
755 fail:
756 	btrfs_subvolume_release_metadata(fs_info,
757 					 &pending_snapshot->block_rsv,
758 					 pending_snapshot->qgroup_reserved);
759 dec_and_free:
760 	if (atomic_dec_and_test(&root->will_be_snapshoted))
761 		wake_up_atomic_t(&root->will_be_snapshoted);
762 free_pending:
763 	kfree(pending_snapshot->root_item);
764 	btrfs_free_path(pending_snapshot->path);
765 	kfree(pending_snapshot);
766 
767 	return ret;
768 }
769 
770 /*  copy of may_delete in fs/namei.c()
771  *	Check whether we can remove a link victim from directory dir, check
772  *  whether the type of victim is right.
773  *  1. We can't do it if dir is read-only (done in permission())
774  *  2. We should have write and exec permissions on dir
775  *  3. We can't remove anything from append-only dir
776  *  4. We can't do anything with immutable dir (done in permission())
777  *  5. If the sticky bit on dir is set we should either
778  *	a. be owner of dir, or
779  *	b. be owner of victim, or
780  *	c. have CAP_FOWNER capability
781  *  6. If the victim is append-only or immutable we can't do anything with
782  *     links pointing to it.
783  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
784  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
785  *  9. We can't remove a root or mountpoint.
786  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
787  *     nfs_async_unlink().
788  */
789 
790 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
791 {
792 	int error;
793 
794 	if (d_really_is_negative(victim))
795 		return -ENOENT;
796 
797 	BUG_ON(d_inode(victim->d_parent) != dir);
798 	audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
799 
800 	error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
801 	if (error)
802 		return error;
803 	if (IS_APPEND(dir))
804 		return -EPERM;
805 	if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
806 	    IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
807 		return -EPERM;
808 	if (isdir) {
809 		if (!d_is_dir(victim))
810 			return -ENOTDIR;
811 		if (IS_ROOT(victim))
812 			return -EBUSY;
813 	} else if (d_is_dir(victim))
814 		return -EISDIR;
815 	if (IS_DEADDIR(dir))
816 		return -ENOENT;
817 	if (victim->d_flags & DCACHE_NFSFS_RENAMED)
818 		return -EBUSY;
819 	return 0;
820 }
821 
822 /* copy of may_create in fs/namei.c() */
823 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
824 {
825 	if (d_really_is_positive(child))
826 		return -EEXIST;
827 	if (IS_DEADDIR(dir))
828 		return -ENOENT;
829 	return inode_permission(dir, MAY_WRITE | MAY_EXEC);
830 }
831 
832 /*
833  * Create a new subvolume below @parent.  This is largely modeled after
834  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
835  * inside this filesystem so it's quite a bit simpler.
836  */
837 static noinline int btrfs_mksubvol(const struct path *parent,
838 				   char *name, int namelen,
839 				   struct btrfs_root *snap_src,
840 				   u64 *async_transid, bool readonly,
841 				   struct btrfs_qgroup_inherit *inherit)
842 {
843 	struct inode *dir = d_inode(parent->dentry);
844 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
845 	struct dentry *dentry;
846 	int error;
847 
848 	error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
849 	if (error == -EINTR)
850 		return error;
851 
852 	dentry = lookup_one_len(name, parent->dentry, namelen);
853 	error = PTR_ERR(dentry);
854 	if (IS_ERR(dentry))
855 		goto out_unlock;
856 
857 	error = btrfs_may_create(dir, dentry);
858 	if (error)
859 		goto out_dput;
860 
861 	/*
862 	 * even if this name doesn't exist, we may get hash collisions.
863 	 * check for them now when we can safely fail
864 	 */
865 	error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
866 					       dir->i_ino, name,
867 					       namelen);
868 	if (error)
869 		goto out_dput;
870 
871 	down_read(&fs_info->subvol_sem);
872 
873 	if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
874 		goto out_up_read;
875 
876 	if (snap_src) {
877 		error = create_snapshot(snap_src, dir, dentry, name, namelen,
878 					async_transid, readonly, inherit);
879 	} else {
880 		error = create_subvol(dir, dentry, name, namelen,
881 				      async_transid, inherit);
882 	}
883 	if (!error)
884 		fsnotify_mkdir(dir, dentry);
885 out_up_read:
886 	up_read(&fs_info->subvol_sem);
887 out_dput:
888 	dput(dentry);
889 out_unlock:
890 	inode_unlock(dir);
891 	return error;
892 }
893 
894 /*
895  * When we're defragging a range, we don't want to kick it off again
896  * if it is really just waiting for delalloc to send it down.
897  * If we find a nice big extent or delalloc range for the bytes in the
898  * file you want to defrag, we return 0 to let you know to skip this
899  * part of the file
900  */
901 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
902 {
903 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
904 	struct extent_map *em = NULL;
905 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
906 	u64 end;
907 
908 	read_lock(&em_tree->lock);
909 	em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
910 	read_unlock(&em_tree->lock);
911 
912 	if (em) {
913 		end = extent_map_end(em);
914 		free_extent_map(em);
915 		if (end - offset > thresh)
916 			return 0;
917 	}
918 	/* if we already have a nice delalloc here, just stop */
919 	thresh /= 2;
920 	end = count_range_bits(io_tree, &offset, offset + thresh,
921 			       thresh, EXTENT_DELALLOC, 1);
922 	if (end >= thresh)
923 		return 0;
924 	return 1;
925 }
926 
927 /*
928  * helper function to walk through a file and find extents
929  * newer than a specific transid, and smaller than thresh.
930  *
931  * This is used by the defragging code to find new and small
932  * extents
933  */
934 static int find_new_extents(struct btrfs_root *root,
935 			    struct inode *inode, u64 newer_than,
936 			    u64 *off, u32 thresh)
937 {
938 	struct btrfs_path *path;
939 	struct btrfs_key min_key;
940 	struct extent_buffer *leaf;
941 	struct btrfs_file_extent_item *extent;
942 	int type;
943 	int ret;
944 	u64 ino = btrfs_ino(inode);
945 
946 	path = btrfs_alloc_path();
947 	if (!path)
948 		return -ENOMEM;
949 
950 	min_key.objectid = ino;
951 	min_key.type = BTRFS_EXTENT_DATA_KEY;
952 	min_key.offset = *off;
953 
954 	while (1) {
955 		ret = btrfs_search_forward(root, &min_key, path, newer_than);
956 		if (ret != 0)
957 			goto none;
958 process_slot:
959 		if (min_key.objectid != ino)
960 			goto none;
961 		if (min_key.type != BTRFS_EXTENT_DATA_KEY)
962 			goto none;
963 
964 		leaf = path->nodes[0];
965 		extent = btrfs_item_ptr(leaf, path->slots[0],
966 					struct btrfs_file_extent_item);
967 
968 		type = btrfs_file_extent_type(leaf, extent);
969 		if (type == BTRFS_FILE_EXTENT_REG &&
970 		    btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
971 		    check_defrag_in_cache(inode, min_key.offset, thresh)) {
972 			*off = min_key.offset;
973 			btrfs_free_path(path);
974 			return 0;
975 		}
976 
977 		path->slots[0]++;
978 		if (path->slots[0] < btrfs_header_nritems(leaf)) {
979 			btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
980 			goto process_slot;
981 		}
982 
983 		if (min_key.offset == (u64)-1)
984 			goto none;
985 
986 		min_key.offset++;
987 		btrfs_release_path(path);
988 	}
989 none:
990 	btrfs_free_path(path);
991 	return -ENOENT;
992 }
993 
994 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
995 {
996 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
997 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
998 	struct extent_map *em;
999 	u64 len = PAGE_SIZE;
1000 
1001 	/*
1002 	 * hopefully we have this extent in the tree already, try without
1003 	 * the full extent lock
1004 	 */
1005 	read_lock(&em_tree->lock);
1006 	em = lookup_extent_mapping(em_tree, start, len);
1007 	read_unlock(&em_tree->lock);
1008 
1009 	if (!em) {
1010 		struct extent_state *cached = NULL;
1011 		u64 end = start + len - 1;
1012 
1013 		/* get the big lock and read metadata off disk */
1014 		lock_extent_bits(io_tree, start, end, &cached);
1015 		em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
1016 		unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
1017 
1018 		if (IS_ERR(em))
1019 			return NULL;
1020 	}
1021 
1022 	return em;
1023 }
1024 
1025 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1026 {
1027 	struct extent_map *next;
1028 	bool ret = true;
1029 
1030 	/* this is the last extent */
1031 	if (em->start + em->len >= i_size_read(inode))
1032 		return false;
1033 
1034 	next = defrag_lookup_extent(inode, em->start + em->len);
1035 	if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1036 		ret = false;
1037 	else if ((em->block_start + em->block_len == next->block_start) &&
1038 		 (em->block_len > SZ_128K && next->block_len > SZ_128K))
1039 		ret = false;
1040 
1041 	free_extent_map(next);
1042 	return ret;
1043 }
1044 
1045 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1046 			       u64 *last_len, u64 *skip, u64 *defrag_end,
1047 			       int compress)
1048 {
1049 	struct extent_map *em;
1050 	int ret = 1;
1051 	bool next_mergeable = true;
1052 	bool prev_mergeable = true;
1053 
1054 	/*
1055 	 * make sure that once we start defragging an extent, we keep on
1056 	 * defragging it
1057 	 */
1058 	if (start < *defrag_end)
1059 		return 1;
1060 
1061 	*skip = 0;
1062 
1063 	em = defrag_lookup_extent(inode, start);
1064 	if (!em)
1065 		return 0;
1066 
1067 	/* this will cover holes, and inline extents */
1068 	if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1069 		ret = 0;
1070 		goto out;
1071 	}
1072 
1073 	if (!*defrag_end)
1074 		prev_mergeable = false;
1075 
1076 	next_mergeable = defrag_check_next_extent(inode, em);
1077 	/*
1078 	 * we hit a real extent, if it is big or the next extent is not a
1079 	 * real extent, don't bother defragging it
1080 	 */
1081 	if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1082 	    (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1083 		ret = 0;
1084 out:
1085 	/*
1086 	 * last_len ends up being a counter of how many bytes we've defragged.
1087 	 * every time we choose not to defrag an extent, we reset *last_len
1088 	 * so that the next tiny extent will force a defrag.
1089 	 *
1090 	 * The end result of this is that tiny extents before a single big
1091 	 * extent will force at least part of that big extent to be defragged.
1092 	 */
1093 	if (ret) {
1094 		*defrag_end = extent_map_end(em);
1095 	} else {
1096 		*last_len = 0;
1097 		*skip = extent_map_end(em);
1098 		*defrag_end = 0;
1099 	}
1100 
1101 	free_extent_map(em);
1102 	return ret;
1103 }
1104 
1105 /*
1106  * it doesn't do much good to defrag one or two pages
1107  * at a time.  This pulls in a nice chunk of pages
1108  * to COW and defrag.
1109  *
1110  * It also makes sure the delalloc code has enough
1111  * dirty data to avoid making new small extents as part
1112  * of the defrag
1113  *
1114  * It's a good idea to start RA on this range
1115  * before calling this.
1116  */
1117 static int cluster_pages_for_defrag(struct inode *inode,
1118 				    struct page **pages,
1119 				    unsigned long start_index,
1120 				    unsigned long num_pages)
1121 {
1122 	unsigned long file_end;
1123 	u64 isize = i_size_read(inode);
1124 	u64 page_start;
1125 	u64 page_end;
1126 	u64 page_cnt;
1127 	int ret;
1128 	int i;
1129 	int i_done;
1130 	struct btrfs_ordered_extent *ordered;
1131 	struct extent_state *cached_state = NULL;
1132 	struct extent_io_tree *tree;
1133 	gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1134 
1135 	file_end = (isize - 1) >> PAGE_SHIFT;
1136 	if (!isize || start_index > file_end)
1137 		return 0;
1138 
1139 	page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1140 
1141 	ret = btrfs_delalloc_reserve_space(inode,
1142 			start_index << PAGE_SHIFT,
1143 			page_cnt << PAGE_SHIFT);
1144 	if (ret)
1145 		return ret;
1146 	i_done = 0;
1147 	tree = &BTRFS_I(inode)->io_tree;
1148 
1149 	/* step one, lock all the pages */
1150 	for (i = 0; i < page_cnt; i++) {
1151 		struct page *page;
1152 again:
1153 		page = find_or_create_page(inode->i_mapping,
1154 					   start_index + i, mask);
1155 		if (!page)
1156 			break;
1157 
1158 		page_start = page_offset(page);
1159 		page_end = page_start + PAGE_SIZE - 1;
1160 		while (1) {
1161 			lock_extent_bits(tree, page_start, page_end,
1162 					 &cached_state);
1163 			ordered = btrfs_lookup_ordered_extent(inode,
1164 							      page_start);
1165 			unlock_extent_cached(tree, page_start, page_end,
1166 					     &cached_state, GFP_NOFS);
1167 			if (!ordered)
1168 				break;
1169 
1170 			unlock_page(page);
1171 			btrfs_start_ordered_extent(inode, ordered, 1);
1172 			btrfs_put_ordered_extent(ordered);
1173 			lock_page(page);
1174 			/*
1175 			 * we unlocked the page above, so we need check if
1176 			 * it was released or not.
1177 			 */
1178 			if (page->mapping != inode->i_mapping) {
1179 				unlock_page(page);
1180 				put_page(page);
1181 				goto again;
1182 			}
1183 		}
1184 
1185 		if (!PageUptodate(page)) {
1186 			btrfs_readpage(NULL, page);
1187 			lock_page(page);
1188 			if (!PageUptodate(page)) {
1189 				unlock_page(page);
1190 				put_page(page);
1191 				ret = -EIO;
1192 				break;
1193 			}
1194 		}
1195 
1196 		if (page->mapping != inode->i_mapping) {
1197 			unlock_page(page);
1198 			put_page(page);
1199 			goto again;
1200 		}
1201 
1202 		pages[i] = page;
1203 		i_done++;
1204 	}
1205 	if (!i_done || ret)
1206 		goto out;
1207 
1208 	if (!(inode->i_sb->s_flags & MS_ACTIVE))
1209 		goto out;
1210 
1211 	/*
1212 	 * so now we have a nice long stream of locked
1213 	 * and up to date pages, lets wait on them
1214 	 */
1215 	for (i = 0; i < i_done; i++)
1216 		wait_on_page_writeback(pages[i]);
1217 
1218 	page_start = page_offset(pages[0]);
1219 	page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1220 
1221 	lock_extent_bits(&BTRFS_I(inode)->io_tree,
1222 			 page_start, page_end - 1, &cached_state);
1223 	clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1224 			  page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1225 			  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1226 			  &cached_state, GFP_NOFS);
1227 
1228 	if (i_done != page_cnt) {
1229 		spin_lock(&BTRFS_I(inode)->lock);
1230 		BTRFS_I(inode)->outstanding_extents++;
1231 		spin_unlock(&BTRFS_I(inode)->lock);
1232 		btrfs_delalloc_release_space(inode,
1233 				start_index << PAGE_SHIFT,
1234 				(page_cnt - i_done) << PAGE_SHIFT);
1235 	}
1236 
1237 
1238 	set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1239 			  &cached_state);
1240 
1241 	unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1242 			     page_start, page_end - 1, &cached_state,
1243 			     GFP_NOFS);
1244 
1245 	for (i = 0; i < i_done; i++) {
1246 		clear_page_dirty_for_io(pages[i]);
1247 		ClearPageChecked(pages[i]);
1248 		set_page_extent_mapped(pages[i]);
1249 		set_page_dirty(pages[i]);
1250 		unlock_page(pages[i]);
1251 		put_page(pages[i]);
1252 	}
1253 	return i_done;
1254 out:
1255 	for (i = 0; i < i_done; i++) {
1256 		unlock_page(pages[i]);
1257 		put_page(pages[i]);
1258 	}
1259 	btrfs_delalloc_release_space(inode,
1260 			start_index << PAGE_SHIFT,
1261 			page_cnt << PAGE_SHIFT);
1262 	return ret;
1263 
1264 }
1265 
1266 int btrfs_defrag_file(struct inode *inode, struct file *file,
1267 		      struct btrfs_ioctl_defrag_range_args *range,
1268 		      u64 newer_than, unsigned long max_to_defrag)
1269 {
1270 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1271 	struct btrfs_root *root = BTRFS_I(inode)->root;
1272 	struct file_ra_state *ra = NULL;
1273 	unsigned long last_index;
1274 	u64 isize = i_size_read(inode);
1275 	u64 last_len = 0;
1276 	u64 skip = 0;
1277 	u64 defrag_end = 0;
1278 	u64 newer_off = range->start;
1279 	unsigned long i;
1280 	unsigned long ra_index = 0;
1281 	int ret;
1282 	int defrag_count = 0;
1283 	int compress_type = BTRFS_COMPRESS_ZLIB;
1284 	u32 extent_thresh = range->extent_thresh;
1285 	unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1286 	unsigned long cluster = max_cluster;
1287 	u64 new_align = ~((u64)SZ_128K - 1);
1288 	struct page **pages = NULL;
1289 
1290 	if (isize == 0)
1291 		return 0;
1292 
1293 	if (range->start >= isize)
1294 		return -EINVAL;
1295 
1296 	if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1297 		if (range->compress_type > BTRFS_COMPRESS_TYPES)
1298 			return -EINVAL;
1299 		if (range->compress_type)
1300 			compress_type = range->compress_type;
1301 	}
1302 
1303 	if (extent_thresh == 0)
1304 		extent_thresh = SZ_256K;
1305 
1306 	/*
1307 	 * if we were not given a file, allocate a readahead
1308 	 * context
1309 	 */
1310 	if (!file) {
1311 		ra = kzalloc(sizeof(*ra), GFP_NOFS);
1312 		if (!ra)
1313 			return -ENOMEM;
1314 		file_ra_state_init(ra, inode->i_mapping);
1315 	} else {
1316 		ra = &file->f_ra;
1317 	}
1318 
1319 	pages = kmalloc_array(max_cluster, sizeof(struct page *),
1320 			GFP_NOFS);
1321 	if (!pages) {
1322 		ret = -ENOMEM;
1323 		goto out_ra;
1324 	}
1325 
1326 	/* find the last page to defrag */
1327 	if (range->start + range->len > range->start) {
1328 		last_index = min_t(u64, isize - 1,
1329 			 range->start + range->len - 1) >> PAGE_SHIFT;
1330 	} else {
1331 		last_index = (isize - 1) >> PAGE_SHIFT;
1332 	}
1333 
1334 	if (newer_than) {
1335 		ret = find_new_extents(root, inode, newer_than,
1336 				       &newer_off, SZ_64K);
1337 		if (!ret) {
1338 			range->start = newer_off;
1339 			/*
1340 			 * we always align our defrag to help keep
1341 			 * the extents in the file evenly spaced
1342 			 */
1343 			i = (newer_off & new_align) >> PAGE_SHIFT;
1344 		} else
1345 			goto out_ra;
1346 	} else {
1347 		i = range->start >> PAGE_SHIFT;
1348 	}
1349 	if (!max_to_defrag)
1350 		max_to_defrag = last_index - i + 1;
1351 
1352 	/*
1353 	 * make writeback starts from i, so the defrag range can be
1354 	 * written sequentially.
1355 	 */
1356 	if (i < inode->i_mapping->writeback_index)
1357 		inode->i_mapping->writeback_index = i;
1358 
1359 	while (i <= last_index && defrag_count < max_to_defrag &&
1360 	       (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1361 		/*
1362 		 * make sure we stop running if someone unmounts
1363 		 * the FS
1364 		 */
1365 		if (!(inode->i_sb->s_flags & MS_ACTIVE))
1366 			break;
1367 
1368 		if (btrfs_defrag_cancelled(fs_info)) {
1369 			btrfs_debug(fs_info, "defrag_file cancelled");
1370 			ret = -EAGAIN;
1371 			break;
1372 		}
1373 
1374 		if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1375 					 extent_thresh, &last_len, &skip,
1376 					 &defrag_end, range->flags &
1377 					 BTRFS_DEFRAG_RANGE_COMPRESS)) {
1378 			unsigned long next;
1379 			/*
1380 			 * the should_defrag function tells us how much to skip
1381 			 * bump our counter by the suggested amount
1382 			 */
1383 			next = DIV_ROUND_UP(skip, PAGE_SIZE);
1384 			i = max(i + 1, next);
1385 			continue;
1386 		}
1387 
1388 		if (!newer_than) {
1389 			cluster = (PAGE_ALIGN(defrag_end) >>
1390 				   PAGE_SHIFT) - i;
1391 			cluster = min(cluster, max_cluster);
1392 		} else {
1393 			cluster = max_cluster;
1394 		}
1395 
1396 		if (i + cluster > ra_index) {
1397 			ra_index = max(i, ra_index);
1398 			btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1399 				       cluster);
1400 			ra_index += cluster;
1401 		}
1402 
1403 		inode_lock(inode);
1404 		if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1405 			BTRFS_I(inode)->force_compress = compress_type;
1406 		ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1407 		if (ret < 0) {
1408 			inode_unlock(inode);
1409 			goto out_ra;
1410 		}
1411 
1412 		defrag_count += ret;
1413 		balance_dirty_pages_ratelimited(inode->i_mapping);
1414 		inode_unlock(inode);
1415 
1416 		if (newer_than) {
1417 			if (newer_off == (u64)-1)
1418 				break;
1419 
1420 			if (ret > 0)
1421 				i += ret;
1422 
1423 			newer_off = max(newer_off + 1,
1424 					(u64)i << PAGE_SHIFT);
1425 
1426 			ret = find_new_extents(root, inode, newer_than,
1427 					       &newer_off, SZ_64K);
1428 			if (!ret) {
1429 				range->start = newer_off;
1430 				i = (newer_off & new_align) >> PAGE_SHIFT;
1431 			} else {
1432 				break;
1433 			}
1434 		} else {
1435 			if (ret > 0) {
1436 				i += ret;
1437 				last_len += ret << PAGE_SHIFT;
1438 			} else {
1439 				i++;
1440 				last_len = 0;
1441 			}
1442 		}
1443 	}
1444 
1445 	if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1446 		filemap_flush(inode->i_mapping);
1447 		if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1448 			     &BTRFS_I(inode)->runtime_flags))
1449 			filemap_flush(inode->i_mapping);
1450 	}
1451 
1452 	if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1453 		/* the filemap_flush will queue IO into the worker threads, but
1454 		 * we have to make sure the IO is actually started and that
1455 		 * ordered extents get created before we return
1456 		 */
1457 		atomic_inc(&fs_info->async_submit_draining);
1458 		while (atomic_read(&fs_info->nr_async_submits) ||
1459 		       atomic_read(&fs_info->async_delalloc_pages)) {
1460 			wait_event(fs_info->async_submit_wait,
1461 				   (atomic_read(&fs_info->nr_async_submits) == 0 &&
1462 				    atomic_read(&fs_info->async_delalloc_pages) == 0));
1463 		}
1464 		atomic_dec(&fs_info->async_submit_draining);
1465 	}
1466 
1467 	if (range->compress_type == BTRFS_COMPRESS_LZO) {
1468 		btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1469 	}
1470 
1471 	ret = defrag_count;
1472 
1473 out_ra:
1474 	if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1475 		inode_lock(inode);
1476 		BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1477 		inode_unlock(inode);
1478 	}
1479 	if (!file)
1480 		kfree(ra);
1481 	kfree(pages);
1482 	return ret;
1483 }
1484 
1485 static noinline int btrfs_ioctl_resize(struct file *file,
1486 					void __user *arg)
1487 {
1488 	struct inode *inode = file_inode(file);
1489 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1490 	u64 new_size;
1491 	u64 old_size;
1492 	u64 devid = 1;
1493 	struct btrfs_root *root = BTRFS_I(inode)->root;
1494 	struct btrfs_ioctl_vol_args *vol_args;
1495 	struct btrfs_trans_handle *trans;
1496 	struct btrfs_device *device = NULL;
1497 	char *sizestr;
1498 	char *retptr;
1499 	char *devstr = NULL;
1500 	int ret = 0;
1501 	int mod = 0;
1502 
1503 	if (!capable(CAP_SYS_ADMIN))
1504 		return -EPERM;
1505 
1506 	ret = mnt_want_write_file(file);
1507 	if (ret)
1508 		return ret;
1509 
1510 	if (atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
1511 		mnt_drop_write_file(file);
1512 		return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1513 	}
1514 
1515 	mutex_lock(&fs_info->volume_mutex);
1516 	vol_args = memdup_user(arg, sizeof(*vol_args));
1517 	if (IS_ERR(vol_args)) {
1518 		ret = PTR_ERR(vol_args);
1519 		goto out;
1520 	}
1521 
1522 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1523 
1524 	sizestr = vol_args->name;
1525 	devstr = strchr(sizestr, ':');
1526 	if (devstr) {
1527 		sizestr = devstr + 1;
1528 		*devstr = '\0';
1529 		devstr = vol_args->name;
1530 		ret = kstrtoull(devstr, 10, &devid);
1531 		if (ret)
1532 			goto out_free;
1533 		if (!devid) {
1534 			ret = -EINVAL;
1535 			goto out_free;
1536 		}
1537 		btrfs_info(fs_info, "resizing devid %llu", devid);
1538 	}
1539 
1540 	device = btrfs_find_device(fs_info, devid, NULL, NULL);
1541 	if (!device) {
1542 		btrfs_info(fs_info, "resizer unable to find device %llu",
1543 			   devid);
1544 		ret = -ENODEV;
1545 		goto out_free;
1546 	}
1547 
1548 	if (!device->writeable) {
1549 		btrfs_info(fs_info,
1550 			   "resizer unable to apply on readonly device %llu",
1551 		       devid);
1552 		ret = -EPERM;
1553 		goto out_free;
1554 	}
1555 
1556 	if (!strcmp(sizestr, "max"))
1557 		new_size = device->bdev->bd_inode->i_size;
1558 	else {
1559 		if (sizestr[0] == '-') {
1560 			mod = -1;
1561 			sizestr++;
1562 		} else if (sizestr[0] == '+') {
1563 			mod = 1;
1564 			sizestr++;
1565 		}
1566 		new_size = memparse(sizestr, &retptr);
1567 		if (*retptr != '\0' || new_size == 0) {
1568 			ret = -EINVAL;
1569 			goto out_free;
1570 		}
1571 	}
1572 
1573 	if (device->is_tgtdev_for_dev_replace) {
1574 		ret = -EPERM;
1575 		goto out_free;
1576 	}
1577 
1578 	old_size = btrfs_device_get_total_bytes(device);
1579 
1580 	if (mod < 0) {
1581 		if (new_size > old_size) {
1582 			ret = -EINVAL;
1583 			goto out_free;
1584 		}
1585 		new_size = old_size - new_size;
1586 	} else if (mod > 0) {
1587 		if (new_size > ULLONG_MAX - old_size) {
1588 			ret = -ERANGE;
1589 			goto out_free;
1590 		}
1591 		new_size = old_size + new_size;
1592 	}
1593 
1594 	if (new_size < SZ_256M) {
1595 		ret = -EINVAL;
1596 		goto out_free;
1597 	}
1598 	if (new_size > device->bdev->bd_inode->i_size) {
1599 		ret = -EFBIG;
1600 		goto out_free;
1601 	}
1602 
1603 	new_size = div_u64(new_size, fs_info->sectorsize);
1604 	new_size *= fs_info->sectorsize;
1605 
1606 	btrfs_info_in_rcu(fs_info, "new size for %s is %llu",
1607 			  rcu_str_deref(device->name), new_size);
1608 
1609 	if (new_size > old_size) {
1610 		trans = btrfs_start_transaction(root, 0);
1611 		if (IS_ERR(trans)) {
1612 			ret = PTR_ERR(trans);
1613 			goto out_free;
1614 		}
1615 		ret = btrfs_grow_device(trans, device, new_size);
1616 		btrfs_commit_transaction(trans);
1617 	} else if (new_size < old_size) {
1618 		ret = btrfs_shrink_device(device, new_size);
1619 	} /* equal, nothing need to do */
1620 
1621 out_free:
1622 	kfree(vol_args);
1623 out:
1624 	mutex_unlock(&fs_info->volume_mutex);
1625 	atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
1626 	mnt_drop_write_file(file);
1627 	return ret;
1628 }
1629 
1630 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1631 				char *name, unsigned long fd, int subvol,
1632 				u64 *transid, bool readonly,
1633 				struct btrfs_qgroup_inherit *inherit)
1634 {
1635 	int namelen;
1636 	int ret = 0;
1637 
1638 	if (!S_ISDIR(file_inode(file)->i_mode))
1639 		return -ENOTDIR;
1640 
1641 	ret = mnt_want_write_file(file);
1642 	if (ret)
1643 		goto out;
1644 
1645 	namelen = strlen(name);
1646 	if (strchr(name, '/')) {
1647 		ret = -EINVAL;
1648 		goto out_drop_write;
1649 	}
1650 
1651 	if (name[0] == '.' &&
1652 	   (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1653 		ret = -EEXIST;
1654 		goto out_drop_write;
1655 	}
1656 
1657 	if (subvol) {
1658 		ret = btrfs_mksubvol(&file->f_path, name, namelen,
1659 				     NULL, transid, readonly, inherit);
1660 	} else {
1661 		struct fd src = fdget(fd);
1662 		struct inode *src_inode;
1663 		if (!src.file) {
1664 			ret = -EINVAL;
1665 			goto out_drop_write;
1666 		}
1667 
1668 		src_inode = file_inode(src.file);
1669 		if (src_inode->i_sb != file_inode(file)->i_sb) {
1670 			btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1671 				   "Snapshot src from another FS");
1672 			ret = -EXDEV;
1673 		} else if (!inode_owner_or_capable(src_inode)) {
1674 			/*
1675 			 * Subvolume creation is not restricted, but snapshots
1676 			 * are limited to own subvolumes only
1677 			 */
1678 			ret = -EPERM;
1679 		} else {
1680 			ret = btrfs_mksubvol(&file->f_path, name, namelen,
1681 					     BTRFS_I(src_inode)->root,
1682 					     transid, readonly, inherit);
1683 		}
1684 		fdput(src);
1685 	}
1686 out_drop_write:
1687 	mnt_drop_write_file(file);
1688 out:
1689 	return ret;
1690 }
1691 
1692 static noinline int btrfs_ioctl_snap_create(struct file *file,
1693 					    void __user *arg, int subvol)
1694 {
1695 	struct btrfs_ioctl_vol_args *vol_args;
1696 	int ret;
1697 
1698 	if (!S_ISDIR(file_inode(file)->i_mode))
1699 		return -ENOTDIR;
1700 
1701 	vol_args = memdup_user(arg, sizeof(*vol_args));
1702 	if (IS_ERR(vol_args))
1703 		return PTR_ERR(vol_args);
1704 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1705 
1706 	ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1707 					      vol_args->fd, subvol,
1708 					      NULL, false, NULL);
1709 
1710 	kfree(vol_args);
1711 	return ret;
1712 }
1713 
1714 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1715 					       void __user *arg, int subvol)
1716 {
1717 	struct btrfs_ioctl_vol_args_v2 *vol_args;
1718 	int ret;
1719 	u64 transid = 0;
1720 	u64 *ptr = NULL;
1721 	bool readonly = false;
1722 	struct btrfs_qgroup_inherit *inherit = NULL;
1723 
1724 	if (!S_ISDIR(file_inode(file)->i_mode))
1725 		return -ENOTDIR;
1726 
1727 	vol_args = memdup_user(arg, sizeof(*vol_args));
1728 	if (IS_ERR(vol_args))
1729 		return PTR_ERR(vol_args);
1730 	vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1731 
1732 	if (vol_args->flags &
1733 	    ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1734 	      BTRFS_SUBVOL_QGROUP_INHERIT)) {
1735 		ret = -EOPNOTSUPP;
1736 		goto free_args;
1737 	}
1738 
1739 	if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1740 		ptr = &transid;
1741 	if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1742 		readonly = true;
1743 	if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1744 		if (vol_args->size > PAGE_SIZE) {
1745 			ret = -EINVAL;
1746 			goto free_args;
1747 		}
1748 		inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1749 		if (IS_ERR(inherit)) {
1750 			ret = PTR_ERR(inherit);
1751 			goto free_args;
1752 		}
1753 	}
1754 
1755 	ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1756 					      vol_args->fd, subvol, ptr,
1757 					      readonly, inherit);
1758 	if (ret)
1759 		goto free_inherit;
1760 
1761 	if (ptr && copy_to_user(arg +
1762 				offsetof(struct btrfs_ioctl_vol_args_v2,
1763 					transid),
1764 				ptr, sizeof(*ptr)))
1765 		ret = -EFAULT;
1766 
1767 free_inherit:
1768 	kfree(inherit);
1769 free_args:
1770 	kfree(vol_args);
1771 	return ret;
1772 }
1773 
1774 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1775 						void __user *arg)
1776 {
1777 	struct inode *inode = file_inode(file);
1778 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1779 	struct btrfs_root *root = BTRFS_I(inode)->root;
1780 	int ret = 0;
1781 	u64 flags = 0;
1782 
1783 	if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1784 		return -EINVAL;
1785 
1786 	down_read(&fs_info->subvol_sem);
1787 	if (btrfs_root_readonly(root))
1788 		flags |= BTRFS_SUBVOL_RDONLY;
1789 	up_read(&fs_info->subvol_sem);
1790 
1791 	if (copy_to_user(arg, &flags, sizeof(flags)))
1792 		ret = -EFAULT;
1793 
1794 	return ret;
1795 }
1796 
1797 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1798 					      void __user *arg)
1799 {
1800 	struct inode *inode = file_inode(file);
1801 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1802 	struct btrfs_root *root = BTRFS_I(inode)->root;
1803 	struct btrfs_trans_handle *trans;
1804 	u64 root_flags;
1805 	u64 flags;
1806 	int ret = 0;
1807 
1808 	if (!inode_owner_or_capable(inode))
1809 		return -EPERM;
1810 
1811 	ret = mnt_want_write_file(file);
1812 	if (ret)
1813 		goto out;
1814 
1815 	if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1816 		ret = -EINVAL;
1817 		goto out_drop_write;
1818 	}
1819 
1820 	if (copy_from_user(&flags, arg, sizeof(flags))) {
1821 		ret = -EFAULT;
1822 		goto out_drop_write;
1823 	}
1824 
1825 	if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1826 		ret = -EINVAL;
1827 		goto out_drop_write;
1828 	}
1829 
1830 	if (flags & ~BTRFS_SUBVOL_RDONLY) {
1831 		ret = -EOPNOTSUPP;
1832 		goto out_drop_write;
1833 	}
1834 
1835 	down_write(&fs_info->subvol_sem);
1836 
1837 	/* nothing to do */
1838 	if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1839 		goto out_drop_sem;
1840 
1841 	root_flags = btrfs_root_flags(&root->root_item);
1842 	if (flags & BTRFS_SUBVOL_RDONLY) {
1843 		btrfs_set_root_flags(&root->root_item,
1844 				     root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1845 	} else {
1846 		/*
1847 		 * Block RO -> RW transition if this subvolume is involved in
1848 		 * send
1849 		 */
1850 		spin_lock(&root->root_item_lock);
1851 		if (root->send_in_progress == 0) {
1852 			btrfs_set_root_flags(&root->root_item,
1853 				     root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1854 			spin_unlock(&root->root_item_lock);
1855 		} else {
1856 			spin_unlock(&root->root_item_lock);
1857 			btrfs_warn(fs_info,
1858 				   "Attempt to set subvolume %llu read-write during send",
1859 				   root->root_key.objectid);
1860 			ret = -EPERM;
1861 			goto out_drop_sem;
1862 		}
1863 	}
1864 
1865 	trans = btrfs_start_transaction(root, 1);
1866 	if (IS_ERR(trans)) {
1867 		ret = PTR_ERR(trans);
1868 		goto out_reset;
1869 	}
1870 
1871 	ret = btrfs_update_root(trans, fs_info->tree_root,
1872 				&root->root_key, &root->root_item);
1873 
1874 	btrfs_commit_transaction(trans);
1875 out_reset:
1876 	if (ret)
1877 		btrfs_set_root_flags(&root->root_item, root_flags);
1878 out_drop_sem:
1879 	up_write(&fs_info->subvol_sem);
1880 out_drop_write:
1881 	mnt_drop_write_file(file);
1882 out:
1883 	return ret;
1884 }
1885 
1886 /*
1887  * helper to check if the subvolume references other subvolumes
1888  */
1889 static noinline int may_destroy_subvol(struct btrfs_root *root)
1890 {
1891 	struct btrfs_fs_info *fs_info = root->fs_info;
1892 	struct btrfs_path *path;
1893 	struct btrfs_dir_item *di;
1894 	struct btrfs_key key;
1895 	u64 dir_id;
1896 	int ret;
1897 
1898 	path = btrfs_alloc_path();
1899 	if (!path)
1900 		return -ENOMEM;
1901 
1902 	/* Make sure this root isn't set as the default subvol */
1903 	dir_id = btrfs_super_root_dir(fs_info->super_copy);
1904 	di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
1905 				   dir_id, "default", 7, 0);
1906 	if (di && !IS_ERR(di)) {
1907 		btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1908 		if (key.objectid == root->root_key.objectid) {
1909 			ret = -EPERM;
1910 			btrfs_err(fs_info,
1911 				  "deleting default subvolume %llu is not allowed",
1912 				  key.objectid);
1913 			goto out;
1914 		}
1915 		btrfs_release_path(path);
1916 	}
1917 
1918 	key.objectid = root->root_key.objectid;
1919 	key.type = BTRFS_ROOT_REF_KEY;
1920 	key.offset = (u64)-1;
1921 
1922 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1923 	if (ret < 0)
1924 		goto out;
1925 	BUG_ON(ret == 0);
1926 
1927 	ret = 0;
1928 	if (path->slots[0] > 0) {
1929 		path->slots[0]--;
1930 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1931 		if (key.objectid == root->root_key.objectid &&
1932 		    key.type == BTRFS_ROOT_REF_KEY)
1933 			ret = -ENOTEMPTY;
1934 	}
1935 out:
1936 	btrfs_free_path(path);
1937 	return ret;
1938 }
1939 
1940 static noinline int key_in_sk(struct btrfs_key *key,
1941 			      struct btrfs_ioctl_search_key *sk)
1942 {
1943 	struct btrfs_key test;
1944 	int ret;
1945 
1946 	test.objectid = sk->min_objectid;
1947 	test.type = sk->min_type;
1948 	test.offset = sk->min_offset;
1949 
1950 	ret = btrfs_comp_cpu_keys(key, &test);
1951 	if (ret < 0)
1952 		return 0;
1953 
1954 	test.objectid = sk->max_objectid;
1955 	test.type = sk->max_type;
1956 	test.offset = sk->max_offset;
1957 
1958 	ret = btrfs_comp_cpu_keys(key, &test);
1959 	if (ret > 0)
1960 		return 0;
1961 	return 1;
1962 }
1963 
1964 static noinline int copy_to_sk(struct btrfs_path *path,
1965 			       struct btrfs_key *key,
1966 			       struct btrfs_ioctl_search_key *sk,
1967 			       size_t *buf_size,
1968 			       char __user *ubuf,
1969 			       unsigned long *sk_offset,
1970 			       int *num_found)
1971 {
1972 	u64 found_transid;
1973 	struct extent_buffer *leaf;
1974 	struct btrfs_ioctl_search_header sh;
1975 	struct btrfs_key test;
1976 	unsigned long item_off;
1977 	unsigned long item_len;
1978 	int nritems;
1979 	int i;
1980 	int slot;
1981 	int ret = 0;
1982 
1983 	leaf = path->nodes[0];
1984 	slot = path->slots[0];
1985 	nritems = btrfs_header_nritems(leaf);
1986 
1987 	if (btrfs_header_generation(leaf) > sk->max_transid) {
1988 		i = nritems;
1989 		goto advance_key;
1990 	}
1991 	found_transid = btrfs_header_generation(leaf);
1992 
1993 	for (i = slot; i < nritems; i++) {
1994 		item_off = btrfs_item_ptr_offset(leaf, i);
1995 		item_len = btrfs_item_size_nr(leaf, i);
1996 
1997 		btrfs_item_key_to_cpu(leaf, key, i);
1998 		if (!key_in_sk(key, sk))
1999 			continue;
2000 
2001 		if (sizeof(sh) + item_len > *buf_size) {
2002 			if (*num_found) {
2003 				ret = 1;
2004 				goto out;
2005 			}
2006 
2007 			/*
2008 			 * return one empty item back for v1, which does not
2009 			 * handle -EOVERFLOW
2010 			 */
2011 
2012 			*buf_size = sizeof(sh) + item_len;
2013 			item_len = 0;
2014 			ret = -EOVERFLOW;
2015 		}
2016 
2017 		if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2018 			ret = 1;
2019 			goto out;
2020 		}
2021 
2022 		sh.objectid = key->objectid;
2023 		sh.offset = key->offset;
2024 		sh.type = key->type;
2025 		sh.len = item_len;
2026 		sh.transid = found_transid;
2027 
2028 		/* copy search result header */
2029 		if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
2030 			ret = -EFAULT;
2031 			goto out;
2032 		}
2033 
2034 		*sk_offset += sizeof(sh);
2035 
2036 		if (item_len) {
2037 			char __user *up = ubuf + *sk_offset;
2038 			/* copy the item */
2039 			if (read_extent_buffer_to_user(leaf, up,
2040 						       item_off, item_len)) {
2041 				ret = -EFAULT;
2042 				goto out;
2043 			}
2044 
2045 			*sk_offset += item_len;
2046 		}
2047 		(*num_found)++;
2048 
2049 		if (ret) /* -EOVERFLOW from above */
2050 			goto out;
2051 
2052 		if (*num_found >= sk->nr_items) {
2053 			ret = 1;
2054 			goto out;
2055 		}
2056 	}
2057 advance_key:
2058 	ret = 0;
2059 	test.objectid = sk->max_objectid;
2060 	test.type = sk->max_type;
2061 	test.offset = sk->max_offset;
2062 	if (btrfs_comp_cpu_keys(key, &test) >= 0)
2063 		ret = 1;
2064 	else if (key->offset < (u64)-1)
2065 		key->offset++;
2066 	else if (key->type < (u8)-1) {
2067 		key->offset = 0;
2068 		key->type++;
2069 	} else if (key->objectid < (u64)-1) {
2070 		key->offset = 0;
2071 		key->type = 0;
2072 		key->objectid++;
2073 	} else
2074 		ret = 1;
2075 out:
2076 	/*
2077 	 *  0: all items from this leaf copied, continue with next
2078 	 *  1: * more items can be copied, but unused buffer is too small
2079 	 *     * all items were found
2080 	 *     Either way, it will stops the loop which iterates to the next
2081 	 *     leaf
2082 	 *  -EOVERFLOW: item was to large for buffer
2083 	 *  -EFAULT: could not copy extent buffer back to userspace
2084 	 */
2085 	return ret;
2086 }
2087 
2088 static noinline int search_ioctl(struct inode *inode,
2089 				 struct btrfs_ioctl_search_key *sk,
2090 				 size_t *buf_size,
2091 				 char __user *ubuf)
2092 {
2093 	struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2094 	struct btrfs_root *root;
2095 	struct btrfs_key key;
2096 	struct btrfs_path *path;
2097 	int ret;
2098 	int num_found = 0;
2099 	unsigned long sk_offset = 0;
2100 
2101 	if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2102 		*buf_size = sizeof(struct btrfs_ioctl_search_header);
2103 		return -EOVERFLOW;
2104 	}
2105 
2106 	path = btrfs_alloc_path();
2107 	if (!path)
2108 		return -ENOMEM;
2109 
2110 	if (sk->tree_id == 0) {
2111 		/* search the root of the inode that was passed */
2112 		root = BTRFS_I(inode)->root;
2113 	} else {
2114 		key.objectid = sk->tree_id;
2115 		key.type = BTRFS_ROOT_ITEM_KEY;
2116 		key.offset = (u64)-1;
2117 		root = btrfs_read_fs_root_no_name(info, &key);
2118 		if (IS_ERR(root)) {
2119 			btrfs_free_path(path);
2120 			return -ENOENT;
2121 		}
2122 	}
2123 
2124 	key.objectid = sk->min_objectid;
2125 	key.type = sk->min_type;
2126 	key.offset = sk->min_offset;
2127 
2128 	while (1) {
2129 		ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2130 		if (ret != 0) {
2131 			if (ret > 0)
2132 				ret = 0;
2133 			goto err;
2134 		}
2135 		ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2136 				 &sk_offset, &num_found);
2137 		btrfs_release_path(path);
2138 		if (ret)
2139 			break;
2140 
2141 	}
2142 	if (ret > 0)
2143 		ret = 0;
2144 err:
2145 	sk->nr_items = num_found;
2146 	btrfs_free_path(path);
2147 	return ret;
2148 }
2149 
2150 static noinline int btrfs_ioctl_tree_search(struct file *file,
2151 					   void __user *argp)
2152 {
2153 	struct btrfs_ioctl_search_args __user *uargs;
2154 	struct btrfs_ioctl_search_key sk;
2155 	struct inode *inode;
2156 	int ret;
2157 	size_t buf_size;
2158 
2159 	if (!capable(CAP_SYS_ADMIN))
2160 		return -EPERM;
2161 
2162 	uargs = (struct btrfs_ioctl_search_args __user *)argp;
2163 
2164 	if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2165 		return -EFAULT;
2166 
2167 	buf_size = sizeof(uargs->buf);
2168 
2169 	inode = file_inode(file);
2170 	ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2171 
2172 	/*
2173 	 * In the origin implementation an overflow is handled by returning a
2174 	 * search header with a len of zero, so reset ret.
2175 	 */
2176 	if (ret == -EOVERFLOW)
2177 		ret = 0;
2178 
2179 	if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2180 		ret = -EFAULT;
2181 	return ret;
2182 }
2183 
2184 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2185 					       void __user *argp)
2186 {
2187 	struct btrfs_ioctl_search_args_v2 __user *uarg;
2188 	struct btrfs_ioctl_search_args_v2 args;
2189 	struct inode *inode;
2190 	int ret;
2191 	size_t buf_size;
2192 	const size_t buf_limit = SZ_16M;
2193 
2194 	if (!capable(CAP_SYS_ADMIN))
2195 		return -EPERM;
2196 
2197 	/* copy search header and buffer size */
2198 	uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2199 	if (copy_from_user(&args, uarg, sizeof(args)))
2200 		return -EFAULT;
2201 
2202 	buf_size = args.buf_size;
2203 
2204 	if (buf_size < sizeof(struct btrfs_ioctl_search_header))
2205 		return -EOVERFLOW;
2206 
2207 	/* limit result size to 16MB */
2208 	if (buf_size > buf_limit)
2209 		buf_size = buf_limit;
2210 
2211 	inode = file_inode(file);
2212 	ret = search_ioctl(inode, &args.key, &buf_size,
2213 			   (char *)(&uarg->buf[0]));
2214 	if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2215 		ret = -EFAULT;
2216 	else if (ret == -EOVERFLOW &&
2217 		copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2218 		ret = -EFAULT;
2219 
2220 	return ret;
2221 }
2222 
2223 /*
2224  * Search INODE_REFs to identify path name of 'dirid' directory
2225  * in a 'tree_id' tree. and sets path name to 'name'.
2226  */
2227 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2228 				u64 tree_id, u64 dirid, char *name)
2229 {
2230 	struct btrfs_root *root;
2231 	struct btrfs_key key;
2232 	char *ptr;
2233 	int ret = -1;
2234 	int slot;
2235 	int len;
2236 	int total_len = 0;
2237 	struct btrfs_inode_ref *iref;
2238 	struct extent_buffer *l;
2239 	struct btrfs_path *path;
2240 
2241 	if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2242 		name[0]='\0';
2243 		return 0;
2244 	}
2245 
2246 	path = btrfs_alloc_path();
2247 	if (!path)
2248 		return -ENOMEM;
2249 
2250 	ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2251 
2252 	key.objectid = tree_id;
2253 	key.type = BTRFS_ROOT_ITEM_KEY;
2254 	key.offset = (u64)-1;
2255 	root = btrfs_read_fs_root_no_name(info, &key);
2256 	if (IS_ERR(root)) {
2257 		btrfs_err(info, "could not find root %llu", tree_id);
2258 		ret = -ENOENT;
2259 		goto out;
2260 	}
2261 
2262 	key.objectid = dirid;
2263 	key.type = BTRFS_INODE_REF_KEY;
2264 	key.offset = (u64)-1;
2265 
2266 	while (1) {
2267 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2268 		if (ret < 0)
2269 			goto out;
2270 		else if (ret > 0) {
2271 			ret = btrfs_previous_item(root, path, dirid,
2272 						  BTRFS_INODE_REF_KEY);
2273 			if (ret < 0)
2274 				goto out;
2275 			else if (ret > 0) {
2276 				ret = -ENOENT;
2277 				goto out;
2278 			}
2279 		}
2280 
2281 		l = path->nodes[0];
2282 		slot = path->slots[0];
2283 		btrfs_item_key_to_cpu(l, &key, slot);
2284 
2285 		iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2286 		len = btrfs_inode_ref_name_len(l, iref);
2287 		ptr -= len + 1;
2288 		total_len += len + 1;
2289 		if (ptr < name) {
2290 			ret = -ENAMETOOLONG;
2291 			goto out;
2292 		}
2293 
2294 		*(ptr + len) = '/';
2295 		read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2296 
2297 		if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2298 			break;
2299 
2300 		btrfs_release_path(path);
2301 		key.objectid = key.offset;
2302 		key.offset = (u64)-1;
2303 		dirid = key.objectid;
2304 	}
2305 	memmove(name, ptr, total_len);
2306 	name[total_len] = '\0';
2307 	ret = 0;
2308 out:
2309 	btrfs_free_path(path);
2310 	return ret;
2311 }
2312 
2313 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2314 					   void __user *argp)
2315 {
2316 	 struct btrfs_ioctl_ino_lookup_args *args;
2317 	 struct inode *inode;
2318 	int ret = 0;
2319 
2320 	args = memdup_user(argp, sizeof(*args));
2321 	if (IS_ERR(args))
2322 		return PTR_ERR(args);
2323 
2324 	inode = file_inode(file);
2325 
2326 	/*
2327 	 * Unprivileged query to obtain the containing subvolume root id. The
2328 	 * path is reset so it's consistent with btrfs_search_path_in_tree.
2329 	 */
2330 	if (args->treeid == 0)
2331 		args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2332 
2333 	if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2334 		args->name[0] = 0;
2335 		goto out;
2336 	}
2337 
2338 	if (!capable(CAP_SYS_ADMIN)) {
2339 		ret = -EPERM;
2340 		goto out;
2341 	}
2342 
2343 	ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2344 					args->treeid, args->objectid,
2345 					args->name);
2346 
2347 out:
2348 	if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2349 		ret = -EFAULT;
2350 
2351 	kfree(args);
2352 	return ret;
2353 }
2354 
2355 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2356 					     void __user *arg)
2357 {
2358 	struct dentry *parent = file->f_path.dentry;
2359 	struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2360 	struct dentry *dentry;
2361 	struct inode *dir = d_inode(parent);
2362 	struct inode *inode;
2363 	struct btrfs_root *root = BTRFS_I(dir)->root;
2364 	struct btrfs_root *dest = NULL;
2365 	struct btrfs_ioctl_vol_args *vol_args;
2366 	struct btrfs_trans_handle *trans;
2367 	struct btrfs_block_rsv block_rsv;
2368 	u64 root_flags;
2369 	u64 qgroup_reserved;
2370 	int namelen;
2371 	int ret;
2372 	int err = 0;
2373 
2374 	if (!S_ISDIR(dir->i_mode))
2375 		return -ENOTDIR;
2376 
2377 	vol_args = memdup_user(arg, sizeof(*vol_args));
2378 	if (IS_ERR(vol_args))
2379 		return PTR_ERR(vol_args);
2380 
2381 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2382 	namelen = strlen(vol_args->name);
2383 	if (strchr(vol_args->name, '/') ||
2384 	    strncmp(vol_args->name, "..", namelen) == 0) {
2385 		err = -EINVAL;
2386 		goto out;
2387 	}
2388 
2389 	err = mnt_want_write_file(file);
2390 	if (err)
2391 		goto out;
2392 
2393 
2394 	err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2395 	if (err == -EINTR)
2396 		goto out_drop_write;
2397 	dentry = lookup_one_len(vol_args->name, parent, namelen);
2398 	if (IS_ERR(dentry)) {
2399 		err = PTR_ERR(dentry);
2400 		goto out_unlock_dir;
2401 	}
2402 
2403 	if (d_really_is_negative(dentry)) {
2404 		err = -ENOENT;
2405 		goto out_dput;
2406 	}
2407 
2408 	inode = d_inode(dentry);
2409 	dest = BTRFS_I(inode)->root;
2410 	if (!capable(CAP_SYS_ADMIN)) {
2411 		/*
2412 		 * Regular user.  Only allow this with a special mount
2413 		 * option, when the user has write+exec access to the
2414 		 * subvol root, and when rmdir(2) would have been
2415 		 * allowed.
2416 		 *
2417 		 * Note that this is _not_ check that the subvol is
2418 		 * empty or doesn't contain data that we wouldn't
2419 		 * otherwise be able to delete.
2420 		 *
2421 		 * Users who want to delete empty subvols should try
2422 		 * rmdir(2).
2423 		 */
2424 		err = -EPERM;
2425 		if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2426 			goto out_dput;
2427 
2428 		/*
2429 		 * Do not allow deletion if the parent dir is the same
2430 		 * as the dir to be deleted.  That means the ioctl
2431 		 * must be called on the dentry referencing the root
2432 		 * of the subvol, not a random directory contained
2433 		 * within it.
2434 		 */
2435 		err = -EINVAL;
2436 		if (root == dest)
2437 			goto out_dput;
2438 
2439 		err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2440 		if (err)
2441 			goto out_dput;
2442 	}
2443 
2444 	/* check if subvolume may be deleted by a user */
2445 	err = btrfs_may_delete(dir, dentry, 1);
2446 	if (err)
2447 		goto out_dput;
2448 
2449 	if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2450 		err = -EINVAL;
2451 		goto out_dput;
2452 	}
2453 
2454 	inode_lock(inode);
2455 
2456 	/*
2457 	 * Don't allow to delete a subvolume with send in progress. This is
2458 	 * inside the i_mutex so the error handling that has to drop the bit
2459 	 * again is not run concurrently.
2460 	 */
2461 	spin_lock(&dest->root_item_lock);
2462 	root_flags = btrfs_root_flags(&dest->root_item);
2463 	if (dest->send_in_progress == 0) {
2464 		btrfs_set_root_flags(&dest->root_item,
2465 				root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2466 		spin_unlock(&dest->root_item_lock);
2467 	} else {
2468 		spin_unlock(&dest->root_item_lock);
2469 		btrfs_warn(fs_info,
2470 			   "Attempt to delete subvolume %llu during send",
2471 			   dest->root_key.objectid);
2472 		err = -EPERM;
2473 		goto out_unlock_inode;
2474 	}
2475 
2476 	down_write(&fs_info->subvol_sem);
2477 
2478 	err = may_destroy_subvol(dest);
2479 	if (err)
2480 		goto out_up_write;
2481 
2482 	btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2483 	/*
2484 	 * One for dir inode, two for dir entries, two for root
2485 	 * ref/backref.
2486 	 */
2487 	err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2488 					       5, &qgroup_reserved, true);
2489 	if (err)
2490 		goto out_up_write;
2491 
2492 	trans = btrfs_start_transaction(root, 0);
2493 	if (IS_ERR(trans)) {
2494 		err = PTR_ERR(trans);
2495 		goto out_release;
2496 	}
2497 	trans->block_rsv = &block_rsv;
2498 	trans->bytes_reserved = block_rsv.size;
2499 
2500 	btrfs_record_snapshot_destroy(trans, dir);
2501 
2502 	ret = btrfs_unlink_subvol(trans, root, dir,
2503 				dest->root_key.objectid,
2504 				dentry->d_name.name,
2505 				dentry->d_name.len);
2506 	if (ret) {
2507 		err = ret;
2508 		btrfs_abort_transaction(trans, ret);
2509 		goto out_end_trans;
2510 	}
2511 
2512 	btrfs_record_root_in_trans(trans, dest);
2513 
2514 	memset(&dest->root_item.drop_progress, 0,
2515 		sizeof(dest->root_item.drop_progress));
2516 	dest->root_item.drop_level = 0;
2517 	btrfs_set_root_refs(&dest->root_item, 0);
2518 
2519 	if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2520 		ret = btrfs_insert_orphan_item(trans,
2521 					fs_info->tree_root,
2522 					dest->root_key.objectid);
2523 		if (ret) {
2524 			btrfs_abort_transaction(trans, ret);
2525 			err = ret;
2526 			goto out_end_trans;
2527 		}
2528 	}
2529 
2530 	ret = btrfs_uuid_tree_rem(trans, fs_info, dest->root_item.uuid,
2531 				  BTRFS_UUID_KEY_SUBVOL,
2532 				  dest->root_key.objectid);
2533 	if (ret && ret != -ENOENT) {
2534 		btrfs_abort_transaction(trans, ret);
2535 		err = ret;
2536 		goto out_end_trans;
2537 	}
2538 	if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2539 		ret = btrfs_uuid_tree_rem(trans, fs_info,
2540 					  dest->root_item.received_uuid,
2541 					  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2542 					  dest->root_key.objectid);
2543 		if (ret && ret != -ENOENT) {
2544 			btrfs_abort_transaction(trans, ret);
2545 			err = ret;
2546 			goto out_end_trans;
2547 		}
2548 	}
2549 
2550 out_end_trans:
2551 	trans->block_rsv = NULL;
2552 	trans->bytes_reserved = 0;
2553 	ret = btrfs_end_transaction(trans);
2554 	if (ret && !err)
2555 		err = ret;
2556 	inode->i_flags |= S_DEAD;
2557 out_release:
2558 	btrfs_subvolume_release_metadata(fs_info, &block_rsv, qgroup_reserved);
2559 out_up_write:
2560 	up_write(&fs_info->subvol_sem);
2561 	if (err) {
2562 		spin_lock(&dest->root_item_lock);
2563 		root_flags = btrfs_root_flags(&dest->root_item);
2564 		btrfs_set_root_flags(&dest->root_item,
2565 				root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2566 		spin_unlock(&dest->root_item_lock);
2567 	}
2568 out_unlock_inode:
2569 	inode_unlock(inode);
2570 	if (!err) {
2571 		d_invalidate(dentry);
2572 		btrfs_invalidate_inodes(dest);
2573 		d_delete(dentry);
2574 		ASSERT(dest->send_in_progress == 0);
2575 
2576 		/* the last ref */
2577 		if (dest->ino_cache_inode) {
2578 			iput(dest->ino_cache_inode);
2579 			dest->ino_cache_inode = NULL;
2580 		}
2581 	}
2582 out_dput:
2583 	dput(dentry);
2584 out_unlock_dir:
2585 	inode_unlock(dir);
2586 out_drop_write:
2587 	mnt_drop_write_file(file);
2588 out:
2589 	kfree(vol_args);
2590 	return err;
2591 }
2592 
2593 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2594 {
2595 	struct inode *inode = file_inode(file);
2596 	struct btrfs_root *root = BTRFS_I(inode)->root;
2597 	struct btrfs_ioctl_defrag_range_args *range;
2598 	int ret;
2599 
2600 	ret = mnt_want_write_file(file);
2601 	if (ret)
2602 		return ret;
2603 
2604 	if (btrfs_root_readonly(root)) {
2605 		ret = -EROFS;
2606 		goto out;
2607 	}
2608 
2609 	switch (inode->i_mode & S_IFMT) {
2610 	case S_IFDIR:
2611 		if (!capable(CAP_SYS_ADMIN)) {
2612 			ret = -EPERM;
2613 			goto out;
2614 		}
2615 		ret = btrfs_defrag_root(root);
2616 		if (ret)
2617 			goto out;
2618 		ret = btrfs_defrag_root(root->fs_info->extent_root);
2619 		break;
2620 	case S_IFREG:
2621 		if (!(file->f_mode & FMODE_WRITE)) {
2622 			ret = -EINVAL;
2623 			goto out;
2624 		}
2625 
2626 		range = kzalloc(sizeof(*range), GFP_KERNEL);
2627 		if (!range) {
2628 			ret = -ENOMEM;
2629 			goto out;
2630 		}
2631 
2632 		if (argp) {
2633 			if (copy_from_user(range, argp,
2634 					   sizeof(*range))) {
2635 				ret = -EFAULT;
2636 				kfree(range);
2637 				goto out;
2638 			}
2639 			/* compression requires us to start the IO */
2640 			if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2641 				range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2642 				range->extent_thresh = (u32)-1;
2643 			}
2644 		} else {
2645 			/* the rest are all set to zero by kzalloc */
2646 			range->len = (u64)-1;
2647 		}
2648 		ret = btrfs_defrag_file(file_inode(file), file,
2649 					range, 0, 0);
2650 		if (ret > 0)
2651 			ret = 0;
2652 		kfree(range);
2653 		break;
2654 	default:
2655 		ret = -EINVAL;
2656 	}
2657 out:
2658 	mnt_drop_write_file(file);
2659 	return ret;
2660 }
2661 
2662 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
2663 {
2664 	struct btrfs_ioctl_vol_args *vol_args;
2665 	int ret;
2666 
2667 	if (!capable(CAP_SYS_ADMIN))
2668 		return -EPERM;
2669 
2670 	if (atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1))
2671 		return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2672 
2673 	mutex_lock(&fs_info->volume_mutex);
2674 	vol_args = memdup_user(arg, sizeof(*vol_args));
2675 	if (IS_ERR(vol_args)) {
2676 		ret = PTR_ERR(vol_args);
2677 		goto out;
2678 	}
2679 
2680 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2681 	ret = btrfs_init_new_device(fs_info, vol_args->name);
2682 
2683 	if (!ret)
2684 		btrfs_info(fs_info, "disk added %s", vol_args->name);
2685 
2686 	kfree(vol_args);
2687 out:
2688 	mutex_unlock(&fs_info->volume_mutex);
2689 	atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
2690 	return ret;
2691 }
2692 
2693 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
2694 {
2695 	struct inode *inode = file_inode(file);
2696 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2697 	struct btrfs_ioctl_vol_args_v2 *vol_args;
2698 	int ret;
2699 
2700 	if (!capable(CAP_SYS_ADMIN))
2701 		return -EPERM;
2702 
2703 	ret = mnt_want_write_file(file);
2704 	if (ret)
2705 		return ret;
2706 
2707 	vol_args = memdup_user(arg, sizeof(*vol_args));
2708 	if (IS_ERR(vol_args)) {
2709 		ret = PTR_ERR(vol_args);
2710 		goto err_drop;
2711 	}
2712 
2713 	/* Check for compatibility reject unknown flags */
2714 	if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED)
2715 		return -EOPNOTSUPP;
2716 
2717 	if (atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
2718 		ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2719 		goto out;
2720 	}
2721 
2722 	mutex_lock(&fs_info->volume_mutex);
2723 	if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
2724 		ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
2725 	} else {
2726 		vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2727 		ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2728 	}
2729 	mutex_unlock(&fs_info->volume_mutex);
2730 	atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
2731 
2732 	if (!ret) {
2733 		if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
2734 			btrfs_info(fs_info, "device deleted: id %llu",
2735 					vol_args->devid);
2736 		else
2737 			btrfs_info(fs_info, "device deleted: %s",
2738 					vol_args->name);
2739 	}
2740 out:
2741 	kfree(vol_args);
2742 err_drop:
2743 	mnt_drop_write_file(file);
2744 	return ret;
2745 }
2746 
2747 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2748 {
2749 	struct inode *inode = file_inode(file);
2750 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2751 	struct btrfs_ioctl_vol_args *vol_args;
2752 	int ret;
2753 
2754 	if (!capable(CAP_SYS_ADMIN))
2755 		return -EPERM;
2756 
2757 	ret = mnt_want_write_file(file);
2758 	if (ret)
2759 		return ret;
2760 
2761 	if (atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
2762 		ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2763 		goto out_drop_write;
2764 	}
2765 
2766 	vol_args = memdup_user(arg, sizeof(*vol_args));
2767 	if (IS_ERR(vol_args)) {
2768 		ret = PTR_ERR(vol_args);
2769 		goto out;
2770 	}
2771 
2772 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2773 	mutex_lock(&fs_info->volume_mutex);
2774 	ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2775 	mutex_unlock(&fs_info->volume_mutex);
2776 
2777 	if (!ret)
2778 		btrfs_info(fs_info, "disk deleted %s", vol_args->name);
2779 	kfree(vol_args);
2780 out:
2781 	atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
2782 out_drop_write:
2783 	mnt_drop_write_file(file);
2784 
2785 	return ret;
2786 }
2787 
2788 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
2789 				void __user *arg)
2790 {
2791 	struct btrfs_ioctl_fs_info_args *fi_args;
2792 	struct btrfs_device *device;
2793 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2794 	int ret = 0;
2795 
2796 	fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2797 	if (!fi_args)
2798 		return -ENOMEM;
2799 
2800 	mutex_lock(&fs_devices->device_list_mutex);
2801 	fi_args->num_devices = fs_devices->num_devices;
2802 	memcpy(&fi_args->fsid, fs_info->fsid, sizeof(fi_args->fsid));
2803 
2804 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
2805 		if (device->devid > fi_args->max_id)
2806 			fi_args->max_id = device->devid;
2807 	}
2808 	mutex_unlock(&fs_devices->device_list_mutex);
2809 
2810 	fi_args->nodesize = fs_info->super_copy->nodesize;
2811 	fi_args->sectorsize = fs_info->super_copy->sectorsize;
2812 	fi_args->clone_alignment = fs_info->super_copy->sectorsize;
2813 
2814 	if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2815 		ret = -EFAULT;
2816 
2817 	kfree(fi_args);
2818 	return ret;
2819 }
2820 
2821 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
2822 				 void __user *arg)
2823 {
2824 	struct btrfs_ioctl_dev_info_args *di_args;
2825 	struct btrfs_device *dev;
2826 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2827 	int ret = 0;
2828 	char *s_uuid = NULL;
2829 
2830 	di_args = memdup_user(arg, sizeof(*di_args));
2831 	if (IS_ERR(di_args))
2832 		return PTR_ERR(di_args);
2833 
2834 	if (!btrfs_is_empty_uuid(di_args->uuid))
2835 		s_uuid = di_args->uuid;
2836 
2837 	mutex_lock(&fs_devices->device_list_mutex);
2838 	dev = btrfs_find_device(fs_info, di_args->devid, s_uuid, NULL);
2839 
2840 	if (!dev) {
2841 		ret = -ENODEV;
2842 		goto out;
2843 	}
2844 
2845 	di_args->devid = dev->devid;
2846 	di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2847 	di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2848 	memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2849 	if (dev->name) {
2850 		struct rcu_string *name;
2851 
2852 		rcu_read_lock();
2853 		name = rcu_dereference(dev->name);
2854 		strncpy(di_args->path, name->str, sizeof(di_args->path));
2855 		rcu_read_unlock();
2856 		di_args->path[sizeof(di_args->path) - 1] = 0;
2857 	} else {
2858 		di_args->path[0] = '\0';
2859 	}
2860 
2861 out:
2862 	mutex_unlock(&fs_devices->device_list_mutex);
2863 	if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2864 		ret = -EFAULT;
2865 
2866 	kfree(di_args);
2867 	return ret;
2868 }
2869 
2870 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
2871 {
2872 	struct page *page;
2873 
2874 	page = grab_cache_page(inode->i_mapping, index);
2875 	if (!page)
2876 		return ERR_PTR(-ENOMEM);
2877 
2878 	if (!PageUptodate(page)) {
2879 		int ret;
2880 
2881 		ret = btrfs_readpage(NULL, page);
2882 		if (ret)
2883 			return ERR_PTR(ret);
2884 		lock_page(page);
2885 		if (!PageUptodate(page)) {
2886 			unlock_page(page);
2887 			put_page(page);
2888 			return ERR_PTR(-EIO);
2889 		}
2890 		if (page->mapping != inode->i_mapping) {
2891 			unlock_page(page);
2892 			put_page(page);
2893 			return ERR_PTR(-EAGAIN);
2894 		}
2895 	}
2896 
2897 	return page;
2898 }
2899 
2900 static int gather_extent_pages(struct inode *inode, struct page **pages,
2901 			       int num_pages, u64 off)
2902 {
2903 	int i;
2904 	pgoff_t index = off >> PAGE_SHIFT;
2905 
2906 	for (i = 0; i < num_pages; i++) {
2907 again:
2908 		pages[i] = extent_same_get_page(inode, index + i);
2909 		if (IS_ERR(pages[i])) {
2910 			int err = PTR_ERR(pages[i]);
2911 
2912 			if (err == -EAGAIN)
2913 				goto again;
2914 			pages[i] = NULL;
2915 			return err;
2916 		}
2917 	}
2918 	return 0;
2919 }
2920 
2921 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
2922 			     bool retry_range_locking)
2923 {
2924 	/*
2925 	 * Do any pending delalloc/csum calculations on inode, one way or
2926 	 * another, and lock file content.
2927 	 * The locking order is:
2928 	 *
2929 	 *   1) pages
2930 	 *   2) range in the inode's io tree
2931 	 */
2932 	while (1) {
2933 		struct btrfs_ordered_extent *ordered;
2934 		lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2935 		ordered = btrfs_lookup_first_ordered_extent(inode,
2936 							    off + len - 1);
2937 		if ((!ordered ||
2938 		     ordered->file_offset + ordered->len <= off ||
2939 		     ordered->file_offset >= off + len) &&
2940 		    !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2941 				    off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2942 			if (ordered)
2943 				btrfs_put_ordered_extent(ordered);
2944 			break;
2945 		}
2946 		unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2947 		if (ordered)
2948 			btrfs_put_ordered_extent(ordered);
2949 		if (!retry_range_locking)
2950 			return -EAGAIN;
2951 		btrfs_wait_ordered_range(inode, off, len);
2952 	}
2953 	return 0;
2954 }
2955 
2956 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
2957 {
2958 	inode_unlock(inode1);
2959 	inode_unlock(inode2);
2960 }
2961 
2962 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
2963 {
2964 	if (inode1 < inode2)
2965 		swap(inode1, inode2);
2966 
2967 	inode_lock_nested(inode1, I_MUTEX_PARENT);
2968 	inode_lock_nested(inode2, I_MUTEX_CHILD);
2969 }
2970 
2971 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
2972 				      struct inode *inode2, u64 loff2, u64 len)
2973 {
2974 	unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2975 	unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2976 }
2977 
2978 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
2979 				    struct inode *inode2, u64 loff2, u64 len,
2980 				    bool retry_range_locking)
2981 {
2982 	int ret;
2983 
2984 	if (inode1 < inode2) {
2985 		swap(inode1, inode2);
2986 		swap(loff1, loff2);
2987 	}
2988 	ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
2989 	if (ret)
2990 		return ret;
2991 	ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
2992 	if (ret)
2993 		unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
2994 			      loff1 + len - 1);
2995 	return ret;
2996 }
2997 
2998 struct cmp_pages {
2999 	int		num_pages;
3000 	struct page	**src_pages;
3001 	struct page	**dst_pages;
3002 };
3003 
3004 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
3005 {
3006 	int i;
3007 	struct page *pg;
3008 
3009 	for (i = 0; i < cmp->num_pages; i++) {
3010 		pg = cmp->src_pages[i];
3011 		if (pg) {
3012 			unlock_page(pg);
3013 			put_page(pg);
3014 		}
3015 		pg = cmp->dst_pages[i];
3016 		if (pg) {
3017 			unlock_page(pg);
3018 			put_page(pg);
3019 		}
3020 	}
3021 	kfree(cmp->src_pages);
3022 	kfree(cmp->dst_pages);
3023 }
3024 
3025 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
3026 				  struct inode *dst, u64 dst_loff,
3027 				  u64 len, struct cmp_pages *cmp)
3028 {
3029 	int ret;
3030 	int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT;
3031 	struct page **src_pgarr, **dst_pgarr;
3032 
3033 	/*
3034 	 * We must gather up all the pages before we initiate our
3035 	 * extent locking. We use an array for the page pointers. Size
3036 	 * of the array is bounded by len, which is in turn bounded by
3037 	 * BTRFS_MAX_DEDUPE_LEN.
3038 	 */
3039 	src_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
3040 	dst_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
3041 	if (!src_pgarr || !dst_pgarr) {
3042 		kfree(src_pgarr);
3043 		kfree(dst_pgarr);
3044 		return -ENOMEM;
3045 	}
3046 	cmp->num_pages = num_pages;
3047 	cmp->src_pages = src_pgarr;
3048 	cmp->dst_pages = dst_pgarr;
3049 
3050 	ret = gather_extent_pages(src, cmp->src_pages, cmp->num_pages, loff);
3051 	if (ret)
3052 		goto out;
3053 
3054 	ret = gather_extent_pages(dst, cmp->dst_pages, cmp->num_pages, dst_loff);
3055 
3056 out:
3057 	if (ret)
3058 		btrfs_cmp_data_free(cmp);
3059 	return 0;
3060 }
3061 
3062 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
3063 			  u64 dst_loff, u64 len, struct cmp_pages *cmp)
3064 {
3065 	int ret = 0;
3066 	int i;
3067 	struct page *src_page, *dst_page;
3068 	unsigned int cmp_len = PAGE_SIZE;
3069 	void *addr, *dst_addr;
3070 
3071 	i = 0;
3072 	while (len) {
3073 		if (len < PAGE_SIZE)
3074 			cmp_len = len;
3075 
3076 		BUG_ON(i >= cmp->num_pages);
3077 
3078 		src_page = cmp->src_pages[i];
3079 		dst_page = cmp->dst_pages[i];
3080 		ASSERT(PageLocked(src_page));
3081 		ASSERT(PageLocked(dst_page));
3082 
3083 		addr = kmap_atomic(src_page);
3084 		dst_addr = kmap_atomic(dst_page);
3085 
3086 		flush_dcache_page(src_page);
3087 		flush_dcache_page(dst_page);
3088 
3089 		if (memcmp(addr, dst_addr, cmp_len))
3090 			ret = -EBADE;
3091 
3092 		kunmap_atomic(addr);
3093 		kunmap_atomic(dst_addr);
3094 
3095 		if (ret)
3096 			break;
3097 
3098 		len -= cmp_len;
3099 		i++;
3100 	}
3101 
3102 	return ret;
3103 }
3104 
3105 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3106 				     u64 olen)
3107 {
3108 	u64 len = *plen;
3109 	u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3110 
3111 	if (off + olen > inode->i_size || off + olen < off)
3112 		return -EINVAL;
3113 
3114 	/* if we extend to eof, continue to block boundary */
3115 	if (off + len == inode->i_size)
3116 		*plen = len = ALIGN(inode->i_size, bs) - off;
3117 
3118 	/* Check that we are block aligned - btrfs_clone() requires this */
3119 	if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3120 		return -EINVAL;
3121 
3122 	return 0;
3123 }
3124 
3125 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3126 			     struct inode *dst, u64 dst_loff)
3127 {
3128 	int ret;
3129 	u64 len = olen;
3130 	struct cmp_pages cmp;
3131 	int same_inode = 0;
3132 	u64 same_lock_start = 0;
3133 	u64 same_lock_len = 0;
3134 
3135 	if (src == dst)
3136 		same_inode = 1;
3137 
3138 	if (len == 0)
3139 		return 0;
3140 
3141 	if (same_inode) {
3142 		inode_lock(src);
3143 
3144 		ret = extent_same_check_offsets(src, loff, &len, olen);
3145 		if (ret)
3146 			goto out_unlock;
3147 		ret = extent_same_check_offsets(src, dst_loff, &len, olen);
3148 		if (ret)
3149 			goto out_unlock;
3150 
3151 		/*
3152 		 * Single inode case wants the same checks, except we
3153 		 * don't want our length pushed out past i_size as
3154 		 * comparing that data range makes no sense.
3155 		 *
3156 		 * extent_same_check_offsets() will do this for an
3157 		 * unaligned length at i_size, so catch it here and
3158 		 * reject the request.
3159 		 *
3160 		 * This effectively means we require aligned extents
3161 		 * for the single-inode case, whereas the other cases
3162 		 * allow an unaligned length so long as it ends at
3163 		 * i_size.
3164 		 */
3165 		if (len != olen) {
3166 			ret = -EINVAL;
3167 			goto out_unlock;
3168 		}
3169 
3170 		/* Check for overlapping ranges */
3171 		if (dst_loff + len > loff && dst_loff < loff + len) {
3172 			ret = -EINVAL;
3173 			goto out_unlock;
3174 		}
3175 
3176 		same_lock_start = min_t(u64, loff, dst_loff);
3177 		same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3178 	} else {
3179 		btrfs_double_inode_lock(src, dst);
3180 
3181 		ret = extent_same_check_offsets(src, loff, &len, olen);
3182 		if (ret)
3183 			goto out_unlock;
3184 
3185 		ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3186 		if (ret)
3187 			goto out_unlock;
3188 	}
3189 
3190 	/* don't make the dst file partly checksummed */
3191 	if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3192 	    (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3193 		ret = -EINVAL;
3194 		goto out_unlock;
3195 	}
3196 
3197 again:
3198 	ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp);
3199 	if (ret)
3200 		goto out_unlock;
3201 
3202 	if (same_inode)
3203 		ret = lock_extent_range(src, same_lock_start, same_lock_len,
3204 					false);
3205 	else
3206 		ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3207 					       false);
3208 	/*
3209 	 * If one of the inodes has dirty pages in the respective range or
3210 	 * ordered extents, we need to flush dellaloc and wait for all ordered
3211 	 * extents in the range. We must unlock the pages and the ranges in the
3212 	 * io trees to avoid deadlocks when flushing delalloc (requires locking
3213 	 * pages) and when waiting for ordered extents to complete (they require
3214 	 * range locking).
3215 	 */
3216 	if (ret == -EAGAIN) {
3217 		/*
3218 		 * Ranges in the io trees already unlocked. Now unlock all
3219 		 * pages before waiting for all IO to complete.
3220 		 */
3221 		btrfs_cmp_data_free(&cmp);
3222 		if (same_inode) {
3223 			btrfs_wait_ordered_range(src, same_lock_start,
3224 						 same_lock_len);
3225 		} else {
3226 			btrfs_wait_ordered_range(src, loff, len);
3227 			btrfs_wait_ordered_range(dst, dst_loff, len);
3228 		}
3229 		goto again;
3230 	}
3231 	ASSERT(ret == 0);
3232 	if (WARN_ON(ret)) {
3233 		/* ranges in the io trees already unlocked */
3234 		btrfs_cmp_data_free(&cmp);
3235 		return ret;
3236 	}
3237 
3238 	/* pass original length for comparison so we stay within i_size */
3239 	ret = btrfs_cmp_data(src, loff, dst, dst_loff, olen, &cmp);
3240 	if (ret == 0)
3241 		ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3242 
3243 	if (same_inode)
3244 		unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3245 			      same_lock_start + same_lock_len - 1);
3246 	else
3247 		btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3248 
3249 	btrfs_cmp_data_free(&cmp);
3250 out_unlock:
3251 	if (same_inode)
3252 		inode_unlock(src);
3253 	else
3254 		btrfs_double_inode_unlock(src, dst);
3255 
3256 	return ret;
3257 }
3258 
3259 #define BTRFS_MAX_DEDUPE_LEN	SZ_16M
3260 
3261 ssize_t btrfs_dedupe_file_range(struct file *src_file, u64 loff, u64 olen,
3262 				struct file *dst_file, u64 dst_loff)
3263 {
3264 	struct inode *src = file_inode(src_file);
3265 	struct inode *dst = file_inode(dst_file);
3266 	u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3267 	ssize_t res;
3268 
3269 	if (olen > BTRFS_MAX_DEDUPE_LEN)
3270 		olen = BTRFS_MAX_DEDUPE_LEN;
3271 
3272 	if (WARN_ON_ONCE(bs < PAGE_SIZE)) {
3273 		/*
3274 		 * Btrfs does not support blocksize < page_size. As a
3275 		 * result, btrfs_cmp_data() won't correctly handle
3276 		 * this situation without an update.
3277 		 */
3278 		return -EINVAL;
3279 	}
3280 
3281 	res = btrfs_extent_same(src, loff, olen, dst, dst_loff);
3282 	if (res)
3283 		return res;
3284 	return olen;
3285 }
3286 
3287 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3288 				     struct inode *inode,
3289 				     u64 endoff,
3290 				     const u64 destoff,
3291 				     const u64 olen,
3292 				     int no_time_update)
3293 {
3294 	struct btrfs_root *root = BTRFS_I(inode)->root;
3295 	int ret;
3296 
3297 	inode_inc_iversion(inode);
3298 	if (!no_time_update)
3299 		inode->i_mtime = inode->i_ctime = current_time(inode);
3300 	/*
3301 	 * We round up to the block size at eof when determining which
3302 	 * extents to clone above, but shouldn't round up the file size.
3303 	 */
3304 	if (endoff > destoff + olen)
3305 		endoff = destoff + olen;
3306 	if (endoff > inode->i_size)
3307 		btrfs_i_size_write(inode, endoff);
3308 
3309 	ret = btrfs_update_inode(trans, root, inode);
3310 	if (ret) {
3311 		btrfs_abort_transaction(trans, ret);
3312 		btrfs_end_transaction(trans);
3313 		goto out;
3314 	}
3315 	ret = btrfs_end_transaction(trans);
3316 out:
3317 	return ret;
3318 }
3319 
3320 static void clone_update_extent_map(struct inode *inode,
3321 				    const struct btrfs_trans_handle *trans,
3322 				    const struct btrfs_path *path,
3323 				    const u64 hole_offset,
3324 				    const u64 hole_len)
3325 {
3326 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3327 	struct extent_map *em;
3328 	int ret;
3329 
3330 	em = alloc_extent_map();
3331 	if (!em) {
3332 		set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3333 			&BTRFS_I(inode)->runtime_flags);
3334 		return;
3335 	}
3336 
3337 	if (path) {
3338 		struct btrfs_file_extent_item *fi;
3339 
3340 		fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3341 				    struct btrfs_file_extent_item);
3342 		btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3343 		em->generation = -1;
3344 		if (btrfs_file_extent_type(path->nodes[0], fi) ==
3345 		    BTRFS_FILE_EXTENT_INLINE)
3346 			set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3347 				&BTRFS_I(inode)->runtime_flags);
3348 	} else {
3349 		em->start = hole_offset;
3350 		em->len = hole_len;
3351 		em->ram_bytes = em->len;
3352 		em->orig_start = hole_offset;
3353 		em->block_start = EXTENT_MAP_HOLE;
3354 		em->block_len = 0;
3355 		em->orig_block_len = 0;
3356 		em->compress_type = BTRFS_COMPRESS_NONE;
3357 		em->generation = trans->transid;
3358 	}
3359 
3360 	while (1) {
3361 		write_lock(&em_tree->lock);
3362 		ret = add_extent_mapping(em_tree, em, 1);
3363 		write_unlock(&em_tree->lock);
3364 		if (ret != -EEXIST) {
3365 			free_extent_map(em);
3366 			break;
3367 		}
3368 		btrfs_drop_extent_cache(inode, em->start,
3369 					em->start + em->len - 1, 0);
3370 	}
3371 
3372 	if (ret)
3373 		set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3374 			&BTRFS_I(inode)->runtime_flags);
3375 }
3376 
3377 /*
3378  * Make sure we do not end up inserting an inline extent into a file that has
3379  * already other (non-inline) extents. If a file has an inline extent it can
3380  * not have any other extents and the (single) inline extent must start at the
3381  * file offset 0. Failing to respect these rules will lead to file corruption,
3382  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3383  *
3384  * We can have extents that have been already written to disk or we can have
3385  * dirty ranges still in delalloc, in which case the extent maps and items are
3386  * created only when we run delalloc, and the delalloc ranges might fall outside
3387  * the range we are currently locking in the inode's io tree. So we check the
3388  * inode's i_size because of that (i_size updates are done while holding the
3389  * i_mutex, which we are holding here).
3390  * We also check to see if the inode has a size not greater than "datal" but has
3391  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3392  * protected against such concurrent fallocate calls by the i_mutex).
3393  *
3394  * If the file has no extents but a size greater than datal, do not allow the
3395  * copy because we would need turn the inline extent into a non-inline one (even
3396  * with NO_HOLES enabled). If we find our destination inode only has one inline
3397  * extent, just overwrite it with the source inline extent if its size is less
3398  * than the source extent's size, or we could copy the source inline extent's
3399  * data into the destination inode's inline extent if the later is greater then
3400  * the former.
3401  */
3402 static int clone_copy_inline_extent(struct inode *src,
3403 				    struct inode *dst,
3404 				    struct btrfs_trans_handle *trans,
3405 				    struct btrfs_path *path,
3406 				    struct btrfs_key *new_key,
3407 				    const u64 drop_start,
3408 				    const u64 datal,
3409 				    const u64 skip,
3410 				    const u64 size,
3411 				    char *inline_data)
3412 {
3413 	struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb);
3414 	struct btrfs_root *root = BTRFS_I(dst)->root;
3415 	const u64 aligned_end = ALIGN(new_key->offset + datal,
3416 				      fs_info->sectorsize);
3417 	int ret;
3418 	struct btrfs_key key;
3419 
3420 	if (new_key->offset > 0)
3421 		return -EOPNOTSUPP;
3422 
3423 	key.objectid = btrfs_ino(dst);
3424 	key.type = BTRFS_EXTENT_DATA_KEY;
3425 	key.offset = 0;
3426 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3427 	if (ret < 0) {
3428 		return ret;
3429 	} else if (ret > 0) {
3430 		if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3431 			ret = btrfs_next_leaf(root, path);
3432 			if (ret < 0)
3433 				return ret;
3434 			else if (ret > 0)
3435 				goto copy_inline_extent;
3436 		}
3437 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3438 		if (key.objectid == btrfs_ino(dst) &&
3439 		    key.type == BTRFS_EXTENT_DATA_KEY) {
3440 			ASSERT(key.offset > 0);
3441 			return -EOPNOTSUPP;
3442 		}
3443 	} else if (i_size_read(dst) <= datal) {
3444 		struct btrfs_file_extent_item *ei;
3445 		u64 ext_len;
3446 
3447 		/*
3448 		 * If the file size is <= datal, make sure there are no other
3449 		 * extents following (can happen do to an fallocate call with
3450 		 * the flag FALLOC_FL_KEEP_SIZE).
3451 		 */
3452 		ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3453 				    struct btrfs_file_extent_item);
3454 		/*
3455 		 * If it's an inline extent, it can not have other extents
3456 		 * following it.
3457 		 */
3458 		if (btrfs_file_extent_type(path->nodes[0], ei) ==
3459 		    BTRFS_FILE_EXTENT_INLINE)
3460 			goto copy_inline_extent;
3461 
3462 		ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3463 		if (ext_len > aligned_end)
3464 			return -EOPNOTSUPP;
3465 
3466 		ret = btrfs_next_item(root, path);
3467 		if (ret < 0) {
3468 			return ret;
3469 		} else if (ret == 0) {
3470 			btrfs_item_key_to_cpu(path->nodes[0], &key,
3471 					      path->slots[0]);
3472 			if (key.objectid == btrfs_ino(dst) &&
3473 			    key.type == BTRFS_EXTENT_DATA_KEY)
3474 				return -EOPNOTSUPP;
3475 		}
3476 	}
3477 
3478 copy_inline_extent:
3479 	/*
3480 	 * We have no extent items, or we have an extent at offset 0 which may
3481 	 * or may not be inlined. All these cases are dealt the same way.
3482 	 */
3483 	if (i_size_read(dst) > datal) {
3484 		/*
3485 		 * If the destination inode has an inline extent...
3486 		 * This would require copying the data from the source inline
3487 		 * extent into the beginning of the destination's inline extent.
3488 		 * But this is really complex, both extents can be compressed
3489 		 * or just one of them, which would require decompressing and
3490 		 * re-compressing data (which could increase the new compressed
3491 		 * size, not allowing the compressed data to fit anymore in an
3492 		 * inline extent).
3493 		 * So just don't support this case for now (it should be rare,
3494 		 * we are not really saving space when cloning inline extents).
3495 		 */
3496 		return -EOPNOTSUPP;
3497 	}
3498 
3499 	btrfs_release_path(path);
3500 	ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3501 	if (ret)
3502 		return ret;
3503 	ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3504 	if (ret)
3505 		return ret;
3506 
3507 	if (skip) {
3508 		const u32 start = btrfs_file_extent_calc_inline_size(0);
3509 
3510 		memmove(inline_data + start, inline_data + start + skip, datal);
3511 	}
3512 
3513 	write_extent_buffer(path->nodes[0], inline_data,
3514 			    btrfs_item_ptr_offset(path->nodes[0],
3515 						  path->slots[0]),
3516 			    size);
3517 	inode_add_bytes(dst, datal);
3518 
3519 	return 0;
3520 }
3521 
3522 /**
3523  * btrfs_clone() - clone a range from inode file to another
3524  *
3525  * @src: Inode to clone from
3526  * @inode: Inode to clone to
3527  * @off: Offset within source to start clone from
3528  * @olen: Original length, passed by user, of range to clone
3529  * @olen_aligned: Block-aligned value of olen
3530  * @destoff: Offset within @inode to start clone
3531  * @no_time_update: Whether to update mtime/ctime on the target inode
3532  */
3533 static int btrfs_clone(struct inode *src, struct inode *inode,
3534 		       const u64 off, const u64 olen, const u64 olen_aligned,
3535 		       const u64 destoff, int no_time_update)
3536 {
3537 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3538 	struct btrfs_root *root = BTRFS_I(inode)->root;
3539 	struct btrfs_path *path = NULL;
3540 	struct extent_buffer *leaf;
3541 	struct btrfs_trans_handle *trans;
3542 	char *buf = NULL;
3543 	struct btrfs_key key;
3544 	u32 nritems;
3545 	int slot;
3546 	int ret;
3547 	const u64 len = olen_aligned;
3548 	u64 last_dest_end = destoff;
3549 
3550 	ret = -ENOMEM;
3551 	buf = kmalloc(fs_info->nodesize, GFP_KERNEL | __GFP_NOWARN);
3552 	if (!buf) {
3553 		buf = vmalloc(fs_info->nodesize);
3554 		if (!buf)
3555 			return ret;
3556 	}
3557 
3558 	path = btrfs_alloc_path();
3559 	if (!path) {
3560 		kvfree(buf);
3561 		return ret;
3562 	}
3563 
3564 	path->reada = READA_FORWARD;
3565 	/* clone data */
3566 	key.objectid = btrfs_ino(src);
3567 	key.type = BTRFS_EXTENT_DATA_KEY;
3568 	key.offset = off;
3569 
3570 	while (1) {
3571 		u64 next_key_min_offset = key.offset + 1;
3572 
3573 		/*
3574 		 * note the key will change type as we walk through the
3575 		 * tree.
3576 		 */
3577 		path->leave_spinning = 1;
3578 		ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3579 				0, 0);
3580 		if (ret < 0)
3581 			goto out;
3582 		/*
3583 		 * First search, if no extent item that starts at offset off was
3584 		 * found but the previous item is an extent item, it's possible
3585 		 * it might overlap our target range, therefore process it.
3586 		 */
3587 		if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3588 			btrfs_item_key_to_cpu(path->nodes[0], &key,
3589 					      path->slots[0] - 1);
3590 			if (key.type == BTRFS_EXTENT_DATA_KEY)
3591 				path->slots[0]--;
3592 		}
3593 
3594 		nritems = btrfs_header_nritems(path->nodes[0]);
3595 process_slot:
3596 		if (path->slots[0] >= nritems) {
3597 			ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3598 			if (ret < 0)
3599 				goto out;
3600 			if (ret > 0)
3601 				break;
3602 			nritems = btrfs_header_nritems(path->nodes[0]);
3603 		}
3604 		leaf = path->nodes[0];
3605 		slot = path->slots[0];
3606 
3607 		btrfs_item_key_to_cpu(leaf, &key, slot);
3608 		if (key.type > BTRFS_EXTENT_DATA_KEY ||
3609 		    key.objectid != btrfs_ino(src))
3610 			break;
3611 
3612 		if (key.type == BTRFS_EXTENT_DATA_KEY) {
3613 			struct btrfs_file_extent_item *extent;
3614 			int type;
3615 			u32 size;
3616 			struct btrfs_key new_key;
3617 			u64 disko = 0, diskl = 0;
3618 			u64 datao = 0, datal = 0;
3619 			u8 comp;
3620 			u64 drop_start;
3621 
3622 			extent = btrfs_item_ptr(leaf, slot,
3623 						struct btrfs_file_extent_item);
3624 			comp = btrfs_file_extent_compression(leaf, extent);
3625 			type = btrfs_file_extent_type(leaf, extent);
3626 			if (type == BTRFS_FILE_EXTENT_REG ||
3627 			    type == BTRFS_FILE_EXTENT_PREALLOC) {
3628 				disko = btrfs_file_extent_disk_bytenr(leaf,
3629 								      extent);
3630 				diskl = btrfs_file_extent_disk_num_bytes(leaf,
3631 								 extent);
3632 				datao = btrfs_file_extent_offset(leaf, extent);
3633 				datal = btrfs_file_extent_num_bytes(leaf,
3634 								    extent);
3635 			} else if (type == BTRFS_FILE_EXTENT_INLINE) {
3636 				/* take upper bound, may be compressed */
3637 				datal = btrfs_file_extent_ram_bytes(leaf,
3638 								    extent);
3639 			}
3640 
3641 			/*
3642 			 * The first search might have left us at an extent
3643 			 * item that ends before our target range's start, can
3644 			 * happen if we have holes and NO_HOLES feature enabled.
3645 			 */
3646 			if (key.offset + datal <= off) {
3647 				path->slots[0]++;
3648 				goto process_slot;
3649 			} else if (key.offset >= off + len) {
3650 				break;
3651 			}
3652 			next_key_min_offset = key.offset + datal;
3653 			size = btrfs_item_size_nr(leaf, slot);
3654 			read_extent_buffer(leaf, buf,
3655 					   btrfs_item_ptr_offset(leaf, slot),
3656 					   size);
3657 
3658 			btrfs_release_path(path);
3659 			path->leave_spinning = 0;
3660 
3661 			memcpy(&new_key, &key, sizeof(new_key));
3662 			new_key.objectid = btrfs_ino(inode);
3663 			if (off <= key.offset)
3664 				new_key.offset = key.offset + destoff - off;
3665 			else
3666 				new_key.offset = destoff;
3667 
3668 			/*
3669 			 * Deal with a hole that doesn't have an extent item
3670 			 * that represents it (NO_HOLES feature enabled).
3671 			 * This hole is either in the middle of the cloning
3672 			 * range or at the beginning (fully overlaps it or
3673 			 * partially overlaps it).
3674 			 */
3675 			if (new_key.offset != last_dest_end)
3676 				drop_start = last_dest_end;
3677 			else
3678 				drop_start = new_key.offset;
3679 
3680 			/*
3681 			 * 1 - adjusting old extent (we may have to split it)
3682 			 * 1 - add new extent
3683 			 * 1 - inode update
3684 			 */
3685 			trans = btrfs_start_transaction(root, 3);
3686 			if (IS_ERR(trans)) {
3687 				ret = PTR_ERR(trans);
3688 				goto out;
3689 			}
3690 
3691 			if (type == BTRFS_FILE_EXTENT_REG ||
3692 			    type == BTRFS_FILE_EXTENT_PREALLOC) {
3693 				/*
3694 				 *    a  | --- range to clone ---|  b
3695 				 * | ------------- extent ------------- |
3696 				 */
3697 
3698 				/* subtract range b */
3699 				if (key.offset + datal > off + len)
3700 					datal = off + len - key.offset;
3701 
3702 				/* subtract range a */
3703 				if (off > key.offset) {
3704 					datao += off - key.offset;
3705 					datal -= off - key.offset;
3706 				}
3707 
3708 				ret = btrfs_drop_extents(trans, root, inode,
3709 							 drop_start,
3710 							 new_key.offset + datal,
3711 							 1);
3712 				if (ret) {
3713 					if (ret != -EOPNOTSUPP)
3714 						btrfs_abort_transaction(trans,
3715 									ret);
3716 					btrfs_end_transaction(trans);
3717 					goto out;
3718 				}
3719 
3720 				ret = btrfs_insert_empty_item(trans, root, path,
3721 							      &new_key, size);
3722 				if (ret) {
3723 					btrfs_abort_transaction(trans, ret);
3724 					btrfs_end_transaction(trans);
3725 					goto out;
3726 				}
3727 
3728 				leaf = path->nodes[0];
3729 				slot = path->slots[0];
3730 				write_extent_buffer(leaf, buf,
3731 					    btrfs_item_ptr_offset(leaf, slot),
3732 					    size);
3733 
3734 				extent = btrfs_item_ptr(leaf, slot,
3735 						struct btrfs_file_extent_item);
3736 
3737 				/* disko == 0 means it's a hole */
3738 				if (!disko)
3739 					datao = 0;
3740 
3741 				btrfs_set_file_extent_offset(leaf, extent,
3742 							     datao);
3743 				btrfs_set_file_extent_num_bytes(leaf, extent,
3744 								datal);
3745 
3746 				if (disko) {
3747 					inode_add_bytes(inode, datal);
3748 					ret = btrfs_inc_extent_ref(trans,
3749 							fs_info,
3750 							disko, diskl, 0,
3751 							root->root_key.objectid,
3752 							btrfs_ino(inode),
3753 							new_key.offset - datao);
3754 					if (ret) {
3755 						btrfs_abort_transaction(trans,
3756 									ret);
3757 						btrfs_end_transaction(trans);
3758 						goto out;
3759 
3760 					}
3761 				}
3762 			} else if (type == BTRFS_FILE_EXTENT_INLINE) {
3763 				u64 skip = 0;
3764 				u64 trim = 0;
3765 
3766 				if (off > key.offset) {
3767 					skip = off - key.offset;
3768 					new_key.offset += skip;
3769 				}
3770 
3771 				if (key.offset + datal > off + len)
3772 					trim = key.offset + datal - (off + len);
3773 
3774 				if (comp && (skip || trim)) {
3775 					ret = -EINVAL;
3776 					btrfs_end_transaction(trans);
3777 					goto out;
3778 				}
3779 				size -= skip + trim;
3780 				datal -= skip + trim;
3781 
3782 				ret = clone_copy_inline_extent(src, inode,
3783 							       trans, path,
3784 							       &new_key,
3785 							       drop_start,
3786 							       datal,
3787 							       skip, size, buf);
3788 				if (ret) {
3789 					if (ret != -EOPNOTSUPP)
3790 						btrfs_abort_transaction(trans,
3791 									ret);
3792 					btrfs_end_transaction(trans);
3793 					goto out;
3794 				}
3795 				leaf = path->nodes[0];
3796 				slot = path->slots[0];
3797 			}
3798 
3799 			/* If we have an implicit hole (NO_HOLES feature). */
3800 			if (drop_start < new_key.offset)
3801 				clone_update_extent_map(inode, trans,
3802 						NULL, drop_start,
3803 						new_key.offset - drop_start);
3804 
3805 			clone_update_extent_map(inode, trans, path, 0, 0);
3806 
3807 			btrfs_mark_buffer_dirty(leaf);
3808 			btrfs_release_path(path);
3809 
3810 			last_dest_end = ALIGN(new_key.offset + datal,
3811 					      fs_info->sectorsize);
3812 			ret = clone_finish_inode_update(trans, inode,
3813 							last_dest_end,
3814 							destoff, olen,
3815 							no_time_update);
3816 			if (ret)
3817 				goto out;
3818 			if (new_key.offset + datal >= destoff + len)
3819 				break;
3820 		}
3821 		btrfs_release_path(path);
3822 		key.offset = next_key_min_offset;
3823 
3824 		if (fatal_signal_pending(current)) {
3825 			ret = -EINTR;
3826 			goto out;
3827 		}
3828 	}
3829 	ret = 0;
3830 
3831 	if (last_dest_end < destoff + len) {
3832 		/*
3833 		 * We have an implicit hole (NO_HOLES feature is enabled) that
3834 		 * fully or partially overlaps our cloning range at its end.
3835 		 */
3836 		btrfs_release_path(path);
3837 
3838 		/*
3839 		 * 1 - remove extent(s)
3840 		 * 1 - inode update
3841 		 */
3842 		trans = btrfs_start_transaction(root, 2);
3843 		if (IS_ERR(trans)) {
3844 			ret = PTR_ERR(trans);
3845 			goto out;
3846 		}
3847 		ret = btrfs_drop_extents(trans, root, inode,
3848 					 last_dest_end, destoff + len, 1);
3849 		if (ret) {
3850 			if (ret != -EOPNOTSUPP)
3851 				btrfs_abort_transaction(trans, ret);
3852 			btrfs_end_transaction(trans);
3853 			goto out;
3854 		}
3855 		clone_update_extent_map(inode, trans, NULL, last_dest_end,
3856 					destoff + len - last_dest_end);
3857 		ret = clone_finish_inode_update(trans, inode, destoff + len,
3858 						destoff, olen, no_time_update);
3859 	}
3860 
3861 out:
3862 	btrfs_free_path(path);
3863 	kvfree(buf);
3864 	return ret;
3865 }
3866 
3867 static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
3868 					u64 off, u64 olen, u64 destoff)
3869 {
3870 	struct inode *inode = file_inode(file);
3871 	struct inode *src = file_inode(file_src);
3872 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3873 	struct btrfs_root *root = BTRFS_I(inode)->root;
3874 	int ret;
3875 	u64 len = olen;
3876 	u64 bs = fs_info->sb->s_blocksize;
3877 	int same_inode = src == inode;
3878 
3879 	/*
3880 	 * TODO:
3881 	 * - split compressed inline extents.  annoying: we need to
3882 	 *   decompress into destination's address_space (the file offset
3883 	 *   may change, so source mapping won't do), then recompress (or
3884 	 *   otherwise reinsert) a subrange.
3885 	 *
3886 	 * - split destination inode's inline extents.  The inline extents can
3887 	 *   be either compressed or non-compressed.
3888 	 */
3889 
3890 	if (btrfs_root_readonly(root))
3891 		return -EROFS;
3892 
3893 	if (file_src->f_path.mnt != file->f_path.mnt ||
3894 	    src->i_sb != inode->i_sb)
3895 		return -EXDEV;
3896 
3897 	/* don't make the dst file partly checksummed */
3898 	if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3899 	    (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3900 		return -EINVAL;
3901 
3902 	if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3903 		return -EISDIR;
3904 
3905 	if (!same_inode) {
3906 		btrfs_double_inode_lock(src, inode);
3907 	} else {
3908 		inode_lock(src);
3909 	}
3910 
3911 	/* determine range to clone */
3912 	ret = -EINVAL;
3913 	if (off + len > src->i_size || off + len < off)
3914 		goto out_unlock;
3915 	if (len == 0)
3916 		olen = len = src->i_size - off;
3917 	/* if we extend to eof, continue to block boundary */
3918 	if (off + len == src->i_size)
3919 		len = ALIGN(src->i_size, bs) - off;
3920 
3921 	if (len == 0) {
3922 		ret = 0;
3923 		goto out_unlock;
3924 	}
3925 
3926 	/* verify the end result is block aligned */
3927 	if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3928 	    !IS_ALIGNED(destoff, bs))
3929 		goto out_unlock;
3930 
3931 	/* verify if ranges are overlapped within the same file */
3932 	if (same_inode) {
3933 		if (destoff + len > off && destoff < off + len)
3934 			goto out_unlock;
3935 	}
3936 
3937 	if (destoff > inode->i_size) {
3938 		ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3939 		if (ret)
3940 			goto out_unlock;
3941 	}
3942 
3943 	/*
3944 	 * Lock the target range too. Right after we replace the file extent
3945 	 * items in the fs tree (which now point to the cloned data), we might
3946 	 * have a worker replace them with extent items relative to a write
3947 	 * operation that was issued before this clone operation (i.e. confront
3948 	 * with inode.c:btrfs_finish_ordered_io).
3949 	 */
3950 	if (same_inode) {
3951 		u64 lock_start = min_t(u64, off, destoff);
3952 		u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
3953 
3954 		ret = lock_extent_range(src, lock_start, lock_len, true);
3955 	} else {
3956 		ret = btrfs_double_extent_lock(src, off, inode, destoff, len,
3957 					       true);
3958 	}
3959 	ASSERT(ret == 0);
3960 	if (WARN_ON(ret)) {
3961 		/* ranges in the io trees already unlocked */
3962 		goto out_unlock;
3963 	}
3964 
3965 	ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
3966 
3967 	if (same_inode) {
3968 		u64 lock_start = min_t(u64, off, destoff);
3969 		u64 lock_end = max_t(u64, off, destoff) + len - 1;
3970 
3971 		unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
3972 	} else {
3973 		btrfs_double_extent_unlock(src, off, inode, destoff, len);
3974 	}
3975 	/*
3976 	 * Truncate page cache pages so that future reads will see the cloned
3977 	 * data immediately and not the previous data.
3978 	 */
3979 	truncate_inode_pages_range(&inode->i_data,
3980 				round_down(destoff, PAGE_SIZE),
3981 				round_up(destoff + len, PAGE_SIZE) - 1);
3982 out_unlock:
3983 	if (!same_inode)
3984 		btrfs_double_inode_unlock(src, inode);
3985 	else
3986 		inode_unlock(src);
3987 	return ret;
3988 }
3989 
3990 int btrfs_clone_file_range(struct file *src_file, loff_t off,
3991 		struct file *dst_file, loff_t destoff, u64 len)
3992 {
3993 	return btrfs_clone_files(dst_file, src_file, off, len, destoff);
3994 }
3995 
3996 /*
3997  * there are many ways the trans_start and trans_end ioctls can lead
3998  * to deadlocks.  They should only be used by applications that
3999  * basically own the machine, and have a very in depth understanding
4000  * of all the possible deadlocks and enospc problems.
4001  */
4002 static long btrfs_ioctl_trans_start(struct file *file)
4003 {
4004 	struct inode *inode = file_inode(file);
4005 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4006 	struct btrfs_root *root = BTRFS_I(inode)->root;
4007 	struct btrfs_trans_handle *trans;
4008 	int ret;
4009 
4010 	ret = -EPERM;
4011 	if (!capable(CAP_SYS_ADMIN))
4012 		goto out;
4013 
4014 	ret = -EINPROGRESS;
4015 	if (file->private_data)
4016 		goto out;
4017 
4018 	ret = -EROFS;
4019 	if (btrfs_root_readonly(root))
4020 		goto out;
4021 
4022 	ret = mnt_want_write_file(file);
4023 	if (ret)
4024 		goto out;
4025 
4026 	atomic_inc(&fs_info->open_ioctl_trans);
4027 
4028 	ret = -ENOMEM;
4029 	trans = btrfs_start_ioctl_transaction(root);
4030 	if (IS_ERR(trans))
4031 		goto out_drop;
4032 
4033 	file->private_data = trans;
4034 	return 0;
4035 
4036 out_drop:
4037 	atomic_dec(&fs_info->open_ioctl_trans);
4038 	mnt_drop_write_file(file);
4039 out:
4040 	return ret;
4041 }
4042 
4043 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
4044 {
4045 	struct inode *inode = file_inode(file);
4046 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4047 	struct btrfs_root *root = BTRFS_I(inode)->root;
4048 	struct btrfs_root *new_root;
4049 	struct btrfs_dir_item *di;
4050 	struct btrfs_trans_handle *trans;
4051 	struct btrfs_path *path;
4052 	struct btrfs_key location;
4053 	struct btrfs_disk_key disk_key;
4054 	u64 objectid = 0;
4055 	u64 dir_id;
4056 	int ret;
4057 
4058 	if (!capable(CAP_SYS_ADMIN))
4059 		return -EPERM;
4060 
4061 	ret = mnt_want_write_file(file);
4062 	if (ret)
4063 		return ret;
4064 
4065 	if (copy_from_user(&objectid, argp, sizeof(objectid))) {
4066 		ret = -EFAULT;
4067 		goto out;
4068 	}
4069 
4070 	if (!objectid)
4071 		objectid = BTRFS_FS_TREE_OBJECTID;
4072 
4073 	location.objectid = objectid;
4074 	location.type = BTRFS_ROOT_ITEM_KEY;
4075 	location.offset = (u64)-1;
4076 
4077 	new_root = btrfs_read_fs_root_no_name(fs_info, &location);
4078 	if (IS_ERR(new_root)) {
4079 		ret = PTR_ERR(new_root);
4080 		goto out;
4081 	}
4082 
4083 	path = btrfs_alloc_path();
4084 	if (!path) {
4085 		ret = -ENOMEM;
4086 		goto out;
4087 	}
4088 	path->leave_spinning = 1;
4089 
4090 	trans = btrfs_start_transaction(root, 1);
4091 	if (IS_ERR(trans)) {
4092 		btrfs_free_path(path);
4093 		ret = PTR_ERR(trans);
4094 		goto out;
4095 	}
4096 
4097 	dir_id = btrfs_super_root_dir(fs_info->super_copy);
4098 	di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
4099 				   dir_id, "default", 7, 1);
4100 	if (IS_ERR_OR_NULL(di)) {
4101 		btrfs_free_path(path);
4102 		btrfs_end_transaction(trans);
4103 		btrfs_err(fs_info,
4104 			  "Umm, you don't have the default diritem, this isn't going to work");
4105 		ret = -ENOENT;
4106 		goto out;
4107 	}
4108 
4109 	btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
4110 	btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
4111 	btrfs_mark_buffer_dirty(path->nodes[0]);
4112 	btrfs_free_path(path);
4113 
4114 	btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
4115 	btrfs_end_transaction(trans);
4116 out:
4117 	mnt_drop_write_file(file);
4118 	return ret;
4119 }
4120 
4121 void btrfs_get_block_group_info(struct list_head *groups_list,
4122 				struct btrfs_ioctl_space_info *space)
4123 {
4124 	struct btrfs_block_group_cache *block_group;
4125 
4126 	space->total_bytes = 0;
4127 	space->used_bytes = 0;
4128 	space->flags = 0;
4129 	list_for_each_entry(block_group, groups_list, list) {
4130 		space->flags = block_group->flags;
4131 		space->total_bytes += block_group->key.offset;
4132 		space->used_bytes +=
4133 			btrfs_block_group_used(&block_group->item);
4134 	}
4135 }
4136 
4137 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
4138 				   void __user *arg)
4139 {
4140 	struct btrfs_ioctl_space_args space_args;
4141 	struct btrfs_ioctl_space_info space;
4142 	struct btrfs_ioctl_space_info *dest;
4143 	struct btrfs_ioctl_space_info *dest_orig;
4144 	struct btrfs_ioctl_space_info __user *user_dest;
4145 	struct btrfs_space_info *info;
4146 	u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
4147 		       BTRFS_BLOCK_GROUP_SYSTEM,
4148 		       BTRFS_BLOCK_GROUP_METADATA,
4149 		       BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
4150 	int num_types = 4;
4151 	int alloc_size;
4152 	int ret = 0;
4153 	u64 slot_count = 0;
4154 	int i, c;
4155 
4156 	if (copy_from_user(&space_args,
4157 			   (struct btrfs_ioctl_space_args __user *)arg,
4158 			   sizeof(space_args)))
4159 		return -EFAULT;
4160 
4161 	for (i = 0; i < num_types; i++) {
4162 		struct btrfs_space_info *tmp;
4163 
4164 		info = NULL;
4165 		rcu_read_lock();
4166 		list_for_each_entry_rcu(tmp, &fs_info->space_info,
4167 					list) {
4168 			if (tmp->flags == types[i]) {
4169 				info = tmp;
4170 				break;
4171 			}
4172 		}
4173 		rcu_read_unlock();
4174 
4175 		if (!info)
4176 			continue;
4177 
4178 		down_read(&info->groups_sem);
4179 		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4180 			if (!list_empty(&info->block_groups[c]))
4181 				slot_count++;
4182 		}
4183 		up_read(&info->groups_sem);
4184 	}
4185 
4186 	/*
4187 	 * Global block reserve, exported as a space_info
4188 	 */
4189 	slot_count++;
4190 
4191 	/* space_slots == 0 means they are asking for a count */
4192 	if (space_args.space_slots == 0) {
4193 		space_args.total_spaces = slot_count;
4194 		goto out;
4195 	}
4196 
4197 	slot_count = min_t(u64, space_args.space_slots, slot_count);
4198 
4199 	alloc_size = sizeof(*dest) * slot_count;
4200 
4201 	/* we generally have at most 6 or so space infos, one for each raid
4202 	 * level.  So, a whole page should be more than enough for everyone
4203 	 */
4204 	if (alloc_size > PAGE_SIZE)
4205 		return -ENOMEM;
4206 
4207 	space_args.total_spaces = 0;
4208 	dest = kmalloc(alloc_size, GFP_KERNEL);
4209 	if (!dest)
4210 		return -ENOMEM;
4211 	dest_orig = dest;
4212 
4213 	/* now we have a buffer to copy into */
4214 	for (i = 0; i < num_types; i++) {
4215 		struct btrfs_space_info *tmp;
4216 
4217 		if (!slot_count)
4218 			break;
4219 
4220 		info = NULL;
4221 		rcu_read_lock();
4222 		list_for_each_entry_rcu(tmp, &fs_info->space_info,
4223 					list) {
4224 			if (tmp->flags == types[i]) {
4225 				info = tmp;
4226 				break;
4227 			}
4228 		}
4229 		rcu_read_unlock();
4230 
4231 		if (!info)
4232 			continue;
4233 		down_read(&info->groups_sem);
4234 		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4235 			if (!list_empty(&info->block_groups[c])) {
4236 				btrfs_get_block_group_info(
4237 					&info->block_groups[c], &space);
4238 				memcpy(dest, &space, sizeof(space));
4239 				dest++;
4240 				space_args.total_spaces++;
4241 				slot_count--;
4242 			}
4243 			if (!slot_count)
4244 				break;
4245 		}
4246 		up_read(&info->groups_sem);
4247 	}
4248 
4249 	/*
4250 	 * Add global block reserve
4251 	 */
4252 	if (slot_count) {
4253 		struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4254 
4255 		spin_lock(&block_rsv->lock);
4256 		space.total_bytes = block_rsv->size;
4257 		space.used_bytes = block_rsv->size - block_rsv->reserved;
4258 		spin_unlock(&block_rsv->lock);
4259 		space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4260 		memcpy(dest, &space, sizeof(space));
4261 		space_args.total_spaces++;
4262 	}
4263 
4264 	user_dest = (struct btrfs_ioctl_space_info __user *)
4265 		(arg + sizeof(struct btrfs_ioctl_space_args));
4266 
4267 	if (copy_to_user(user_dest, dest_orig, alloc_size))
4268 		ret = -EFAULT;
4269 
4270 	kfree(dest_orig);
4271 out:
4272 	if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4273 		ret = -EFAULT;
4274 
4275 	return ret;
4276 }
4277 
4278 /*
4279  * there are many ways the trans_start and trans_end ioctls can lead
4280  * to deadlocks.  They should only be used by applications that
4281  * basically own the machine, and have a very in depth understanding
4282  * of all the possible deadlocks and enospc problems.
4283  */
4284 long btrfs_ioctl_trans_end(struct file *file)
4285 {
4286 	struct inode *inode = file_inode(file);
4287 	struct btrfs_root *root = BTRFS_I(inode)->root;
4288 	struct btrfs_trans_handle *trans;
4289 
4290 	trans = file->private_data;
4291 	if (!trans)
4292 		return -EINVAL;
4293 	file->private_data = NULL;
4294 
4295 	btrfs_end_transaction(trans);
4296 
4297 	atomic_dec(&root->fs_info->open_ioctl_trans);
4298 
4299 	mnt_drop_write_file(file);
4300 	return 0;
4301 }
4302 
4303 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4304 					    void __user *argp)
4305 {
4306 	struct btrfs_trans_handle *trans;
4307 	u64 transid;
4308 	int ret;
4309 
4310 	trans = btrfs_attach_transaction_barrier(root);
4311 	if (IS_ERR(trans)) {
4312 		if (PTR_ERR(trans) != -ENOENT)
4313 			return PTR_ERR(trans);
4314 
4315 		/* No running transaction, don't bother */
4316 		transid = root->fs_info->last_trans_committed;
4317 		goto out;
4318 	}
4319 	transid = trans->transid;
4320 	ret = btrfs_commit_transaction_async(trans, 0);
4321 	if (ret) {
4322 		btrfs_end_transaction(trans);
4323 		return ret;
4324 	}
4325 out:
4326 	if (argp)
4327 		if (copy_to_user(argp, &transid, sizeof(transid)))
4328 			return -EFAULT;
4329 	return 0;
4330 }
4331 
4332 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
4333 					   void __user *argp)
4334 {
4335 	u64 transid;
4336 
4337 	if (argp) {
4338 		if (copy_from_user(&transid, argp, sizeof(transid)))
4339 			return -EFAULT;
4340 	} else {
4341 		transid = 0;  /* current trans */
4342 	}
4343 	return btrfs_wait_for_commit(fs_info, transid);
4344 }
4345 
4346 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4347 {
4348 	struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
4349 	struct btrfs_ioctl_scrub_args *sa;
4350 	int ret;
4351 
4352 	if (!capable(CAP_SYS_ADMIN))
4353 		return -EPERM;
4354 
4355 	sa = memdup_user(arg, sizeof(*sa));
4356 	if (IS_ERR(sa))
4357 		return PTR_ERR(sa);
4358 
4359 	if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4360 		ret = mnt_want_write_file(file);
4361 		if (ret)
4362 			goto out;
4363 	}
4364 
4365 	ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
4366 			      &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4367 			      0);
4368 
4369 	if (copy_to_user(arg, sa, sizeof(*sa)))
4370 		ret = -EFAULT;
4371 
4372 	if (!(sa->flags & BTRFS_SCRUB_READONLY))
4373 		mnt_drop_write_file(file);
4374 out:
4375 	kfree(sa);
4376 	return ret;
4377 }
4378 
4379 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
4380 {
4381 	if (!capable(CAP_SYS_ADMIN))
4382 		return -EPERM;
4383 
4384 	return btrfs_scrub_cancel(fs_info);
4385 }
4386 
4387 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
4388 				       void __user *arg)
4389 {
4390 	struct btrfs_ioctl_scrub_args *sa;
4391 	int ret;
4392 
4393 	if (!capable(CAP_SYS_ADMIN))
4394 		return -EPERM;
4395 
4396 	sa = memdup_user(arg, sizeof(*sa));
4397 	if (IS_ERR(sa))
4398 		return PTR_ERR(sa);
4399 
4400 	ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
4401 
4402 	if (copy_to_user(arg, sa, sizeof(*sa)))
4403 		ret = -EFAULT;
4404 
4405 	kfree(sa);
4406 	return ret;
4407 }
4408 
4409 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
4410 				      void __user *arg)
4411 {
4412 	struct btrfs_ioctl_get_dev_stats *sa;
4413 	int ret;
4414 
4415 	sa = memdup_user(arg, sizeof(*sa));
4416 	if (IS_ERR(sa))
4417 		return PTR_ERR(sa);
4418 
4419 	if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4420 		kfree(sa);
4421 		return -EPERM;
4422 	}
4423 
4424 	ret = btrfs_get_dev_stats(fs_info, sa);
4425 
4426 	if (copy_to_user(arg, sa, sizeof(*sa)))
4427 		ret = -EFAULT;
4428 
4429 	kfree(sa);
4430 	return ret;
4431 }
4432 
4433 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
4434 				    void __user *arg)
4435 {
4436 	struct btrfs_ioctl_dev_replace_args *p;
4437 	int ret;
4438 
4439 	if (!capable(CAP_SYS_ADMIN))
4440 		return -EPERM;
4441 
4442 	p = memdup_user(arg, sizeof(*p));
4443 	if (IS_ERR(p))
4444 		return PTR_ERR(p);
4445 
4446 	switch (p->cmd) {
4447 	case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4448 		if (fs_info->sb->s_flags & MS_RDONLY) {
4449 			ret = -EROFS;
4450 			goto out;
4451 		}
4452 		if (atomic_xchg(
4453 			&fs_info->mutually_exclusive_operation_running, 1)) {
4454 			ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4455 		} else {
4456 			ret = btrfs_dev_replace_by_ioctl(fs_info, p);
4457 			atomic_set(
4458 			 &fs_info->mutually_exclusive_operation_running, 0);
4459 		}
4460 		break;
4461 	case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4462 		btrfs_dev_replace_status(fs_info, p);
4463 		ret = 0;
4464 		break;
4465 	case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4466 		ret = btrfs_dev_replace_cancel(fs_info, p);
4467 		break;
4468 	default:
4469 		ret = -EINVAL;
4470 		break;
4471 	}
4472 
4473 	if (copy_to_user(arg, p, sizeof(*p)))
4474 		ret = -EFAULT;
4475 out:
4476 	kfree(p);
4477 	return ret;
4478 }
4479 
4480 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4481 {
4482 	int ret = 0;
4483 	int i;
4484 	u64 rel_ptr;
4485 	int size;
4486 	struct btrfs_ioctl_ino_path_args *ipa = NULL;
4487 	struct inode_fs_paths *ipath = NULL;
4488 	struct btrfs_path *path;
4489 
4490 	if (!capable(CAP_DAC_READ_SEARCH))
4491 		return -EPERM;
4492 
4493 	path = btrfs_alloc_path();
4494 	if (!path) {
4495 		ret = -ENOMEM;
4496 		goto out;
4497 	}
4498 
4499 	ipa = memdup_user(arg, sizeof(*ipa));
4500 	if (IS_ERR(ipa)) {
4501 		ret = PTR_ERR(ipa);
4502 		ipa = NULL;
4503 		goto out;
4504 	}
4505 
4506 	size = min_t(u32, ipa->size, 4096);
4507 	ipath = init_ipath(size, root, path);
4508 	if (IS_ERR(ipath)) {
4509 		ret = PTR_ERR(ipath);
4510 		ipath = NULL;
4511 		goto out;
4512 	}
4513 
4514 	ret = paths_from_inode(ipa->inum, ipath);
4515 	if (ret < 0)
4516 		goto out;
4517 
4518 	for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4519 		rel_ptr = ipath->fspath->val[i] -
4520 			  (u64)(unsigned long)ipath->fspath->val;
4521 		ipath->fspath->val[i] = rel_ptr;
4522 	}
4523 
4524 	ret = copy_to_user((void *)(unsigned long)ipa->fspath,
4525 			   (void *)(unsigned long)ipath->fspath, size);
4526 	if (ret) {
4527 		ret = -EFAULT;
4528 		goto out;
4529 	}
4530 
4531 out:
4532 	btrfs_free_path(path);
4533 	free_ipath(ipath);
4534 	kfree(ipa);
4535 
4536 	return ret;
4537 }
4538 
4539 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4540 {
4541 	struct btrfs_data_container *inodes = ctx;
4542 	const size_t c = 3 * sizeof(u64);
4543 
4544 	if (inodes->bytes_left >= c) {
4545 		inodes->bytes_left -= c;
4546 		inodes->val[inodes->elem_cnt] = inum;
4547 		inodes->val[inodes->elem_cnt + 1] = offset;
4548 		inodes->val[inodes->elem_cnt + 2] = root;
4549 		inodes->elem_cnt += 3;
4550 	} else {
4551 		inodes->bytes_missing += c - inodes->bytes_left;
4552 		inodes->bytes_left = 0;
4553 		inodes->elem_missed += 3;
4554 	}
4555 
4556 	return 0;
4557 }
4558 
4559 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
4560 					void __user *arg)
4561 {
4562 	int ret = 0;
4563 	int size;
4564 	struct btrfs_ioctl_logical_ino_args *loi;
4565 	struct btrfs_data_container *inodes = NULL;
4566 	struct btrfs_path *path = NULL;
4567 
4568 	if (!capable(CAP_SYS_ADMIN))
4569 		return -EPERM;
4570 
4571 	loi = memdup_user(arg, sizeof(*loi));
4572 	if (IS_ERR(loi))
4573 		return PTR_ERR(loi);
4574 
4575 	path = btrfs_alloc_path();
4576 	if (!path) {
4577 		ret = -ENOMEM;
4578 		goto out;
4579 	}
4580 
4581 	size = min_t(u32, loi->size, SZ_64K);
4582 	inodes = init_data_container(size);
4583 	if (IS_ERR(inodes)) {
4584 		ret = PTR_ERR(inodes);
4585 		inodes = NULL;
4586 		goto out;
4587 	}
4588 
4589 	ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
4590 					  build_ino_list, inodes);
4591 	if (ret == -EINVAL)
4592 		ret = -ENOENT;
4593 	if (ret < 0)
4594 		goto out;
4595 
4596 	ret = copy_to_user((void *)(unsigned long)loi->inodes,
4597 			   (void *)(unsigned long)inodes, size);
4598 	if (ret)
4599 		ret = -EFAULT;
4600 
4601 out:
4602 	btrfs_free_path(path);
4603 	vfree(inodes);
4604 	kfree(loi);
4605 
4606 	return ret;
4607 }
4608 
4609 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4610 			       struct btrfs_ioctl_balance_args *bargs)
4611 {
4612 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4613 
4614 	bargs->flags = bctl->flags;
4615 
4616 	if (atomic_read(&fs_info->balance_running))
4617 		bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4618 	if (atomic_read(&fs_info->balance_pause_req))
4619 		bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4620 	if (atomic_read(&fs_info->balance_cancel_req))
4621 		bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4622 
4623 	memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4624 	memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4625 	memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4626 
4627 	if (lock) {
4628 		spin_lock(&fs_info->balance_lock);
4629 		memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4630 		spin_unlock(&fs_info->balance_lock);
4631 	} else {
4632 		memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4633 	}
4634 }
4635 
4636 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4637 {
4638 	struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4639 	struct btrfs_fs_info *fs_info = root->fs_info;
4640 	struct btrfs_ioctl_balance_args *bargs;
4641 	struct btrfs_balance_control *bctl;
4642 	bool need_unlock; /* for mut. excl. ops lock */
4643 	int ret;
4644 
4645 	if (!capable(CAP_SYS_ADMIN))
4646 		return -EPERM;
4647 
4648 	ret = mnt_want_write_file(file);
4649 	if (ret)
4650 		return ret;
4651 
4652 again:
4653 	if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
4654 		mutex_lock(&fs_info->volume_mutex);
4655 		mutex_lock(&fs_info->balance_mutex);
4656 		need_unlock = true;
4657 		goto locked;
4658 	}
4659 
4660 	/*
4661 	 * mut. excl. ops lock is locked.  Three possibilities:
4662 	 *   (1) some other op is running
4663 	 *   (2) balance is running
4664 	 *   (3) balance is paused -- special case (think resume)
4665 	 */
4666 	mutex_lock(&fs_info->balance_mutex);
4667 	if (fs_info->balance_ctl) {
4668 		/* this is either (2) or (3) */
4669 		if (!atomic_read(&fs_info->balance_running)) {
4670 			mutex_unlock(&fs_info->balance_mutex);
4671 			if (!mutex_trylock(&fs_info->volume_mutex))
4672 				goto again;
4673 			mutex_lock(&fs_info->balance_mutex);
4674 
4675 			if (fs_info->balance_ctl &&
4676 			    !atomic_read(&fs_info->balance_running)) {
4677 				/* this is (3) */
4678 				need_unlock = false;
4679 				goto locked;
4680 			}
4681 
4682 			mutex_unlock(&fs_info->balance_mutex);
4683 			mutex_unlock(&fs_info->volume_mutex);
4684 			goto again;
4685 		} else {
4686 			/* this is (2) */
4687 			mutex_unlock(&fs_info->balance_mutex);
4688 			ret = -EINPROGRESS;
4689 			goto out;
4690 		}
4691 	} else {
4692 		/* this is (1) */
4693 		mutex_unlock(&fs_info->balance_mutex);
4694 		ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4695 		goto out;
4696 	}
4697 
4698 locked:
4699 	BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
4700 
4701 	if (arg) {
4702 		bargs = memdup_user(arg, sizeof(*bargs));
4703 		if (IS_ERR(bargs)) {
4704 			ret = PTR_ERR(bargs);
4705 			goto out_unlock;
4706 		}
4707 
4708 		if (bargs->flags & BTRFS_BALANCE_RESUME) {
4709 			if (!fs_info->balance_ctl) {
4710 				ret = -ENOTCONN;
4711 				goto out_bargs;
4712 			}
4713 
4714 			bctl = fs_info->balance_ctl;
4715 			spin_lock(&fs_info->balance_lock);
4716 			bctl->flags |= BTRFS_BALANCE_RESUME;
4717 			spin_unlock(&fs_info->balance_lock);
4718 
4719 			goto do_balance;
4720 		}
4721 	} else {
4722 		bargs = NULL;
4723 	}
4724 
4725 	if (fs_info->balance_ctl) {
4726 		ret = -EINPROGRESS;
4727 		goto out_bargs;
4728 	}
4729 
4730 	bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4731 	if (!bctl) {
4732 		ret = -ENOMEM;
4733 		goto out_bargs;
4734 	}
4735 
4736 	bctl->fs_info = fs_info;
4737 	if (arg) {
4738 		memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4739 		memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4740 		memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4741 
4742 		bctl->flags = bargs->flags;
4743 	} else {
4744 		/* balance everything - no filters */
4745 		bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4746 	}
4747 
4748 	if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4749 		ret = -EINVAL;
4750 		goto out_bctl;
4751 	}
4752 
4753 do_balance:
4754 	/*
4755 	 * Ownership of bctl and mutually_exclusive_operation_running
4756 	 * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4757 	 * or, if restriper was paused all the way until unmount, in
4758 	 * free_fs_info.  mutually_exclusive_operation_running is
4759 	 * cleared in __cancel_balance.
4760 	 */
4761 	need_unlock = false;
4762 
4763 	ret = btrfs_balance(bctl, bargs);
4764 	bctl = NULL;
4765 
4766 	if (arg) {
4767 		if (copy_to_user(arg, bargs, sizeof(*bargs)))
4768 			ret = -EFAULT;
4769 	}
4770 
4771 out_bctl:
4772 	kfree(bctl);
4773 out_bargs:
4774 	kfree(bargs);
4775 out_unlock:
4776 	mutex_unlock(&fs_info->balance_mutex);
4777 	mutex_unlock(&fs_info->volume_mutex);
4778 	if (need_unlock)
4779 		atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
4780 out:
4781 	mnt_drop_write_file(file);
4782 	return ret;
4783 }
4784 
4785 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4786 {
4787 	if (!capable(CAP_SYS_ADMIN))
4788 		return -EPERM;
4789 
4790 	switch (cmd) {
4791 	case BTRFS_BALANCE_CTL_PAUSE:
4792 		return btrfs_pause_balance(fs_info);
4793 	case BTRFS_BALANCE_CTL_CANCEL:
4794 		return btrfs_cancel_balance(fs_info);
4795 	}
4796 
4797 	return -EINVAL;
4798 }
4799 
4800 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4801 					 void __user *arg)
4802 {
4803 	struct btrfs_ioctl_balance_args *bargs;
4804 	int ret = 0;
4805 
4806 	if (!capable(CAP_SYS_ADMIN))
4807 		return -EPERM;
4808 
4809 	mutex_lock(&fs_info->balance_mutex);
4810 	if (!fs_info->balance_ctl) {
4811 		ret = -ENOTCONN;
4812 		goto out;
4813 	}
4814 
4815 	bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4816 	if (!bargs) {
4817 		ret = -ENOMEM;
4818 		goto out;
4819 	}
4820 
4821 	update_ioctl_balance_args(fs_info, 1, bargs);
4822 
4823 	if (copy_to_user(arg, bargs, sizeof(*bargs)))
4824 		ret = -EFAULT;
4825 
4826 	kfree(bargs);
4827 out:
4828 	mutex_unlock(&fs_info->balance_mutex);
4829 	return ret;
4830 }
4831 
4832 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4833 {
4834 	struct inode *inode = file_inode(file);
4835 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4836 	struct btrfs_ioctl_quota_ctl_args *sa;
4837 	struct btrfs_trans_handle *trans = NULL;
4838 	int ret;
4839 	int err;
4840 
4841 	if (!capable(CAP_SYS_ADMIN))
4842 		return -EPERM;
4843 
4844 	ret = mnt_want_write_file(file);
4845 	if (ret)
4846 		return ret;
4847 
4848 	sa = memdup_user(arg, sizeof(*sa));
4849 	if (IS_ERR(sa)) {
4850 		ret = PTR_ERR(sa);
4851 		goto drop_write;
4852 	}
4853 
4854 	down_write(&fs_info->subvol_sem);
4855 	trans = btrfs_start_transaction(fs_info->tree_root, 2);
4856 	if (IS_ERR(trans)) {
4857 		ret = PTR_ERR(trans);
4858 		goto out;
4859 	}
4860 
4861 	switch (sa->cmd) {
4862 	case BTRFS_QUOTA_CTL_ENABLE:
4863 		ret = btrfs_quota_enable(trans, fs_info);
4864 		break;
4865 	case BTRFS_QUOTA_CTL_DISABLE:
4866 		ret = btrfs_quota_disable(trans, fs_info);
4867 		break;
4868 	default:
4869 		ret = -EINVAL;
4870 		break;
4871 	}
4872 
4873 	err = btrfs_commit_transaction(trans);
4874 	if (err && !ret)
4875 		ret = err;
4876 out:
4877 	kfree(sa);
4878 	up_write(&fs_info->subvol_sem);
4879 drop_write:
4880 	mnt_drop_write_file(file);
4881 	return ret;
4882 }
4883 
4884 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4885 {
4886 	struct inode *inode = file_inode(file);
4887 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4888 	struct btrfs_root *root = BTRFS_I(inode)->root;
4889 	struct btrfs_ioctl_qgroup_assign_args *sa;
4890 	struct btrfs_trans_handle *trans;
4891 	int ret;
4892 	int err;
4893 
4894 	if (!capable(CAP_SYS_ADMIN))
4895 		return -EPERM;
4896 
4897 	ret = mnt_want_write_file(file);
4898 	if (ret)
4899 		return ret;
4900 
4901 	sa = memdup_user(arg, sizeof(*sa));
4902 	if (IS_ERR(sa)) {
4903 		ret = PTR_ERR(sa);
4904 		goto drop_write;
4905 	}
4906 
4907 	trans = btrfs_join_transaction(root);
4908 	if (IS_ERR(trans)) {
4909 		ret = PTR_ERR(trans);
4910 		goto out;
4911 	}
4912 
4913 	/* FIXME: check if the IDs really exist */
4914 	if (sa->assign) {
4915 		ret = btrfs_add_qgroup_relation(trans, fs_info,
4916 						sa->src, sa->dst);
4917 	} else {
4918 		ret = btrfs_del_qgroup_relation(trans, fs_info,
4919 						sa->src, sa->dst);
4920 	}
4921 
4922 	/* update qgroup status and info */
4923 	err = btrfs_run_qgroups(trans, fs_info);
4924 	if (err < 0)
4925 		btrfs_handle_fs_error(fs_info, err,
4926 				      "failed to update qgroup status and info");
4927 	err = btrfs_end_transaction(trans);
4928 	if (err && !ret)
4929 		ret = err;
4930 
4931 out:
4932 	kfree(sa);
4933 drop_write:
4934 	mnt_drop_write_file(file);
4935 	return ret;
4936 }
4937 
4938 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4939 {
4940 	struct inode *inode = file_inode(file);
4941 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4942 	struct btrfs_root *root = BTRFS_I(inode)->root;
4943 	struct btrfs_ioctl_qgroup_create_args *sa;
4944 	struct btrfs_trans_handle *trans;
4945 	int ret;
4946 	int err;
4947 
4948 	if (!capable(CAP_SYS_ADMIN))
4949 		return -EPERM;
4950 
4951 	ret = mnt_want_write_file(file);
4952 	if (ret)
4953 		return ret;
4954 
4955 	sa = memdup_user(arg, sizeof(*sa));
4956 	if (IS_ERR(sa)) {
4957 		ret = PTR_ERR(sa);
4958 		goto drop_write;
4959 	}
4960 
4961 	if (!sa->qgroupid) {
4962 		ret = -EINVAL;
4963 		goto out;
4964 	}
4965 
4966 	trans = btrfs_join_transaction(root);
4967 	if (IS_ERR(trans)) {
4968 		ret = PTR_ERR(trans);
4969 		goto out;
4970 	}
4971 
4972 	/* FIXME: check if the IDs really exist */
4973 	if (sa->create) {
4974 		ret = btrfs_create_qgroup(trans, fs_info, sa->qgroupid);
4975 	} else {
4976 		ret = btrfs_remove_qgroup(trans, fs_info, sa->qgroupid);
4977 	}
4978 
4979 	err = btrfs_end_transaction(trans);
4980 	if (err && !ret)
4981 		ret = err;
4982 
4983 out:
4984 	kfree(sa);
4985 drop_write:
4986 	mnt_drop_write_file(file);
4987 	return ret;
4988 }
4989 
4990 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4991 {
4992 	struct inode *inode = file_inode(file);
4993 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4994 	struct btrfs_root *root = BTRFS_I(inode)->root;
4995 	struct btrfs_ioctl_qgroup_limit_args *sa;
4996 	struct btrfs_trans_handle *trans;
4997 	int ret;
4998 	int err;
4999 	u64 qgroupid;
5000 
5001 	if (!capable(CAP_SYS_ADMIN))
5002 		return -EPERM;
5003 
5004 	ret = mnt_want_write_file(file);
5005 	if (ret)
5006 		return ret;
5007 
5008 	sa = memdup_user(arg, sizeof(*sa));
5009 	if (IS_ERR(sa)) {
5010 		ret = PTR_ERR(sa);
5011 		goto drop_write;
5012 	}
5013 
5014 	trans = btrfs_join_transaction(root);
5015 	if (IS_ERR(trans)) {
5016 		ret = PTR_ERR(trans);
5017 		goto out;
5018 	}
5019 
5020 	qgroupid = sa->qgroupid;
5021 	if (!qgroupid) {
5022 		/* take the current subvol as qgroup */
5023 		qgroupid = root->root_key.objectid;
5024 	}
5025 
5026 	/* FIXME: check if the IDs really exist */
5027 	ret = btrfs_limit_qgroup(trans, fs_info, qgroupid, &sa->lim);
5028 
5029 	err = btrfs_end_transaction(trans);
5030 	if (err && !ret)
5031 		ret = err;
5032 
5033 out:
5034 	kfree(sa);
5035 drop_write:
5036 	mnt_drop_write_file(file);
5037 	return ret;
5038 }
5039 
5040 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
5041 {
5042 	struct inode *inode = file_inode(file);
5043 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5044 	struct btrfs_ioctl_quota_rescan_args *qsa;
5045 	int ret;
5046 
5047 	if (!capable(CAP_SYS_ADMIN))
5048 		return -EPERM;
5049 
5050 	ret = mnt_want_write_file(file);
5051 	if (ret)
5052 		return ret;
5053 
5054 	qsa = memdup_user(arg, sizeof(*qsa));
5055 	if (IS_ERR(qsa)) {
5056 		ret = PTR_ERR(qsa);
5057 		goto drop_write;
5058 	}
5059 
5060 	if (qsa->flags) {
5061 		ret = -EINVAL;
5062 		goto out;
5063 	}
5064 
5065 	ret = btrfs_qgroup_rescan(fs_info);
5066 
5067 out:
5068 	kfree(qsa);
5069 drop_write:
5070 	mnt_drop_write_file(file);
5071 	return ret;
5072 }
5073 
5074 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
5075 {
5076 	struct inode *inode = file_inode(file);
5077 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5078 	struct btrfs_ioctl_quota_rescan_args *qsa;
5079 	int ret = 0;
5080 
5081 	if (!capable(CAP_SYS_ADMIN))
5082 		return -EPERM;
5083 
5084 	qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
5085 	if (!qsa)
5086 		return -ENOMEM;
5087 
5088 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
5089 		qsa->flags = 1;
5090 		qsa->progress = fs_info->qgroup_rescan_progress.objectid;
5091 	}
5092 
5093 	if (copy_to_user(arg, qsa, sizeof(*qsa)))
5094 		ret = -EFAULT;
5095 
5096 	kfree(qsa);
5097 	return ret;
5098 }
5099 
5100 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
5101 {
5102 	struct inode *inode = file_inode(file);
5103 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5104 
5105 	if (!capable(CAP_SYS_ADMIN))
5106 		return -EPERM;
5107 
5108 	return btrfs_qgroup_wait_for_completion(fs_info, true);
5109 }
5110 
5111 static long _btrfs_ioctl_set_received_subvol(struct file *file,
5112 					    struct btrfs_ioctl_received_subvol_args *sa)
5113 {
5114 	struct inode *inode = file_inode(file);
5115 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5116 	struct btrfs_root *root = BTRFS_I(inode)->root;
5117 	struct btrfs_root_item *root_item = &root->root_item;
5118 	struct btrfs_trans_handle *trans;
5119 	struct timespec ct = current_time(inode);
5120 	int ret = 0;
5121 	int received_uuid_changed;
5122 
5123 	if (!inode_owner_or_capable(inode))
5124 		return -EPERM;
5125 
5126 	ret = mnt_want_write_file(file);
5127 	if (ret < 0)
5128 		return ret;
5129 
5130 	down_write(&fs_info->subvol_sem);
5131 
5132 	if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
5133 		ret = -EINVAL;
5134 		goto out;
5135 	}
5136 
5137 	if (btrfs_root_readonly(root)) {
5138 		ret = -EROFS;
5139 		goto out;
5140 	}
5141 
5142 	/*
5143 	 * 1 - root item
5144 	 * 2 - uuid items (received uuid + subvol uuid)
5145 	 */
5146 	trans = btrfs_start_transaction(root, 3);
5147 	if (IS_ERR(trans)) {
5148 		ret = PTR_ERR(trans);
5149 		trans = NULL;
5150 		goto out;
5151 	}
5152 
5153 	sa->rtransid = trans->transid;
5154 	sa->rtime.sec = ct.tv_sec;
5155 	sa->rtime.nsec = ct.tv_nsec;
5156 
5157 	received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5158 				       BTRFS_UUID_SIZE);
5159 	if (received_uuid_changed &&
5160 	    !btrfs_is_empty_uuid(root_item->received_uuid))
5161 		btrfs_uuid_tree_rem(trans, fs_info, root_item->received_uuid,
5162 				    BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5163 				    root->root_key.objectid);
5164 	memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5165 	btrfs_set_root_stransid(root_item, sa->stransid);
5166 	btrfs_set_root_rtransid(root_item, sa->rtransid);
5167 	btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5168 	btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5169 	btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5170 	btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5171 
5172 	ret = btrfs_update_root(trans, fs_info->tree_root,
5173 				&root->root_key, &root->root_item);
5174 	if (ret < 0) {
5175 		btrfs_end_transaction(trans);
5176 		goto out;
5177 	}
5178 	if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5179 		ret = btrfs_uuid_tree_add(trans, fs_info, sa->uuid,
5180 					  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5181 					  root->root_key.objectid);
5182 		if (ret < 0 && ret != -EEXIST) {
5183 			btrfs_abort_transaction(trans, ret);
5184 			goto out;
5185 		}
5186 	}
5187 	ret = btrfs_commit_transaction(trans);
5188 	if (ret < 0) {
5189 		btrfs_abort_transaction(trans, ret);
5190 		goto out;
5191 	}
5192 
5193 out:
5194 	up_write(&fs_info->subvol_sem);
5195 	mnt_drop_write_file(file);
5196 	return ret;
5197 }
5198 
5199 #ifdef CONFIG_64BIT
5200 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5201 						void __user *arg)
5202 {
5203 	struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5204 	struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5205 	int ret = 0;
5206 
5207 	args32 = memdup_user(arg, sizeof(*args32));
5208 	if (IS_ERR(args32))
5209 		return PTR_ERR(args32);
5210 
5211 	args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
5212 	if (!args64) {
5213 		ret = -ENOMEM;
5214 		goto out;
5215 	}
5216 
5217 	memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5218 	args64->stransid = args32->stransid;
5219 	args64->rtransid = args32->rtransid;
5220 	args64->stime.sec = args32->stime.sec;
5221 	args64->stime.nsec = args32->stime.nsec;
5222 	args64->rtime.sec = args32->rtime.sec;
5223 	args64->rtime.nsec = args32->rtime.nsec;
5224 	args64->flags = args32->flags;
5225 
5226 	ret = _btrfs_ioctl_set_received_subvol(file, args64);
5227 	if (ret)
5228 		goto out;
5229 
5230 	memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5231 	args32->stransid = args64->stransid;
5232 	args32->rtransid = args64->rtransid;
5233 	args32->stime.sec = args64->stime.sec;
5234 	args32->stime.nsec = args64->stime.nsec;
5235 	args32->rtime.sec = args64->rtime.sec;
5236 	args32->rtime.nsec = args64->rtime.nsec;
5237 	args32->flags = args64->flags;
5238 
5239 	ret = copy_to_user(arg, args32, sizeof(*args32));
5240 	if (ret)
5241 		ret = -EFAULT;
5242 
5243 out:
5244 	kfree(args32);
5245 	kfree(args64);
5246 	return ret;
5247 }
5248 #endif
5249 
5250 static long btrfs_ioctl_set_received_subvol(struct file *file,
5251 					    void __user *arg)
5252 {
5253 	struct btrfs_ioctl_received_subvol_args *sa = NULL;
5254 	int ret = 0;
5255 
5256 	sa = memdup_user(arg, sizeof(*sa));
5257 	if (IS_ERR(sa))
5258 		return PTR_ERR(sa);
5259 
5260 	ret = _btrfs_ioctl_set_received_subvol(file, sa);
5261 
5262 	if (ret)
5263 		goto out;
5264 
5265 	ret = copy_to_user(arg, sa, sizeof(*sa));
5266 	if (ret)
5267 		ret = -EFAULT;
5268 
5269 out:
5270 	kfree(sa);
5271 	return ret;
5272 }
5273 
5274 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5275 {
5276 	struct inode *inode = file_inode(file);
5277 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5278 	size_t len;
5279 	int ret;
5280 	char label[BTRFS_LABEL_SIZE];
5281 
5282 	spin_lock(&fs_info->super_lock);
5283 	memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5284 	spin_unlock(&fs_info->super_lock);
5285 
5286 	len = strnlen(label, BTRFS_LABEL_SIZE);
5287 
5288 	if (len == BTRFS_LABEL_SIZE) {
5289 		btrfs_warn(fs_info,
5290 			   "label is too long, return the first %zu bytes",
5291 			   --len);
5292 	}
5293 
5294 	ret = copy_to_user(arg, label, len);
5295 
5296 	return ret ? -EFAULT : 0;
5297 }
5298 
5299 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5300 {
5301 	struct inode *inode = file_inode(file);
5302 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5303 	struct btrfs_root *root = BTRFS_I(inode)->root;
5304 	struct btrfs_super_block *super_block = fs_info->super_copy;
5305 	struct btrfs_trans_handle *trans;
5306 	char label[BTRFS_LABEL_SIZE];
5307 	int ret;
5308 
5309 	if (!capable(CAP_SYS_ADMIN))
5310 		return -EPERM;
5311 
5312 	if (copy_from_user(label, arg, sizeof(label)))
5313 		return -EFAULT;
5314 
5315 	if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5316 		btrfs_err(fs_info,
5317 			  "unable to set label with more than %d bytes",
5318 			  BTRFS_LABEL_SIZE - 1);
5319 		return -EINVAL;
5320 	}
5321 
5322 	ret = mnt_want_write_file(file);
5323 	if (ret)
5324 		return ret;
5325 
5326 	trans = btrfs_start_transaction(root, 0);
5327 	if (IS_ERR(trans)) {
5328 		ret = PTR_ERR(trans);
5329 		goto out_unlock;
5330 	}
5331 
5332 	spin_lock(&fs_info->super_lock);
5333 	strcpy(super_block->label, label);
5334 	spin_unlock(&fs_info->super_lock);
5335 	ret = btrfs_commit_transaction(trans);
5336 
5337 out_unlock:
5338 	mnt_drop_write_file(file);
5339 	return ret;
5340 }
5341 
5342 #define INIT_FEATURE_FLAGS(suffix) \
5343 	{ .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5344 	  .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5345 	  .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5346 
5347 int btrfs_ioctl_get_supported_features(void __user *arg)
5348 {
5349 	static const struct btrfs_ioctl_feature_flags features[3] = {
5350 		INIT_FEATURE_FLAGS(SUPP),
5351 		INIT_FEATURE_FLAGS(SAFE_SET),
5352 		INIT_FEATURE_FLAGS(SAFE_CLEAR)
5353 	};
5354 
5355 	if (copy_to_user(arg, &features, sizeof(features)))
5356 		return -EFAULT;
5357 
5358 	return 0;
5359 }
5360 
5361 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5362 {
5363 	struct inode *inode = file_inode(file);
5364 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5365 	struct btrfs_super_block *super_block = fs_info->super_copy;
5366 	struct btrfs_ioctl_feature_flags features;
5367 
5368 	features.compat_flags = btrfs_super_compat_flags(super_block);
5369 	features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5370 	features.incompat_flags = btrfs_super_incompat_flags(super_block);
5371 
5372 	if (copy_to_user(arg, &features, sizeof(features)))
5373 		return -EFAULT;
5374 
5375 	return 0;
5376 }
5377 
5378 static int check_feature_bits(struct btrfs_fs_info *fs_info,
5379 			      enum btrfs_feature_set set,
5380 			      u64 change_mask, u64 flags, u64 supported_flags,
5381 			      u64 safe_set, u64 safe_clear)
5382 {
5383 	const char *type = btrfs_feature_set_names[set];
5384 	char *names;
5385 	u64 disallowed, unsupported;
5386 	u64 set_mask = flags & change_mask;
5387 	u64 clear_mask = ~flags & change_mask;
5388 
5389 	unsupported = set_mask & ~supported_flags;
5390 	if (unsupported) {
5391 		names = btrfs_printable_features(set, unsupported);
5392 		if (names) {
5393 			btrfs_warn(fs_info,
5394 				   "this kernel does not support the %s feature bit%s",
5395 				   names, strchr(names, ',') ? "s" : "");
5396 			kfree(names);
5397 		} else
5398 			btrfs_warn(fs_info,
5399 				   "this kernel does not support %s bits 0x%llx",
5400 				   type, unsupported);
5401 		return -EOPNOTSUPP;
5402 	}
5403 
5404 	disallowed = set_mask & ~safe_set;
5405 	if (disallowed) {
5406 		names = btrfs_printable_features(set, disallowed);
5407 		if (names) {
5408 			btrfs_warn(fs_info,
5409 				   "can't set the %s feature bit%s while mounted",
5410 				   names, strchr(names, ',') ? "s" : "");
5411 			kfree(names);
5412 		} else
5413 			btrfs_warn(fs_info,
5414 				   "can't set %s bits 0x%llx while mounted",
5415 				   type, disallowed);
5416 		return -EPERM;
5417 	}
5418 
5419 	disallowed = clear_mask & ~safe_clear;
5420 	if (disallowed) {
5421 		names = btrfs_printable_features(set, disallowed);
5422 		if (names) {
5423 			btrfs_warn(fs_info,
5424 				   "can't clear the %s feature bit%s while mounted",
5425 				   names, strchr(names, ',') ? "s" : "");
5426 			kfree(names);
5427 		} else
5428 			btrfs_warn(fs_info,
5429 				   "can't clear %s bits 0x%llx while mounted",
5430 				   type, disallowed);
5431 		return -EPERM;
5432 	}
5433 
5434 	return 0;
5435 }
5436 
5437 #define check_feature(fs_info, change_mask, flags, mask_base)	\
5438 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,	\
5439 		   BTRFS_FEATURE_ ## mask_base ## _SUPP,	\
5440 		   BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,	\
5441 		   BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5442 
5443 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5444 {
5445 	struct inode *inode = file_inode(file);
5446 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5447 	struct btrfs_root *root = BTRFS_I(inode)->root;
5448 	struct btrfs_super_block *super_block = fs_info->super_copy;
5449 	struct btrfs_ioctl_feature_flags flags[2];
5450 	struct btrfs_trans_handle *trans;
5451 	u64 newflags;
5452 	int ret;
5453 
5454 	if (!capable(CAP_SYS_ADMIN))
5455 		return -EPERM;
5456 
5457 	if (copy_from_user(flags, arg, sizeof(flags)))
5458 		return -EFAULT;
5459 
5460 	/* Nothing to do */
5461 	if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5462 	    !flags[0].incompat_flags)
5463 		return 0;
5464 
5465 	ret = check_feature(fs_info, flags[0].compat_flags,
5466 			    flags[1].compat_flags, COMPAT);
5467 	if (ret)
5468 		return ret;
5469 
5470 	ret = check_feature(fs_info, flags[0].compat_ro_flags,
5471 			    flags[1].compat_ro_flags, COMPAT_RO);
5472 	if (ret)
5473 		return ret;
5474 
5475 	ret = check_feature(fs_info, flags[0].incompat_flags,
5476 			    flags[1].incompat_flags, INCOMPAT);
5477 	if (ret)
5478 		return ret;
5479 
5480 	ret = mnt_want_write_file(file);
5481 	if (ret)
5482 		return ret;
5483 
5484 	trans = btrfs_start_transaction(root, 0);
5485 	if (IS_ERR(trans)) {
5486 		ret = PTR_ERR(trans);
5487 		goto out_drop_write;
5488 	}
5489 
5490 	spin_lock(&fs_info->super_lock);
5491 	newflags = btrfs_super_compat_flags(super_block);
5492 	newflags |= flags[0].compat_flags & flags[1].compat_flags;
5493 	newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5494 	btrfs_set_super_compat_flags(super_block, newflags);
5495 
5496 	newflags = btrfs_super_compat_ro_flags(super_block);
5497 	newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5498 	newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5499 	btrfs_set_super_compat_ro_flags(super_block, newflags);
5500 
5501 	newflags = btrfs_super_incompat_flags(super_block);
5502 	newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5503 	newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5504 	btrfs_set_super_incompat_flags(super_block, newflags);
5505 	spin_unlock(&fs_info->super_lock);
5506 
5507 	ret = btrfs_commit_transaction(trans);
5508 out_drop_write:
5509 	mnt_drop_write_file(file);
5510 
5511 	return ret;
5512 }
5513 
5514 long btrfs_ioctl(struct file *file, unsigned int
5515 		cmd, unsigned long arg)
5516 {
5517 	struct inode *inode = file_inode(file);
5518 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5519 	struct btrfs_root *root = BTRFS_I(inode)->root;
5520 	void __user *argp = (void __user *)arg;
5521 
5522 	switch (cmd) {
5523 	case FS_IOC_GETFLAGS:
5524 		return btrfs_ioctl_getflags(file, argp);
5525 	case FS_IOC_SETFLAGS:
5526 		return btrfs_ioctl_setflags(file, argp);
5527 	case FS_IOC_GETVERSION:
5528 		return btrfs_ioctl_getversion(file, argp);
5529 	case FITRIM:
5530 		return btrfs_ioctl_fitrim(file, argp);
5531 	case BTRFS_IOC_SNAP_CREATE:
5532 		return btrfs_ioctl_snap_create(file, argp, 0);
5533 	case BTRFS_IOC_SNAP_CREATE_V2:
5534 		return btrfs_ioctl_snap_create_v2(file, argp, 0);
5535 	case BTRFS_IOC_SUBVOL_CREATE:
5536 		return btrfs_ioctl_snap_create(file, argp, 1);
5537 	case BTRFS_IOC_SUBVOL_CREATE_V2:
5538 		return btrfs_ioctl_snap_create_v2(file, argp, 1);
5539 	case BTRFS_IOC_SNAP_DESTROY:
5540 		return btrfs_ioctl_snap_destroy(file, argp);
5541 	case BTRFS_IOC_SUBVOL_GETFLAGS:
5542 		return btrfs_ioctl_subvol_getflags(file, argp);
5543 	case BTRFS_IOC_SUBVOL_SETFLAGS:
5544 		return btrfs_ioctl_subvol_setflags(file, argp);
5545 	case BTRFS_IOC_DEFAULT_SUBVOL:
5546 		return btrfs_ioctl_default_subvol(file, argp);
5547 	case BTRFS_IOC_DEFRAG:
5548 		return btrfs_ioctl_defrag(file, NULL);
5549 	case BTRFS_IOC_DEFRAG_RANGE:
5550 		return btrfs_ioctl_defrag(file, argp);
5551 	case BTRFS_IOC_RESIZE:
5552 		return btrfs_ioctl_resize(file, argp);
5553 	case BTRFS_IOC_ADD_DEV:
5554 		return btrfs_ioctl_add_dev(fs_info, argp);
5555 	case BTRFS_IOC_RM_DEV:
5556 		return btrfs_ioctl_rm_dev(file, argp);
5557 	case BTRFS_IOC_RM_DEV_V2:
5558 		return btrfs_ioctl_rm_dev_v2(file, argp);
5559 	case BTRFS_IOC_FS_INFO:
5560 		return btrfs_ioctl_fs_info(fs_info, argp);
5561 	case BTRFS_IOC_DEV_INFO:
5562 		return btrfs_ioctl_dev_info(fs_info, argp);
5563 	case BTRFS_IOC_BALANCE:
5564 		return btrfs_ioctl_balance(file, NULL);
5565 	case BTRFS_IOC_TRANS_START:
5566 		return btrfs_ioctl_trans_start(file);
5567 	case BTRFS_IOC_TRANS_END:
5568 		return btrfs_ioctl_trans_end(file);
5569 	case BTRFS_IOC_TREE_SEARCH:
5570 		return btrfs_ioctl_tree_search(file, argp);
5571 	case BTRFS_IOC_TREE_SEARCH_V2:
5572 		return btrfs_ioctl_tree_search_v2(file, argp);
5573 	case BTRFS_IOC_INO_LOOKUP:
5574 		return btrfs_ioctl_ino_lookup(file, argp);
5575 	case BTRFS_IOC_INO_PATHS:
5576 		return btrfs_ioctl_ino_to_path(root, argp);
5577 	case BTRFS_IOC_LOGICAL_INO:
5578 		return btrfs_ioctl_logical_to_ino(fs_info, argp);
5579 	case BTRFS_IOC_SPACE_INFO:
5580 		return btrfs_ioctl_space_info(fs_info, argp);
5581 	case BTRFS_IOC_SYNC: {
5582 		int ret;
5583 
5584 		ret = btrfs_start_delalloc_roots(fs_info, 0, -1);
5585 		if (ret)
5586 			return ret;
5587 		ret = btrfs_sync_fs(inode->i_sb, 1);
5588 		/*
5589 		 * The transaction thread may want to do more work,
5590 		 * namely it pokes the cleaner kthread that will start
5591 		 * processing uncleaned subvols.
5592 		 */
5593 		wake_up_process(fs_info->transaction_kthread);
5594 		return ret;
5595 	}
5596 	case BTRFS_IOC_START_SYNC:
5597 		return btrfs_ioctl_start_sync(root, argp);
5598 	case BTRFS_IOC_WAIT_SYNC:
5599 		return btrfs_ioctl_wait_sync(fs_info, argp);
5600 	case BTRFS_IOC_SCRUB:
5601 		return btrfs_ioctl_scrub(file, argp);
5602 	case BTRFS_IOC_SCRUB_CANCEL:
5603 		return btrfs_ioctl_scrub_cancel(fs_info);
5604 	case BTRFS_IOC_SCRUB_PROGRESS:
5605 		return btrfs_ioctl_scrub_progress(fs_info, argp);
5606 	case BTRFS_IOC_BALANCE_V2:
5607 		return btrfs_ioctl_balance(file, argp);
5608 	case BTRFS_IOC_BALANCE_CTL:
5609 		return btrfs_ioctl_balance_ctl(fs_info, arg);
5610 	case BTRFS_IOC_BALANCE_PROGRESS:
5611 		return btrfs_ioctl_balance_progress(fs_info, argp);
5612 	case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5613 		return btrfs_ioctl_set_received_subvol(file, argp);
5614 #ifdef CONFIG_64BIT
5615 	case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5616 		return btrfs_ioctl_set_received_subvol_32(file, argp);
5617 #endif
5618 	case BTRFS_IOC_SEND:
5619 		return btrfs_ioctl_send(file, argp);
5620 	case BTRFS_IOC_GET_DEV_STATS:
5621 		return btrfs_ioctl_get_dev_stats(fs_info, argp);
5622 	case BTRFS_IOC_QUOTA_CTL:
5623 		return btrfs_ioctl_quota_ctl(file, argp);
5624 	case BTRFS_IOC_QGROUP_ASSIGN:
5625 		return btrfs_ioctl_qgroup_assign(file, argp);
5626 	case BTRFS_IOC_QGROUP_CREATE:
5627 		return btrfs_ioctl_qgroup_create(file, argp);
5628 	case BTRFS_IOC_QGROUP_LIMIT:
5629 		return btrfs_ioctl_qgroup_limit(file, argp);
5630 	case BTRFS_IOC_QUOTA_RESCAN:
5631 		return btrfs_ioctl_quota_rescan(file, argp);
5632 	case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5633 		return btrfs_ioctl_quota_rescan_status(file, argp);
5634 	case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5635 		return btrfs_ioctl_quota_rescan_wait(file, argp);
5636 	case BTRFS_IOC_DEV_REPLACE:
5637 		return btrfs_ioctl_dev_replace(fs_info, argp);
5638 	case BTRFS_IOC_GET_FSLABEL:
5639 		return btrfs_ioctl_get_fslabel(file, argp);
5640 	case BTRFS_IOC_SET_FSLABEL:
5641 		return btrfs_ioctl_set_fslabel(file, argp);
5642 	case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5643 		return btrfs_ioctl_get_supported_features(argp);
5644 	case BTRFS_IOC_GET_FEATURES:
5645 		return btrfs_ioctl_get_features(file, argp);
5646 	case BTRFS_IOC_SET_FEATURES:
5647 		return btrfs_ioctl_set_features(file, argp);
5648 	}
5649 
5650 	return -ENOTTY;
5651 }
5652 
5653 #ifdef CONFIG_COMPAT
5654 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5655 {
5656 	switch (cmd) {
5657 	case FS_IOC32_GETFLAGS:
5658 		cmd = FS_IOC_GETFLAGS;
5659 		break;
5660 	case FS_IOC32_SETFLAGS:
5661 		cmd = FS_IOC_SETFLAGS;
5662 		break;
5663 	case FS_IOC32_GETVERSION:
5664 		cmd = FS_IOC_GETVERSION;
5665 		break;
5666 	default:
5667 		return -ENOIOCTLCMD;
5668 	}
5669 
5670 	return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5671 }
5672 #endif
5673