xref: /openbmc/linux/fs/btrfs/ioctl.c (revision 68198dca)
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/fsnotify.h>
25 #include <linux/pagemap.h>
26 #include <linux/highmem.h>
27 #include <linux/time.h>
28 #include <linux/init.h>
29 #include <linux/string.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mount.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/swap.h>
35 #include <linux/writeback.h>
36 #include <linux/compat.h>
37 #include <linux/bit_spinlock.h>
38 #include <linux/security.h>
39 #include <linux/xattr.h>
40 #include <linux/mm.h>
41 #include <linux/slab.h>
42 #include <linux/blkdev.h>
43 #include <linux/uuid.h>
44 #include <linux/btrfs.h>
45 #include <linux/uaccess.h>
46 #include "ctree.h"
47 #include "disk-io.h"
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "volumes.h"
52 #include "locking.h"
53 #include "inode-map.h"
54 #include "backref.h"
55 #include "rcu-string.h"
56 #include "send.h"
57 #include "dev-replace.h"
58 #include "props.h"
59 #include "sysfs.h"
60 #include "qgroup.h"
61 #include "tree-log.h"
62 #include "compression.h"
63 
64 #ifdef CONFIG_64BIT
65 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
66  * structures are incorrect, as the timespec structure from userspace
67  * is 4 bytes too small. We define these alternatives here to teach
68  * the kernel about the 32-bit struct packing.
69  */
70 struct btrfs_ioctl_timespec_32 {
71 	__u64 sec;
72 	__u32 nsec;
73 } __attribute__ ((__packed__));
74 
75 struct btrfs_ioctl_received_subvol_args_32 {
76 	char	uuid[BTRFS_UUID_SIZE];	/* in */
77 	__u64	stransid;		/* in */
78 	__u64	rtransid;		/* out */
79 	struct btrfs_ioctl_timespec_32 stime; /* in */
80 	struct btrfs_ioctl_timespec_32 rtime; /* out */
81 	__u64	flags;			/* in */
82 	__u64	reserved[16];		/* in */
83 } __attribute__ ((__packed__));
84 
85 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
86 				struct btrfs_ioctl_received_subvol_args_32)
87 #endif
88 
89 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
90 struct btrfs_ioctl_send_args_32 {
91 	__s64 send_fd;			/* in */
92 	__u64 clone_sources_count;	/* in */
93 	compat_uptr_t clone_sources;	/* in */
94 	__u64 parent_root;		/* in */
95 	__u64 flags;			/* in */
96 	__u64 reserved[4];		/* in */
97 } __attribute__ ((__packed__));
98 
99 #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \
100 			       struct btrfs_ioctl_send_args_32)
101 #endif
102 
103 static int btrfs_clone(struct inode *src, struct inode *inode,
104 		       u64 off, u64 olen, u64 olen_aligned, u64 destoff,
105 		       int no_time_update);
106 
107 /* Mask out flags that are inappropriate for the given type of inode. */
108 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
109 {
110 	if (S_ISDIR(mode))
111 		return flags;
112 	else if (S_ISREG(mode))
113 		return flags & ~FS_DIRSYNC_FL;
114 	else
115 		return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
116 }
117 
118 /*
119  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
120  */
121 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
122 {
123 	unsigned int iflags = 0;
124 
125 	if (flags & BTRFS_INODE_SYNC)
126 		iflags |= FS_SYNC_FL;
127 	if (flags & BTRFS_INODE_IMMUTABLE)
128 		iflags |= FS_IMMUTABLE_FL;
129 	if (flags & BTRFS_INODE_APPEND)
130 		iflags |= FS_APPEND_FL;
131 	if (flags & BTRFS_INODE_NODUMP)
132 		iflags |= FS_NODUMP_FL;
133 	if (flags & BTRFS_INODE_NOATIME)
134 		iflags |= FS_NOATIME_FL;
135 	if (flags & BTRFS_INODE_DIRSYNC)
136 		iflags |= FS_DIRSYNC_FL;
137 	if (flags & BTRFS_INODE_NODATACOW)
138 		iflags |= FS_NOCOW_FL;
139 
140 	if (flags & BTRFS_INODE_NOCOMPRESS)
141 		iflags |= FS_NOCOMP_FL;
142 	else if (flags & BTRFS_INODE_COMPRESS)
143 		iflags |= FS_COMPR_FL;
144 
145 	return iflags;
146 }
147 
148 /*
149  * Update inode->i_flags based on the btrfs internal flags.
150  */
151 void btrfs_update_iflags(struct inode *inode)
152 {
153 	struct btrfs_inode *ip = BTRFS_I(inode);
154 	unsigned int new_fl = 0;
155 
156 	if (ip->flags & BTRFS_INODE_SYNC)
157 		new_fl |= S_SYNC;
158 	if (ip->flags & BTRFS_INODE_IMMUTABLE)
159 		new_fl |= S_IMMUTABLE;
160 	if (ip->flags & BTRFS_INODE_APPEND)
161 		new_fl |= S_APPEND;
162 	if (ip->flags & BTRFS_INODE_NOATIME)
163 		new_fl |= S_NOATIME;
164 	if (ip->flags & BTRFS_INODE_DIRSYNC)
165 		new_fl |= S_DIRSYNC;
166 
167 	set_mask_bits(&inode->i_flags,
168 		      S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
169 		      new_fl);
170 }
171 
172 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
173 {
174 	struct btrfs_inode *ip = BTRFS_I(file_inode(file));
175 	unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
176 
177 	if (copy_to_user(arg, &flags, sizeof(flags)))
178 		return -EFAULT;
179 	return 0;
180 }
181 
182 static int check_flags(unsigned int flags)
183 {
184 	if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
185 		      FS_NOATIME_FL | FS_NODUMP_FL | \
186 		      FS_SYNC_FL | FS_DIRSYNC_FL | \
187 		      FS_NOCOMP_FL | FS_COMPR_FL |
188 		      FS_NOCOW_FL))
189 		return -EOPNOTSUPP;
190 
191 	if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
192 		return -EINVAL;
193 
194 	return 0;
195 }
196 
197 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
198 {
199 	struct inode *inode = file_inode(file);
200 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
201 	struct btrfs_inode *ip = BTRFS_I(inode);
202 	struct btrfs_root *root = ip->root;
203 	struct btrfs_trans_handle *trans;
204 	unsigned int flags, oldflags;
205 	int ret;
206 	u64 ip_oldflags;
207 	unsigned int i_oldflags;
208 	umode_t mode;
209 
210 	if (!inode_owner_or_capable(inode))
211 		return -EPERM;
212 
213 	if (btrfs_root_readonly(root))
214 		return -EROFS;
215 
216 	if (copy_from_user(&flags, arg, sizeof(flags)))
217 		return -EFAULT;
218 
219 	ret = check_flags(flags);
220 	if (ret)
221 		return ret;
222 
223 	ret = mnt_want_write_file(file);
224 	if (ret)
225 		return ret;
226 
227 	inode_lock(inode);
228 
229 	ip_oldflags = ip->flags;
230 	i_oldflags = inode->i_flags;
231 	mode = inode->i_mode;
232 
233 	flags = btrfs_mask_flags(inode->i_mode, flags);
234 	oldflags = btrfs_flags_to_ioctl(ip->flags);
235 	if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
236 		if (!capable(CAP_LINUX_IMMUTABLE)) {
237 			ret = -EPERM;
238 			goto out_unlock;
239 		}
240 	}
241 
242 	if (flags & FS_SYNC_FL)
243 		ip->flags |= BTRFS_INODE_SYNC;
244 	else
245 		ip->flags &= ~BTRFS_INODE_SYNC;
246 	if (flags & FS_IMMUTABLE_FL)
247 		ip->flags |= BTRFS_INODE_IMMUTABLE;
248 	else
249 		ip->flags &= ~BTRFS_INODE_IMMUTABLE;
250 	if (flags & FS_APPEND_FL)
251 		ip->flags |= BTRFS_INODE_APPEND;
252 	else
253 		ip->flags &= ~BTRFS_INODE_APPEND;
254 	if (flags & FS_NODUMP_FL)
255 		ip->flags |= BTRFS_INODE_NODUMP;
256 	else
257 		ip->flags &= ~BTRFS_INODE_NODUMP;
258 	if (flags & FS_NOATIME_FL)
259 		ip->flags |= BTRFS_INODE_NOATIME;
260 	else
261 		ip->flags &= ~BTRFS_INODE_NOATIME;
262 	if (flags & FS_DIRSYNC_FL)
263 		ip->flags |= BTRFS_INODE_DIRSYNC;
264 	else
265 		ip->flags &= ~BTRFS_INODE_DIRSYNC;
266 	if (flags & FS_NOCOW_FL) {
267 		if (S_ISREG(mode)) {
268 			/*
269 			 * It's safe to turn csums off here, no extents exist.
270 			 * Otherwise we want the flag to reflect the real COW
271 			 * status of the file and will not set it.
272 			 */
273 			if (inode->i_size == 0)
274 				ip->flags |= BTRFS_INODE_NODATACOW
275 					   | BTRFS_INODE_NODATASUM;
276 		} else {
277 			ip->flags |= BTRFS_INODE_NODATACOW;
278 		}
279 	} else {
280 		/*
281 		 * Revert back under same assumptions as above
282 		 */
283 		if (S_ISREG(mode)) {
284 			if (inode->i_size == 0)
285 				ip->flags &= ~(BTRFS_INODE_NODATACOW
286 				             | BTRFS_INODE_NODATASUM);
287 		} else {
288 			ip->flags &= ~BTRFS_INODE_NODATACOW;
289 		}
290 	}
291 
292 	/*
293 	 * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
294 	 * flag may be changed automatically if compression code won't make
295 	 * things smaller.
296 	 */
297 	if (flags & FS_NOCOMP_FL) {
298 		ip->flags &= ~BTRFS_INODE_COMPRESS;
299 		ip->flags |= BTRFS_INODE_NOCOMPRESS;
300 
301 		ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
302 		if (ret && ret != -ENODATA)
303 			goto out_drop;
304 	} else if (flags & FS_COMPR_FL) {
305 		const char *comp;
306 
307 		ip->flags |= BTRFS_INODE_COMPRESS;
308 		ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
309 
310 		if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
311 			comp = "lzo";
312 		else if (fs_info->compress_type == BTRFS_COMPRESS_ZLIB)
313 			comp = "zlib";
314 		else
315 			comp = "zstd";
316 		ret = btrfs_set_prop(inode, "btrfs.compression",
317 				     comp, strlen(comp), 0);
318 		if (ret)
319 			goto out_drop;
320 
321 	} else {
322 		ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
323 		if (ret && ret != -ENODATA)
324 			goto out_drop;
325 		ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
326 	}
327 
328 	trans = btrfs_start_transaction(root, 1);
329 	if (IS_ERR(trans)) {
330 		ret = PTR_ERR(trans);
331 		goto out_drop;
332 	}
333 
334 	btrfs_update_iflags(inode);
335 	inode_inc_iversion(inode);
336 	inode->i_ctime = current_time(inode);
337 	ret = btrfs_update_inode(trans, root, inode);
338 
339 	btrfs_end_transaction(trans);
340  out_drop:
341 	if (ret) {
342 		ip->flags = ip_oldflags;
343 		inode->i_flags = i_oldflags;
344 	}
345 
346  out_unlock:
347 	inode_unlock(inode);
348 	mnt_drop_write_file(file);
349 	return ret;
350 }
351 
352 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
353 {
354 	struct inode *inode = file_inode(file);
355 
356 	return put_user(inode->i_generation, arg);
357 }
358 
359 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
360 {
361 	struct inode *inode = file_inode(file);
362 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
363 	struct btrfs_device *device;
364 	struct request_queue *q;
365 	struct fstrim_range range;
366 	u64 minlen = ULLONG_MAX;
367 	u64 num_devices = 0;
368 	u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
369 	int ret;
370 
371 	if (!capable(CAP_SYS_ADMIN))
372 		return -EPERM;
373 
374 	rcu_read_lock();
375 	list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
376 				dev_list) {
377 		if (!device->bdev)
378 			continue;
379 		q = bdev_get_queue(device->bdev);
380 		if (blk_queue_discard(q)) {
381 			num_devices++;
382 			minlen = min_t(u64, q->limits.discard_granularity,
383 				     minlen);
384 		}
385 	}
386 	rcu_read_unlock();
387 
388 	if (!num_devices)
389 		return -EOPNOTSUPP;
390 	if (copy_from_user(&range, arg, sizeof(range)))
391 		return -EFAULT;
392 	if (range.start > total_bytes ||
393 	    range.len < fs_info->sb->s_blocksize)
394 		return -EINVAL;
395 
396 	range.len = min(range.len, total_bytes - range.start);
397 	range.minlen = max(range.minlen, minlen);
398 	ret = btrfs_trim_fs(fs_info, &range);
399 	if (ret < 0)
400 		return ret;
401 
402 	if (copy_to_user(arg, &range, sizeof(range)))
403 		return -EFAULT;
404 
405 	return 0;
406 }
407 
408 int btrfs_is_empty_uuid(u8 *uuid)
409 {
410 	int i;
411 
412 	for (i = 0; i < BTRFS_UUID_SIZE; i++) {
413 		if (uuid[i])
414 			return 0;
415 	}
416 	return 1;
417 }
418 
419 static noinline int create_subvol(struct inode *dir,
420 				  struct dentry *dentry,
421 				  const char *name, int namelen,
422 				  u64 *async_transid,
423 				  struct btrfs_qgroup_inherit *inherit)
424 {
425 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
426 	struct btrfs_trans_handle *trans;
427 	struct btrfs_key key;
428 	struct btrfs_root_item *root_item;
429 	struct btrfs_inode_item *inode_item;
430 	struct extent_buffer *leaf;
431 	struct btrfs_root *root = BTRFS_I(dir)->root;
432 	struct btrfs_root *new_root;
433 	struct btrfs_block_rsv block_rsv;
434 	struct timespec cur_time = current_time(dir);
435 	struct inode *inode;
436 	int ret;
437 	int err;
438 	u64 objectid;
439 	u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
440 	u64 index = 0;
441 	u64 qgroup_reserved;
442 	uuid_le new_uuid;
443 
444 	root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
445 	if (!root_item)
446 		return -ENOMEM;
447 
448 	ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid);
449 	if (ret)
450 		goto fail_free;
451 
452 	/*
453 	 * Don't create subvolume whose level is not zero. Or qgroup will be
454 	 * screwed up since it assumes subvolume qgroup's level to be 0.
455 	 */
456 	if (btrfs_qgroup_level(objectid)) {
457 		ret = -ENOSPC;
458 		goto fail_free;
459 	}
460 
461 	btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
462 	/*
463 	 * The same as the snapshot creation, please see the comment
464 	 * of create_snapshot().
465 	 */
466 	ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
467 					       8, &qgroup_reserved, false);
468 	if (ret)
469 		goto fail_free;
470 
471 	trans = btrfs_start_transaction(root, 0);
472 	if (IS_ERR(trans)) {
473 		ret = PTR_ERR(trans);
474 		btrfs_subvolume_release_metadata(fs_info, &block_rsv);
475 		goto fail_free;
476 	}
477 	trans->block_rsv = &block_rsv;
478 	trans->bytes_reserved = block_rsv.size;
479 
480 	ret = btrfs_qgroup_inherit(trans, fs_info, 0, objectid, inherit);
481 	if (ret)
482 		goto fail;
483 
484 	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
485 	if (IS_ERR(leaf)) {
486 		ret = PTR_ERR(leaf);
487 		goto fail;
488 	}
489 
490 	memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
491 	btrfs_set_header_bytenr(leaf, leaf->start);
492 	btrfs_set_header_generation(leaf, trans->transid);
493 	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
494 	btrfs_set_header_owner(leaf, objectid);
495 
496 	write_extent_buffer_fsid(leaf, fs_info->fsid);
497 	write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
498 	btrfs_mark_buffer_dirty(leaf);
499 
500 	inode_item = &root_item->inode;
501 	btrfs_set_stack_inode_generation(inode_item, 1);
502 	btrfs_set_stack_inode_size(inode_item, 3);
503 	btrfs_set_stack_inode_nlink(inode_item, 1);
504 	btrfs_set_stack_inode_nbytes(inode_item,
505 				     fs_info->nodesize);
506 	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
507 
508 	btrfs_set_root_flags(root_item, 0);
509 	btrfs_set_root_limit(root_item, 0);
510 	btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
511 
512 	btrfs_set_root_bytenr(root_item, leaf->start);
513 	btrfs_set_root_generation(root_item, trans->transid);
514 	btrfs_set_root_level(root_item, 0);
515 	btrfs_set_root_refs(root_item, 1);
516 	btrfs_set_root_used(root_item, leaf->len);
517 	btrfs_set_root_last_snapshot(root_item, 0);
518 
519 	btrfs_set_root_generation_v2(root_item,
520 			btrfs_root_generation(root_item));
521 	uuid_le_gen(&new_uuid);
522 	memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
523 	btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
524 	btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
525 	root_item->ctime = root_item->otime;
526 	btrfs_set_root_ctransid(root_item, trans->transid);
527 	btrfs_set_root_otransid(root_item, trans->transid);
528 
529 	btrfs_tree_unlock(leaf);
530 	free_extent_buffer(leaf);
531 	leaf = NULL;
532 
533 	btrfs_set_root_dirid(root_item, new_dirid);
534 
535 	key.objectid = objectid;
536 	key.offset = 0;
537 	key.type = BTRFS_ROOT_ITEM_KEY;
538 	ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
539 				root_item);
540 	if (ret)
541 		goto fail;
542 
543 	key.offset = (u64)-1;
544 	new_root = btrfs_read_fs_root_no_name(fs_info, &key);
545 	if (IS_ERR(new_root)) {
546 		ret = PTR_ERR(new_root);
547 		btrfs_abort_transaction(trans, ret);
548 		goto fail;
549 	}
550 
551 	btrfs_record_root_in_trans(trans, new_root);
552 
553 	ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
554 	if (ret) {
555 		/* We potentially lose an unused inode item here */
556 		btrfs_abort_transaction(trans, ret);
557 		goto fail;
558 	}
559 
560 	mutex_lock(&new_root->objectid_mutex);
561 	new_root->highest_objectid = new_dirid;
562 	mutex_unlock(&new_root->objectid_mutex);
563 
564 	/*
565 	 * insert the directory item
566 	 */
567 	ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
568 	if (ret) {
569 		btrfs_abort_transaction(trans, ret);
570 		goto fail;
571 	}
572 
573 	ret = btrfs_insert_dir_item(trans, root,
574 				    name, namelen, BTRFS_I(dir), &key,
575 				    BTRFS_FT_DIR, index);
576 	if (ret) {
577 		btrfs_abort_transaction(trans, ret);
578 		goto fail;
579 	}
580 
581 	btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
582 	ret = btrfs_update_inode(trans, root, dir);
583 	BUG_ON(ret);
584 
585 	ret = btrfs_add_root_ref(trans, fs_info,
586 				 objectid, root->root_key.objectid,
587 				 btrfs_ino(BTRFS_I(dir)), index, name, namelen);
588 	BUG_ON(ret);
589 
590 	ret = btrfs_uuid_tree_add(trans, fs_info, root_item->uuid,
591 				  BTRFS_UUID_KEY_SUBVOL, objectid);
592 	if (ret)
593 		btrfs_abort_transaction(trans, ret);
594 
595 fail:
596 	kfree(root_item);
597 	trans->block_rsv = NULL;
598 	trans->bytes_reserved = 0;
599 	btrfs_subvolume_release_metadata(fs_info, &block_rsv);
600 
601 	if (async_transid) {
602 		*async_transid = trans->transid;
603 		err = btrfs_commit_transaction_async(trans, 1);
604 		if (err)
605 			err = btrfs_commit_transaction(trans);
606 	} else {
607 		err = btrfs_commit_transaction(trans);
608 	}
609 	if (err && !ret)
610 		ret = err;
611 
612 	if (!ret) {
613 		inode = btrfs_lookup_dentry(dir, dentry);
614 		if (IS_ERR(inode))
615 			return PTR_ERR(inode);
616 		d_instantiate(dentry, inode);
617 	}
618 	return ret;
619 
620 fail_free:
621 	kfree(root_item);
622 	return ret;
623 }
624 
625 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
626 			   struct dentry *dentry,
627 			   u64 *async_transid, bool readonly,
628 			   struct btrfs_qgroup_inherit *inherit)
629 {
630 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
631 	struct inode *inode;
632 	struct btrfs_pending_snapshot *pending_snapshot;
633 	struct btrfs_trans_handle *trans;
634 	int ret;
635 
636 	if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
637 		return -EINVAL;
638 
639 	pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
640 	if (!pending_snapshot)
641 		return -ENOMEM;
642 
643 	pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
644 			GFP_KERNEL);
645 	pending_snapshot->path = btrfs_alloc_path();
646 	if (!pending_snapshot->root_item || !pending_snapshot->path) {
647 		ret = -ENOMEM;
648 		goto free_pending;
649 	}
650 
651 	atomic_inc(&root->will_be_snapshotted);
652 	smp_mb__after_atomic();
653 	/* wait for no snapshot writes */
654 	wait_event(root->subv_writers->wait,
655 		   percpu_counter_sum(&root->subv_writers->counter) == 0);
656 
657 	ret = btrfs_start_delalloc_inodes(root, 0);
658 	if (ret)
659 		goto dec_and_free;
660 
661 	btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
662 
663 	btrfs_init_block_rsv(&pending_snapshot->block_rsv,
664 			     BTRFS_BLOCK_RSV_TEMP);
665 	/*
666 	 * 1 - parent dir inode
667 	 * 2 - dir entries
668 	 * 1 - root item
669 	 * 2 - root ref/backref
670 	 * 1 - root of snapshot
671 	 * 1 - UUID item
672 	 */
673 	ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
674 					&pending_snapshot->block_rsv, 8,
675 					&pending_snapshot->qgroup_reserved,
676 					false);
677 	if (ret)
678 		goto dec_and_free;
679 
680 	pending_snapshot->dentry = dentry;
681 	pending_snapshot->root = root;
682 	pending_snapshot->readonly = readonly;
683 	pending_snapshot->dir = dir;
684 	pending_snapshot->inherit = inherit;
685 
686 	trans = btrfs_start_transaction(root, 0);
687 	if (IS_ERR(trans)) {
688 		ret = PTR_ERR(trans);
689 		goto fail;
690 	}
691 
692 	spin_lock(&fs_info->trans_lock);
693 	list_add(&pending_snapshot->list,
694 		 &trans->transaction->pending_snapshots);
695 	spin_unlock(&fs_info->trans_lock);
696 	if (async_transid) {
697 		*async_transid = trans->transid;
698 		ret = btrfs_commit_transaction_async(trans, 1);
699 		if (ret)
700 			ret = btrfs_commit_transaction(trans);
701 	} else {
702 		ret = btrfs_commit_transaction(trans);
703 	}
704 	if (ret)
705 		goto fail;
706 
707 	ret = pending_snapshot->error;
708 	if (ret)
709 		goto fail;
710 
711 	ret = btrfs_orphan_cleanup(pending_snapshot->snap);
712 	if (ret)
713 		goto fail;
714 
715 	inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
716 	if (IS_ERR(inode)) {
717 		ret = PTR_ERR(inode);
718 		goto fail;
719 	}
720 
721 	d_instantiate(dentry, inode);
722 	ret = 0;
723 fail:
724 	btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
725 dec_and_free:
726 	if (atomic_dec_and_test(&root->will_be_snapshotted))
727 		wake_up_atomic_t(&root->will_be_snapshotted);
728 free_pending:
729 	kfree(pending_snapshot->root_item);
730 	btrfs_free_path(pending_snapshot->path);
731 	kfree(pending_snapshot);
732 
733 	return ret;
734 }
735 
736 /*  copy of may_delete in fs/namei.c()
737  *	Check whether we can remove a link victim from directory dir, check
738  *  whether the type of victim is right.
739  *  1. We can't do it if dir is read-only (done in permission())
740  *  2. We should have write and exec permissions on dir
741  *  3. We can't remove anything from append-only dir
742  *  4. We can't do anything with immutable dir (done in permission())
743  *  5. If the sticky bit on dir is set we should either
744  *	a. be owner of dir, or
745  *	b. be owner of victim, or
746  *	c. have CAP_FOWNER capability
747  *  6. If the victim is append-only or immutable we can't do anything with
748  *     links pointing to it.
749  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
750  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
751  *  9. We can't remove a root or mountpoint.
752  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
753  *     nfs_async_unlink().
754  */
755 
756 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
757 {
758 	int error;
759 
760 	if (d_really_is_negative(victim))
761 		return -ENOENT;
762 
763 	BUG_ON(d_inode(victim->d_parent) != dir);
764 	audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
765 
766 	error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
767 	if (error)
768 		return error;
769 	if (IS_APPEND(dir))
770 		return -EPERM;
771 	if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
772 	    IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
773 		return -EPERM;
774 	if (isdir) {
775 		if (!d_is_dir(victim))
776 			return -ENOTDIR;
777 		if (IS_ROOT(victim))
778 			return -EBUSY;
779 	} else if (d_is_dir(victim))
780 		return -EISDIR;
781 	if (IS_DEADDIR(dir))
782 		return -ENOENT;
783 	if (victim->d_flags & DCACHE_NFSFS_RENAMED)
784 		return -EBUSY;
785 	return 0;
786 }
787 
788 /* copy of may_create in fs/namei.c() */
789 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
790 {
791 	if (d_really_is_positive(child))
792 		return -EEXIST;
793 	if (IS_DEADDIR(dir))
794 		return -ENOENT;
795 	return inode_permission(dir, MAY_WRITE | MAY_EXEC);
796 }
797 
798 /*
799  * Create a new subvolume below @parent.  This is largely modeled after
800  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
801  * inside this filesystem so it's quite a bit simpler.
802  */
803 static noinline int btrfs_mksubvol(const struct path *parent,
804 				   const char *name, int namelen,
805 				   struct btrfs_root *snap_src,
806 				   u64 *async_transid, bool readonly,
807 				   struct btrfs_qgroup_inherit *inherit)
808 {
809 	struct inode *dir = d_inode(parent->dentry);
810 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
811 	struct dentry *dentry;
812 	int error;
813 
814 	error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
815 	if (error == -EINTR)
816 		return error;
817 
818 	dentry = lookup_one_len(name, parent->dentry, namelen);
819 	error = PTR_ERR(dentry);
820 	if (IS_ERR(dentry))
821 		goto out_unlock;
822 
823 	error = btrfs_may_create(dir, dentry);
824 	if (error)
825 		goto out_dput;
826 
827 	/*
828 	 * even if this name doesn't exist, we may get hash collisions.
829 	 * check for them now when we can safely fail
830 	 */
831 	error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
832 					       dir->i_ino, name,
833 					       namelen);
834 	if (error)
835 		goto out_dput;
836 
837 	down_read(&fs_info->subvol_sem);
838 
839 	if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
840 		goto out_up_read;
841 
842 	if (snap_src) {
843 		error = create_snapshot(snap_src, dir, dentry,
844 					async_transid, readonly, inherit);
845 	} else {
846 		error = create_subvol(dir, dentry, name, namelen,
847 				      async_transid, inherit);
848 	}
849 	if (!error)
850 		fsnotify_mkdir(dir, dentry);
851 out_up_read:
852 	up_read(&fs_info->subvol_sem);
853 out_dput:
854 	dput(dentry);
855 out_unlock:
856 	inode_unlock(dir);
857 	return error;
858 }
859 
860 /*
861  * When we're defragging a range, we don't want to kick it off again
862  * if it is really just waiting for delalloc to send it down.
863  * If we find a nice big extent or delalloc range for the bytes in the
864  * file you want to defrag, we return 0 to let you know to skip this
865  * part of the file
866  */
867 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
868 {
869 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
870 	struct extent_map *em = NULL;
871 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
872 	u64 end;
873 
874 	read_lock(&em_tree->lock);
875 	em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
876 	read_unlock(&em_tree->lock);
877 
878 	if (em) {
879 		end = extent_map_end(em);
880 		free_extent_map(em);
881 		if (end - offset > thresh)
882 			return 0;
883 	}
884 	/* if we already have a nice delalloc here, just stop */
885 	thresh /= 2;
886 	end = count_range_bits(io_tree, &offset, offset + thresh,
887 			       thresh, EXTENT_DELALLOC, 1);
888 	if (end >= thresh)
889 		return 0;
890 	return 1;
891 }
892 
893 /*
894  * helper function to walk through a file and find extents
895  * newer than a specific transid, and smaller than thresh.
896  *
897  * This is used by the defragging code to find new and small
898  * extents
899  */
900 static int find_new_extents(struct btrfs_root *root,
901 			    struct inode *inode, u64 newer_than,
902 			    u64 *off, u32 thresh)
903 {
904 	struct btrfs_path *path;
905 	struct btrfs_key min_key;
906 	struct extent_buffer *leaf;
907 	struct btrfs_file_extent_item *extent;
908 	int type;
909 	int ret;
910 	u64 ino = btrfs_ino(BTRFS_I(inode));
911 
912 	path = btrfs_alloc_path();
913 	if (!path)
914 		return -ENOMEM;
915 
916 	min_key.objectid = ino;
917 	min_key.type = BTRFS_EXTENT_DATA_KEY;
918 	min_key.offset = *off;
919 
920 	while (1) {
921 		ret = btrfs_search_forward(root, &min_key, path, newer_than);
922 		if (ret != 0)
923 			goto none;
924 process_slot:
925 		if (min_key.objectid != ino)
926 			goto none;
927 		if (min_key.type != BTRFS_EXTENT_DATA_KEY)
928 			goto none;
929 
930 		leaf = path->nodes[0];
931 		extent = btrfs_item_ptr(leaf, path->slots[0],
932 					struct btrfs_file_extent_item);
933 
934 		type = btrfs_file_extent_type(leaf, extent);
935 		if (type == BTRFS_FILE_EXTENT_REG &&
936 		    btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
937 		    check_defrag_in_cache(inode, min_key.offset, thresh)) {
938 			*off = min_key.offset;
939 			btrfs_free_path(path);
940 			return 0;
941 		}
942 
943 		path->slots[0]++;
944 		if (path->slots[0] < btrfs_header_nritems(leaf)) {
945 			btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
946 			goto process_slot;
947 		}
948 
949 		if (min_key.offset == (u64)-1)
950 			goto none;
951 
952 		min_key.offset++;
953 		btrfs_release_path(path);
954 	}
955 none:
956 	btrfs_free_path(path);
957 	return -ENOENT;
958 }
959 
960 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
961 {
962 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
963 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
964 	struct extent_map *em;
965 	u64 len = PAGE_SIZE;
966 
967 	/*
968 	 * hopefully we have this extent in the tree already, try without
969 	 * the full extent lock
970 	 */
971 	read_lock(&em_tree->lock);
972 	em = lookup_extent_mapping(em_tree, start, len);
973 	read_unlock(&em_tree->lock);
974 
975 	if (!em) {
976 		struct extent_state *cached = NULL;
977 		u64 end = start + len - 1;
978 
979 		/* get the big lock and read metadata off disk */
980 		lock_extent_bits(io_tree, start, end, &cached);
981 		em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
982 		unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
983 
984 		if (IS_ERR(em))
985 			return NULL;
986 	}
987 
988 	return em;
989 }
990 
991 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
992 {
993 	struct extent_map *next;
994 	bool ret = true;
995 
996 	/* this is the last extent */
997 	if (em->start + em->len >= i_size_read(inode))
998 		return false;
999 
1000 	next = defrag_lookup_extent(inode, em->start + em->len);
1001 	if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1002 		ret = false;
1003 	else if ((em->block_start + em->block_len == next->block_start) &&
1004 		 (em->block_len > SZ_128K && next->block_len > SZ_128K))
1005 		ret = false;
1006 
1007 	free_extent_map(next);
1008 	return ret;
1009 }
1010 
1011 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1012 			       u64 *last_len, u64 *skip, u64 *defrag_end,
1013 			       int compress)
1014 {
1015 	struct extent_map *em;
1016 	int ret = 1;
1017 	bool next_mergeable = true;
1018 	bool prev_mergeable = true;
1019 
1020 	/*
1021 	 * make sure that once we start defragging an extent, we keep on
1022 	 * defragging it
1023 	 */
1024 	if (start < *defrag_end)
1025 		return 1;
1026 
1027 	*skip = 0;
1028 
1029 	em = defrag_lookup_extent(inode, start);
1030 	if (!em)
1031 		return 0;
1032 
1033 	/* this will cover holes, and inline extents */
1034 	if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1035 		ret = 0;
1036 		goto out;
1037 	}
1038 
1039 	if (!*defrag_end)
1040 		prev_mergeable = false;
1041 
1042 	next_mergeable = defrag_check_next_extent(inode, em);
1043 	/*
1044 	 * we hit a real extent, if it is big or the next extent is not a
1045 	 * real extent, don't bother defragging it
1046 	 */
1047 	if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1048 	    (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1049 		ret = 0;
1050 out:
1051 	/*
1052 	 * last_len ends up being a counter of how many bytes we've defragged.
1053 	 * every time we choose not to defrag an extent, we reset *last_len
1054 	 * so that the next tiny extent will force a defrag.
1055 	 *
1056 	 * The end result of this is that tiny extents before a single big
1057 	 * extent will force at least part of that big extent to be defragged.
1058 	 */
1059 	if (ret) {
1060 		*defrag_end = extent_map_end(em);
1061 	} else {
1062 		*last_len = 0;
1063 		*skip = extent_map_end(em);
1064 		*defrag_end = 0;
1065 	}
1066 
1067 	free_extent_map(em);
1068 	return ret;
1069 }
1070 
1071 /*
1072  * it doesn't do much good to defrag one or two pages
1073  * at a time.  This pulls in a nice chunk of pages
1074  * to COW and defrag.
1075  *
1076  * It also makes sure the delalloc code has enough
1077  * dirty data to avoid making new small extents as part
1078  * of the defrag
1079  *
1080  * It's a good idea to start RA on this range
1081  * before calling this.
1082  */
1083 static int cluster_pages_for_defrag(struct inode *inode,
1084 				    struct page **pages,
1085 				    unsigned long start_index,
1086 				    unsigned long num_pages)
1087 {
1088 	unsigned long file_end;
1089 	u64 isize = i_size_read(inode);
1090 	u64 page_start;
1091 	u64 page_end;
1092 	u64 page_cnt;
1093 	int ret;
1094 	int i;
1095 	int i_done;
1096 	struct btrfs_ordered_extent *ordered;
1097 	struct extent_state *cached_state = NULL;
1098 	struct extent_io_tree *tree;
1099 	struct extent_changeset *data_reserved = NULL;
1100 	gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1101 
1102 	file_end = (isize - 1) >> PAGE_SHIFT;
1103 	if (!isize || start_index > file_end)
1104 		return 0;
1105 
1106 	page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1107 
1108 	ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
1109 			start_index << PAGE_SHIFT,
1110 			page_cnt << PAGE_SHIFT);
1111 	if (ret)
1112 		return ret;
1113 	i_done = 0;
1114 	tree = &BTRFS_I(inode)->io_tree;
1115 
1116 	/* step one, lock all the pages */
1117 	for (i = 0; i < page_cnt; i++) {
1118 		struct page *page;
1119 again:
1120 		page = find_or_create_page(inode->i_mapping,
1121 					   start_index + i, mask);
1122 		if (!page)
1123 			break;
1124 
1125 		page_start = page_offset(page);
1126 		page_end = page_start + PAGE_SIZE - 1;
1127 		while (1) {
1128 			lock_extent_bits(tree, page_start, page_end,
1129 					 &cached_state);
1130 			ordered = btrfs_lookup_ordered_extent(inode,
1131 							      page_start);
1132 			unlock_extent_cached(tree, page_start, page_end,
1133 					     &cached_state, GFP_NOFS);
1134 			if (!ordered)
1135 				break;
1136 
1137 			unlock_page(page);
1138 			btrfs_start_ordered_extent(inode, ordered, 1);
1139 			btrfs_put_ordered_extent(ordered);
1140 			lock_page(page);
1141 			/*
1142 			 * we unlocked the page above, so we need check if
1143 			 * it was released or not.
1144 			 */
1145 			if (page->mapping != inode->i_mapping) {
1146 				unlock_page(page);
1147 				put_page(page);
1148 				goto again;
1149 			}
1150 		}
1151 
1152 		if (!PageUptodate(page)) {
1153 			btrfs_readpage(NULL, page);
1154 			lock_page(page);
1155 			if (!PageUptodate(page)) {
1156 				unlock_page(page);
1157 				put_page(page);
1158 				ret = -EIO;
1159 				break;
1160 			}
1161 		}
1162 
1163 		if (page->mapping != inode->i_mapping) {
1164 			unlock_page(page);
1165 			put_page(page);
1166 			goto again;
1167 		}
1168 
1169 		pages[i] = page;
1170 		i_done++;
1171 	}
1172 	if (!i_done || ret)
1173 		goto out;
1174 
1175 	if (!(inode->i_sb->s_flags & SB_ACTIVE))
1176 		goto out;
1177 
1178 	/*
1179 	 * so now we have a nice long stream of locked
1180 	 * and up to date pages, lets wait on them
1181 	 */
1182 	for (i = 0; i < i_done; i++)
1183 		wait_on_page_writeback(pages[i]);
1184 
1185 	page_start = page_offset(pages[0]);
1186 	page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1187 
1188 	lock_extent_bits(&BTRFS_I(inode)->io_tree,
1189 			 page_start, page_end - 1, &cached_state);
1190 	clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1191 			  page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1192 			  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1193 			  &cached_state, GFP_NOFS);
1194 
1195 	if (i_done != page_cnt) {
1196 		spin_lock(&BTRFS_I(inode)->lock);
1197 		BTRFS_I(inode)->outstanding_extents++;
1198 		spin_unlock(&BTRFS_I(inode)->lock);
1199 		btrfs_delalloc_release_space(inode, data_reserved,
1200 				start_index << PAGE_SHIFT,
1201 				(page_cnt - i_done) << PAGE_SHIFT);
1202 	}
1203 
1204 
1205 	set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1206 			  &cached_state);
1207 
1208 	unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1209 			     page_start, page_end - 1, &cached_state,
1210 			     GFP_NOFS);
1211 
1212 	for (i = 0; i < i_done; i++) {
1213 		clear_page_dirty_for_io(pages[i]);
1214 		ClearPageChecked(pages[i]);
1215 		set_page_extent_mapped(pages[i]);
1216 		set_page_dirty(pages[i]);
1217 		unlock_page(pages[i]);
1218 		put_page(pages[i]);
1219 	}
1220 	btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1221 	extent_changeset_free(data_reserved);
1222 	return i_done;
1223 out:
1224 	for (i = 0; i < i_done; i++) {
1225 		unlock_page(pages[i]);
1226 		put_page(pages[i]);
1227 	}
1228 	btrfs_delalloc_release_space(inode, data_reserved,
1229 			start_index << PAGE_SHIFT,
1230 			page_cnt << PAGE_SHIFT);
1231 	btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1232 	extent_changeset_free(data_reserved);
1233 	return ret;
1234 
1235 }
1236 
1237 int btrfs_defrag_file(struct inode *inode, struct file *file,
1238 		      struct btrfs_ioctl_defrag_range_args *range,
1239 		      u64 newer_than, unsigned long max_to_defrag)
1240 {
1241 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1242 	struct btrfs_root *root = BTRFS_I(inode)->root;
1243 	struct file_ra_state *ra = NULL;
1244 	unsigned long last_index;
1245 	u64 isize = i_size_read(inode);
1246 	u64 last_len = 0;
1247 	u64 skip = 0;
1248 	u64 defrag_end = 0;
1249 	u64 newer_off = range->start;
1250 	unsigned long i;
1251 	unsigned long ra_index = 0;
1252 	int ret;
1253 	int defrag_count = 0;
1254 	int compress_type = BTRFS_COMPRESS_ZLIB;
1255 	u32 extent_thresh = range->extent_thresh;
1256 	unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1257 	unsigned long cluster = max_cluster;
1258 	u64 new_align = ~((u64)SZ_128K - 1);
1259 	struct page **pages = NULL;
1260 	bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1261 
1262 	if (isize == 0)
1263 		return 0;
1264 
1265 	if (range->start >= isize)
1266 		return -EINVAL;
1267 
1268 	if (do_compress) {
1269 		if (range->compress_type > BTRFS_COMPRESS_TYPES)
1270 			return -EINVAL;
1271 		if (range->compress_type)
1272 			compress_type = range->compress_type;
1273 	}
1274 
1275 	if (extent_thresh == 0)
1276 		extent_thresh = SZ_256K;
1277 
1278 	/*
1279 	 * If we were not given a file, allocate a readahead context. As
1280 	 * readahead is just an optimization, defrag will work without it so
1281 	 * we don't error out.
1282 	 */
1283 	if (!file) {
1284 		ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1285 		if (ra)
1286 			file_ra_state_init(ra, inode->i_mapping);
1287 	} else {
1288 		ra = &file->f_ra;
1289 	}
1290 
1291 	pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1292 	if (!pages) {
1293 		ret = -ENOMEM;
1294 		goto out_ra;
1295 	}
1296 
1297 	/* find the last page to defrag */
1298 	if (range->start + range->len > range->start) {
1299 		last_index = min_t(u64, isize - 1,
1300 			 range->start + range->len - 1) >> PAGE_SHIFT;
1301 	} else {
1302 		last_index = (isize - 1) >> PAGE_SHIFT;
1303 	}
1304 
1305 	if (newer_than) {
1306 		ret = find_new_extents(root, inode, newer_than,
1307 				       &newer_off, SZ_64K);
1308 		if (!ret) {
1309 			range->start = newer_off;
1310 			/*
1311 			 * we always align our defrag to help keep
1312 			 * the extents in the file evenly spaced
1313 			 */
1314 			i = (newer_off & new_align) >> PAGE_SHIFT;
1315 		} else
1316 			goto out_ra;
1317 	} else {
1318 		i = range->start >> PAGE_SHIFT;
1319 	}
1320 	if (!max_to_defrag)
1321 		max_to_defrag = last_index - i + 1;
1322 
1323 	/*
1324 	 * make writeback starts from i, so the defrag range can be
1325 	 * written sequentially.
1326 	 */
1327 	if (i < inode->i_mapping->writeback_index)
1328 		inode->i_mapping->writeback_index = i;
1329 
1330 	while (i <= last_index && defrag_count < max_to_defrag &&
1331 	       (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1332 		/*
1333 		 * make sure we stop running if someone unmounts
1334 		 * the FS
1335 		 */
1336 		if (!(inode->i_sb->s_flags & SB_ACTIVE))
1337 			break;
1338 
1339 		if (btrfs_defrag_cancelled(fs_info)) {
1340 			btrfs_debug(fs_info, "defrag_file cancelled");
1341 			ret = -EAGAIN;
1342 			break;
1343 		}
1344 
1345 		if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1346 					 extent_thresh, &last_len, &skip,
1347 					 &defrag_end, do_compress)){
1348 			unsigned long next;
1349 			/*
1350 			 * the should_defrag function tells us how much to skip
1351 			 * bump our counter by the suggested amount
1352 			 */
1353 			next = DIV_ROUND_UP(skip, PAGE_SIZE);
1354 			i = max(i + 1, next);
1355 			continue;
1356 		}
1357 
1358 		if (!newer_than) {
1359 			cluster = (PAGE_ALIGN(defrag_end) >>
1360 				   PAGE_SHIFT) - i;
1361 			cluster = min(cluster, max_cluster);
1362 		} else {
1363 			cluster = max_cluster;
1364 		}
1365 
1366 		if (i + cluster > ra_index) {
1367 			ra_index = max(i, ra_index);
1368 			if (ra)
1369 				page_cache_sync_readahead(inode->i_mapping, ra,
1370 						file, ra_index, cluster);
1371 			ra_index += cluster;
1372 		}
1373 
1374 		inode_lock(inode);
1375 		if (do_compress)
1376 			BTRFS_I(inode)->defrag_compress = compress_type;
1377 		ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1378 		if (ret < 0) {
1379 			inode_unlock(inode);
1380 			goto out_ra;
1381 		}
1382 
1383 		defrag_count += ret;
1384 		balance_dirty_pages_ratelimited(inode->i_mapping);
1385 		inode_unlock(inode);
1386 
1387 		if (newer_than) {
1388 			if (newer_off == (u64)-1)
1389 				break;
1390 
1391 			if (ret > 0)
1392 				i += ret;
1393 
1394 			newer_off = max(newer_off + 1,
1395 					(u64)i << PAGE_SHIFT);
1396 
1397 			ret = find_new_extents(root, inode, newer_than,
1398 					       &newer_off, SZ_64K);
1399 			if (!ret) {
1400 				range->start = newer_off;
1401 				i = (newer_off & new_align) >> PAGE_SHIFT;
1402 			} else {
1403 				break;
1404 			}
1405 		} else {
1406 			if (ret > 0) {
1407 				i += ret;
1408 				last_len += ret << PAGE_SHIFT;
1409 			} else {
1410 				i++;
1411 				last_len = 0;
1412 			}
1413 		}
1414 	}
1415 
1416 	if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1417 		filemap_flush(inode->i_mapping);
1418 		if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1419 			     &BTRFS_I(inode)->runtime_flags))
1420 			filemap_flush(inode->i_mapping);
1421 	}
1422 
1423 	if (range->compress_type == BTRFS_COMPRESS_LZO) {
1424 		btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1425 	} else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1426 		btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1427 	}
1428 
1429 	ret = defrag_count;
1430 
1431 out_ra:
1432 	if (do_compress) {
1433 		inode_lock(inode);
1434 		BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1435 		inode_unlock(inode);
1436 	}
1437 	if (!file)
1438 		kfree(ra);
1439 	kfree(pages);
1440 	return ret;
1441 }
1442 
1443 static noinline int btrfs_ioctl_resize(struct file *file,
1444 					void __user *arg)
1445 {
1446 	struct inode *inode = file_inode(file);
1447 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1448 	u64 new_size;
1449 	u64 old_size;
1450 	u64 devid = 1;
1451 	struct btrfs_root *root = BTRFS_I(inode)->root;
1452 	struct btrfs_ioctl_vol_args *vol_args;
1453 	struct btrfs_trans_handle *trans;
1454 	struct btrfs_device *device = NULL;
1455 	char *sizestr;
1456 	char *retptr;
1457 	char *devstr = NULL;
1458 	int ret = 0;
1459 	int mod = 0;
1460 
1461 	if (!capable(CAP_SYS_ADMIN))
1462 		return -EPERM;
1463 
1464 	ret = mnt_want_write_file(file);
1465 	if (ret)
1466 		return ret;
1467 
1468 	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
1469 		mnt_drop_write_file(file);
1470 		return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1471 	}
1472 
1473 	mutex_lock(&fs_info->volume_mutex);
1474 	vol_args = memdup_user(arg, sizeof(*vol_args));
1475 	if (IS_ERR(vol_args)) {
1476 		ret = PTR_ERR(vol_args);
1477 		goto out;
1478 	}
1479 
1480 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1481 
1482 	sizestr = vol_args->name;
1483 	devstr = strchr(sizestr, ':');
1484 	if (devstr) {
1485 		sizestr = devstr + 1;
1486 		*devstr = '\0';
1487 		devstr = vol_args->name;
1488 		ret = kstrtoull(devstr, 10, &devid);
1489 		if (ret)
1490 			goto out_free;
1491 		if (!devid) {
1492 			ret = -EINVAL;
1493 			goto out_free;
1494 		}
1495 		btrfs_info(fs_info, "resizing devid %llu", devid);
1496 	}
1497 
1498 	device = btrfs_find_device(fs_info, devid, NULL, NULL);
1499 	if (!device) {
1500 		btrfs_info(fs_info, "resizer unable to find device %llu",
1501 			   devid);
1502 		ret = -ENODEV;
1503 		goto out_free;
1504 	}
1505 
1506 	if (!device->writeable) {
1507 		btrfs_info(fs_info,
1508 			   "resizer unable to apply on readonly device %llu",
1509 		       devid);
1510 		ret = -EPERM;
1511 		goto out_free;
1512 	}
1513 
1514 	if (!strcmp(sizestr, "max"))
1515 		new_size = device->bdev->bd_inode->i_size;
1516 	else {
1517 		if (sizestr[0] == '-') {
1518 			mod = -1;
1519 			sizestr++;
1520 		} else if (sizestr[0] == '+') {
1521 			mod = 1;
1522 			sizestr++;
1523 		}
1524 		new_size = memparse(sizestr, &retptr);
1525 		if (*retptr != '\0' || new_size == 0) {
1526 			ret = -EINVAL;
1527 			goto out_free;
1528 		}
1529 	}
1530 
1531 	if (device->is_tgtdev_for_dev_replace) {
1532 		ret = -EPERM;
1533 		goto out_free;
1534 	}
1535 
1536 	old_size = btrfs_device_get_total_bytes(device);
1537 
1538 	if (mod < 0) {
1539 		if (new_size > old_size) {
1540 			ret = -EINVAL;
1541 			goto out_free;
1542 		}
1543 		new_size = old_size - new_size;
1544 	} else if (mod > 0) {
1545 		if (new_size > ULLONG_MAX - old_size) {
1546 			ret = -ERANGE;
1547 			goto out_free;
1548 		}
1549 		new_size = old_size + new_size;
1550 	}
1551 
1552 	if (new_size < SZ_256M) {
1553 		ret = -EINVAL;
1554 		goto out_free;
1555 	}
1556 	if (new_size > device->bdev->bd_inode->i_size) {
1557 		ret = -EFBIG;
1558 		goto out_free;
1559 	}
1560 
1561 	new_size = round_down(new_size, fs_info->sectorsize);
1562 
1563 	btrfs_info_in_rcu(fs_info, "new size for %s is %llu",
1564 			  rcu_str_deref(device->name), new_size);
1565 
1566 	if (new_size > old_size) {
1567 		trans = btrfs_start_transaction(root, 0);
1568 		if (IS_ERR(trans)) {
1569 			ret = PTR_ERR(trans);
1570 			goto out_free;
1571 		}
1572 		ret = btrfs_grow_device(trans, device, new_size);
1573 		btrfs_commit_transaction(trans);
1574 	} else if (new_size < old_size) {
1575 		ret = btrfs_shrink_device(device, new_size);
1576 	} /* equal, nothing need to do */
1577 
1578 out_free:
1579 	kfree(vol_args);
1580 out:
1581 	mutex_unlock(&fs_info->volume_mutex);
1582 	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
1583 	mnt_drop_write_file(file);
1584 	return ret;
1585 }
1586 
1587 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1588 				const char *name, unsigned long fd, int subvol,
1589 				u64 *transid, bool readonly,
1590 				struct btrfs_qgroup_inherit *inherit)
1591 {
1592 	int namelen;
1593 	int ret = 0;
1594 
1595 	if (!S_ISDIR(file_inode(file)->i_mode))
1596 		return -ENOTDIR;
1597 
1598 	ret = mnt_want_write_file(file);
1599 	if (ret)
1600 		goto out;
1601 
1602 	namelen = strlen(name);
1603 	if (strchr(name, '/')) {
1604 		ret = -EINVAL;
1605 		goto out_drop_write;
1606 	}
1607 
1608 	if (name[0] == '.' &&
1609 	   (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1610 		ret = -EEXIST;
1611 		goto out_drop_write;
1612 	}
1613 
1614 	if (subvol) {
1615 		ret = btrfs_mksubvol(&file->f_path, name, namelen,
1616 				     NULL, transid, readonly, inherit);
1617 	} else {
1618 		struct fd src = fdget(fd);
1619 		struct inode *src_inode;
1620 		if (!src.file) {
1621 			ret = -EINVAL;
1622 			goto out_drop_write;
1623 		}
1624 
1625 		src_inode = file_inode(src.file);
1626 		if (src_inode->i_sb != file_inode(file)->i_sb) {
1627 			btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1628 				   "Snapshot src from another FS");
1629 			ret = -EXDEV;
1630 		} else if (!inode_owner_or_capable(src_inode)) {
1631 			/*
1632 			 * Subvolume creation is not restricted, but snapshots
1633 			 * are limited to own subvolumes only
1634 			 */
1635 			ret = -EPERM;
1636 		} else {
1637 			ret = btrfs_mksubvol(&file->f_path, name, namelen,
1638 					     BTRFS_I(src_inode)->root,
1639 					     transid, readonly, inherit);
1640 		}
1641 		fdput(src);
1642 	}
1643 out_drop_write:
1644 	mnt_drop_write_file(file);
1645 out:
1646 	return ret;
1647 }
1648 
1649 static noinline int btrfs_ioctl_snap_create(struct file *file,
1650 					    void __user *arg, int subvol)
1651 {
1652 	struct btrfs_ioctl_vol_args *vol_args;
1653 	int ret;
1654 
1655 	if (!S_ISDIR(file_inode(file)->i_mode))
1656 		return -ENOTDIR;
1657 
1658 	vol_args = memdup_user(arg, sizeof(*vol_args));
1659 	if (IS_ERR(vol_args))
1660 		return PTR_ERR(vol_args);
1661 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1662 
1663 	ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1664 					      vol_args->fd, subvol,
1665 					      NULL, false, NULL);
1666 
1667 	kfree(vol_args);
1668 	return ret;
1669 }
1670 
1671 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1672 					       void __user *arg, int subvol)
1673 {
1674 	struct btrfs_ioctl_vol_args_v2 *vol_args;
1675 	int ret;
1676 	u64 transid = 0;
1677 	u64 *ptr = NULL;
1678 	bool readonly = false;
1679 	struct btrfs_qgroup_inherit *inherit = NULL;
1680 
1681 	if (!S_ISDIR(file_inode(file)->i_mode))
1682 		return -ENOTDIR;
1683 
1684 	vol_args = memdup_user(arg, sizeof(*vol_args));
1685 	if (IS_ERR(vol_args))
1686 		return PTR_ERR(vol_args);
1687 	vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1688 
1689 	if (vol_args->flags &
1690 	    ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1691 	      BTRFS_SUBVOL_QGROUP_INHERIT)) {
1692 		ret = -EOPNOTSUPP;
1693 		goto free_args;
1694 	}
1695 
1696 	if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1697 		ptr = &transid;
1698 	if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1699 		readonly = true;
1700 	if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1701 		if (vol_args->size > PAGE_SIZE) {
1702 			ret = -EINVAL;
1703 			goto free_args;
1704 		}
1705 		inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1706 		if (IS_ERR(inherit)) {
1707 			ret = PTR_ERR(inherit);
1708 			goto free_args;
1709 		}
1710 	}
1711 
1712 	ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1713 					      vol_args->fd, subvol, ptr,
1714 					      readonly, inherit);
1715 	if (ret)
1716 		goto free_inherit;
1717 
1718 	if (ptr && copy_to_user(arg +
1719 				offsetof(struct btrfs_ioctl_vol_args_v2,
1720 					transid),
1721 				ptr, sizeof(*ptr)))
1722 		ret = -EFAULT;
1723 
1724 free_inherit:
1725 	kfree(inherit);
1726 free_args:
1727 	kfree(vol_args);
1728 	return ret;
1729 }
1730 
1731 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1732 						void __user *arg)
1733 {
1734 	struct inode *inode = file_inode(file);
1735 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1736 	struct btrfs_root *root = BTRFS_I(inode)->root;
1737 	int ret = 0;
1738 	u64 flags = 0;
1739 
1740 	if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1741 		return -EINVAL;
1742 
1743 	down_read(&fs_info->subvol_sem);
1744 	if (btrfs_root_readonly(root))
1745 		flags |= BTRFS_SUBVOL_RDONLY;
1746 	up_read(&fs_info->subvol_sem);
1747 
1748 	if (copy_to_user(arg, &flags, sizeof(flags)))
1749 		ret = -EFAULT;
1750 
1751 	return ret;
1752 }
1753 
1754 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1755 					      void __user *arg)
1756 {
1757 	struct inode *inode = file_inode(file);
1758 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1759 	struct btrfs_root *root = BTRFS_I(inode)->root;
1760 	struct btrfs_trans_handle *trans;
1761 	u64 root_flags;
1762 	u64 flags;
1763 	int ret = 0;
1764 
1765 	if (!inode_owner_or_capable(inode))
1766 		return -EPERM;
1767 
1768 	ret = mnt_want_write_file(file);
1769 	if (ret)
1770 		goto out;
1771 
1772 	if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1773 		ret = -EINVAL;
1774 		goto out_drop_write;
1775 	}
1776 
1777 	if (copy_from_user(&flags, arg, sizeof(flags))) {
1778 		ret = -EFAULT;
1779 		goto out_drop_write;
1780 	}
1781 
1782 	if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1783 		ret = -EINVAL;
1784 		goto out_drop_write;
1785 	}
1786 
1787 	if (flags & ~BTRFS_SUBVOL_RDONLY) {
1788 		ret = -EOPNOTSUPP;
1789 		goto out_drop_write;
1790 	}
1791 
1792 	down_write(&fs_info->subvol_sem);
1793 
1794 	/* nothing to do */
1795 	if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1796 		goto out_drop_sem;
1797 
1798 	root_flags = btrfs_root_flags(&root->root_item);
1799 	if (flags & BTRFS_SUBVOL_RDONLY) {
1800 		btrfs_set_root_flags(&root->root_item,
1801 				     root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1802 	} else {
1803 		/*
1804 		 * Block RO -> RW transition if this subvolume is involved in
1805 		 * send
1806 		 */
1807 		spin_lock(&root->root_item_lock);
1808 		if (root->send_in_progress == 0) {
1809 			btrfs_set_root_flags(&root->root_item,
1810 				     root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1811 			spin_unlock(&root->root_item_lock);
1812 		} else {
1813 			spin_unlock(&root->root_item_lock);
1814 			btrfs_warn(fs_info,
1815 				   "Attempt to set subvolume %llu read-write during send",
1816 				   root->root_key.objectid);
1817 			ret = -EPERM;
1818 			goto out_drop_sem;
1819 		}
1820 	}
1821 
1822 	trans = btrfs_start_transaction(root, 1);
1823 	if (IS_ERR(trans)) {
1824 		ret = PTR_ERR(trans);
1825 		goto out_reset;
1826 	}
1827 
1828 	ret = btrfs_update_root(trans, fs_info->tree_root,
1829 				&root->root_key, &root->root_item);
1830 	if (ret < 0) {
1831 		btrfs_end_transaction(trans);
1832 		goto out_reset;
1833 	}
1834 
1835 	ret = btrfs_commit_transaction(trans);
1836 
1837 out_reset:
1838 	if (ret)
1839 		btrfs_set_root_flags(&root->root_item, root_flags);
1840 out_drop_sem:
1841 	up_write(&fs_info->subvol_sem);
1842 out_drop_write:
1843 	mnt_drop_write_file(file);
1844 out:
1845 	return ret;
1846 }
1847 
1848 /*
1849  * helper to check if the subvolume references other subvolumes
1850  */
1851 static noinline int may_destroy_subvol(struct btrfs_root *root)
1852 {
1853 	struct btrfs_fs_info *fs_info = root->fs_info;
1854 	struct btrfs_path *path;
1855 	struct btrfs_dir_item *di;
1856 	struct btrfs_key key;
1857 	u64 dir_id;
1858 	int ret;
1859 
1860 	path = btrfs_alloc_path();
1861 	if (!path)
1862 		return -ENOMEM;
1863 
1864 	/* Make sure this root isn't set as the default subvol */
1865 	dir_id = btrfs_super_root_dir(fs_info->super_copy);
1866 	di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
1867 				   dir_id, "default", 7, 0);
1868 	if (di && !IS_ERR(di)) {
1869 		btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1870 		if (key.objectid == root->root_key.objectid) {
1871 			ret = -EPERM;
1872 			btrfs_err(fs_info,
1873 				  "deleting default subvolume %llu is not allowed",
1874 				  key.objectid);
1875 			goto out;
1876 		}
1877 		btrfs_release_path(path);
1878 	}
1879 
1880 	key.objectid = root->root_key.objectid;
1881 	key.type = BTRFS_ROOT_REF_KEY;
1882 	key.offset = (u64)-1;
1883 
1884 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1885 	if (ret < 0)
1886 		goto out;
1887 	BUG_ON(ret == 0);
1888 
1889 	ret = 0;
1890 	if (path->slots[0] > 0) {
1891 		path->slots[0]--;
1892 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1893 		if (key.objectid == root->root_key.objectid &&
1894 		    key.type == BTRFS_ROOT_REF_KEY)
1895 			ret = -ENOTEMPTY;
1896 	}
1897 out:
1898 	btrfs_free_path(path);
1899 	return ret;
1900 }
1901 
1902 static noinline int key_in_sk(struct btrfs_key *key,
1903 			      struct btrfs_ioctl_search_key *sk)
1904 {
1905 	struct btrfs_key test;
1906 	int ret;
1907 
1908 	test.objectid = sk->min_objectid;
1909 	test.type = sk->min_type;
1910 	test.offset = sk->min_offset;
1911 
1912 	ret = btrfs_comp_cpu_keys(key, &test);
1913 	if (ret < 0)
1914 		return 0;
1915 
1916 	test.objectid = sk->max_objectid;
1917 	test.type = sk->max_type;
1918 	test.offset = sk->max_offset;
1919 
1920 	ret = btrfs_comp_cpu_keys(key, &test);
1921 	if (ret > 0)
1922 		return 0;
1923 	return 1;
1924 }
1925 
1926 static noinline int copy_to_sk(struct btrfs_path *path,
1927 			       struct btrfs_key *key,
1928 			       struct btrfs_ioctl_search_key *sk,
1929 			       size_t *buf_size,
1930 			       char __user *ubuf,
1931 			       unsigned long *sk_offset,
1932 			       int *num_found)
1933 {
1934 	u64 found_transid;
1935 	struct extent_buffer *leaf;
1936 	struct btrfs_ioctl_search_header sh;
1937 	struct btrfs_key test;
1938 	unsigned long item_off;
1939 	unsigned long item_len;
1940 	int nritems;
1941 	int i;
1942 	int slot;
1943 	int ret = 0;
1944 
1945 	leaf = path->nodes[0];
1946 	slot = path->slots[0];
1947 	nritems = btrfs_header_nritems(leaf);
1948 
1949 	if (btrfs_header_generation(leaf) > sk->max_transid) {
1950 		i = nritems;
1951 		goto advance_key;
1952 	}
1953 	found_transid = btrfs_header_generation(leaf);
1954 
1955 	for (i = slot; i < nritems; i++) {
1956 		item_off = btrfs_item_ptr_offset(leaf, i);
1957 		item_len = btrfs_item_size_nr(leaf, i);
1958 
1959 		btrfs_item_key_to_cpu(leaf, key, i);
1960 		if (!key_in_sk(key, sk))
1961 			continue;
1962 
1963 		if (sizeof(sh) + item_len > *buf_size) {
1964 			if (*num_found) {
1965 				ret = 1;
1966 				goto out;
1967 			}
1968 
1969 			/*
1970 			 * return one empty item back for v1, which does not
1971 			 * handle -EOVERFLOW
1972 			 */
1973 
1974 			*buf_size = sizeof(sh) + item_len;
1975 			item_len = 0;
1976 			ret = -EOVERFLOW;
1977 		}
1978 
1979 		if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
1980 			ret = 1;
1981 			goto out;
1982 		}
1983 
1984 		sh.objectid = key->objectid;
1985 		sh.offset = key->offset;
1986 		sh.type = key->type;
1987 		sh.len = item_len;
1988 		sh.transid = found_transid;
1989 
1990 		/* copy search result header */
1991 		if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
1992 			ret = -EFAULT;
1993 			goto out;
1994 		}
1995 
1996 		*sk_offset += sizeof(sh);
1997 
1998 		if (item_len) {
1999 			char __user *up = ubuf + *sk_offset;
2000 			/* copy the item */
2001 			if (read_extent_buffer_to_user(leaf, up,
2002 						       item_off, item_len)) {
2003 				ret = -EFAULT;
2004 				goto out;
2005 			}
2006 
2007 			*sk_offset += item_len;
2008 		}
2009 		(*num_found)++;
2010 
2011 		if (ret) /* -EOVERFLOW from above */
2012 			goto out;
2013 
2014 		if (*num_found >= sk->nr_items) {
2015 			ret = 1;
2016 			goto out;
2017 		}
2018 	}
2019 advance_key:
2020 	ret = 0;
2021 	test.objectid = sk->max_objectid;
2022 	test.type = sk->max_type;
2023 	test.offset = sk->max_offset;
2024 	if (btrfs_comp_cpu_keys(key, &test) >= 0)
2025 		ret = 1;
2026 	else if (key->offset < (u64)-1)
2027 		key->offset++;
2028 	else if (key->type < (u8)-1) {
2029 		key->offset = 0;
2030 		key->type++;
2031 	} else if (key->objectid < (u64)-1) {
2032 		key->offset = 0;
2033 		key->type = 0;
2034 		key->objectid++;
2035 	} else
2036 		ret = 1;
2037 out:
2038 	/*
2039 	 *  0: all items from this leaf copied, continue with next
2040 	 *  1: * more items can be copied, but unused buffer is too small
2041 	 *     * all items were found
2042 	 *     Either way, it will stops the loop which iterates to the next
2043 	 *     leaf
2044 	 *  -EOVERFLOW: item was to large for buffer
2045 	 *  -EFAULT: could not copy extent buffer back to userspace
2046 	 */
2047 	return ret;
2048 }
2049 
2050 static noinline int search_ioctl(struct inode *inode,
2051 				 struct btrfs_ioctl_search_key *sk,
2052 				 size_t *buf_size,
2053 				 char __user *ubuf)
2054 {
2055 	struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2056 	struct btrfs_root *root;
2057 	struct btrfs_key key;
2058 	struct btrfs_path *path;
2059 	int ret;
2060 	int num_found = 0;
2061 	unsigned long sk_offset = 0;
2062 
2063 	if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2064 		*buf_size = sizeof(struct btrfs_ioctl_search_header);
2065 		return -EOVERFLOW;
2066 	}
2067 
2068 	path = btrfs_alloc_path();
2069 	if (!path)
2070 		return -ENOMEM;
2071 
2072 	if (sk->tree_id == 0) {
2073 		/* search the root of the inode that was passed */
2074 		root = BTRFS_I(inode)->root;
2075 	} else {
2076 		key.objectid = sk->tree_id;
2077 		key.type = BTRFS_ROOT_ITEM_KEY;
2078 		key.offset = (u64)-1;
2079 		root = btrfs_read_fs_root_no_name(info, &key);
2080 		if (IS_ERR(root)) {
2081 			btrfs_free_path(path);
2082 			return -ENOENT;
2083 		}
2084 	}
2085 
2086 	key.objectid = sk->min_objectid;
2087 	key.type = sk->min_type;
2088 	key.offset = sk->min_offset;
2089 
2090 	while (1) {
2091 		ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2092 		if (ret != 0) {
2093 			if (ret > 0)
2094 				ret = 0;
2095 			goto err;
2096 		}
2097 		ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2098 				 &sk_offset, &num_found);
2099 		btrfs_release_path(path);
2100 		if (ret)
2101 			break;
2102 
2103 	}
2104 	if (ret > 0)
2105 		ret = 0;
2106 err:
2107 	sk->nr_items = num_found;
2108 	btrfs_free_path(path);
2109 	return ret;
2110 }
2111 
2112 static noinline int btrfs_ioctl_tree_search(struct file *file,
2113 					   void __user *argp)
2114 {
2115 	struct btrfs_ioctl_search_args __user *uargs;
2116 	struct btrfs_ioctl_search_key sk;
2117 	struct inode *inode;
2118 	int ret;
2119 	size_t buf_size;
2120 
2121 	if (!capable(CAP_SYS_ADMIN))
2122 		return -EPERM;
2123 
2124 	uargs = (struct btrfs_ioctl_search_args __user *)argp;
2125 
2126 	if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2127 		return -EFAULT;
2128 
2129 	buf_size = sizeof(uargs->buf);
2130 
2131 	inode = file_inode(file);
2132 	ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2133 
2134 	/*
2135 	 * In the origin implementation an overflow is handled by returning a
2136 	 * search header with a len of zero, so reset ret.
2137 	 */
2138 	if (ret == -EOVERFLOW)
2139 		ret = 0;
2140 
2141 	if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2142 		ret = -EFAULT;
2143 	return ret;
2144 }
2145 
2146 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2147 					       void __user *argp)
2148 {
2149 	struct btrfs_ioctl_search_args_v2 __user *uarg;
2150 	struct btrfs_ioctl_search_args_v2 args;
2151 	struct inode *inode;
2152 	int ret;
2153 	size_t buf_size;
2154 	const size_t buf_limit = SZ_16M;
2155 
2156 	if (!capable(CAP_SYS_ADMIN))
2157 		return -EPERM;
2158 
2159 	/* copy search header and buffer size */
2160 	uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2161 	if (copy_from_user(&args, uarg, sizeof(args)))
2162 		return -EFAULT;
2163 
2164 	buf_size = args.buf_size;
2165 
2166 	/* limit result size to 16MB */
2167 	if (buf_size > buf_limit)
2168 		buf_size = buf_limit;
2169 
2170 	inode = file_inode(file);
2171 	ret = search_ioctl(inode, &args.key, &buf_size,
2172 			   (char __user *)(&uarg->buf[0]));
2173 	if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2174 		ret = -EFAULT;
2175 	else if (ret == -EOVERFLOW &&
2176 		copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2177 		ret = -EFAULT;
2178 
2179 	return ret;
2180 }
2181 
2182 /*
2183  * Search INODE_REFs to identify path name of 'dirid' directory
2184  * in a 'tree_id' tree. and sets path name to 'name'.
2185  */
2186 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2187 				u64 tree_id, u64 dirid, char *name)
2188 {
2189 	struct btrfs_root *root;
2190 	struct btrfs_key key;
2191 	char *ptr;
2192 	int ret = -1;
2193 	int slot;
2194 	int len;
2195 	int total_len = 0;
2196 	struct btrfs_inode_ref *iref;
2197 	struct extent_buffer *l;
2198 	struct btrfs_path *path;
2199 
2200 	if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2201 		name[0]='\0';
2202 		return 0;
2203 	}
2204 
2205 	path = btrfs_alloc_path();
2206 	if (!path)
2207 		return -ENOMEM;
2208 
2209 	ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2210 
2211 	key.objectid = tree_id;
2212 	key.type = BTRFS_ROOT_ITEM_KEY;
2213 	key.offset = (u64)-1;
2214 	root = btrfs_read_fs_root_no_name(info, &key);
2215 	if (IS_ERR(root)) {
2216 		btrfs_err(info, "could not find root %llu", tree_id);
2217 		ret = -ENOENT;
2218 		goto out;
2219 	}
2220 
2221 	key.objectid = dirid;
2222 	key.type = BTRFS_INODE_REF_KEY;
2223 	key.offset = (u64)-1;
2224 
2225 	while (1) {
2226 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2227 		if (ret < 0)
2228 			goto out;
2229 		else if (ret > 0) {
2230 			ret = btrfs_previous_item(root, path, dirid,
2231 						  BTRFS_INODE_REF_KEY);
2232 			if (ret < 0)
2233 				goto out;
2234 			else if (ret > 0) {
2235 				ret = -ENOENT;
2236 				goto out;
2237 			}
2238 		}
2239 
2240 		l = path->nodes[0];
2241 		slot = path->slots[0];
2242 		btrfs_item_key_to_cpu(l, &key, slot);
2243 
2244 		iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2245 		len = btrfs_inode_ref_name_len(l, iref);
2246 		ptr -= len + 1;
2247 		total_len += len + 1;
2248 		if (ptr < name) {
2249 			ret = -ENAMETOOLONG;
2250 			goto out;
2251 		}
2252 
2253 		*(ptr + len) = '/';
2254 		read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2255 
2256 		if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2257 			break;
2258 
2259 		btrfs_release_path(path);
2260 		key.objectid = key.offset;
2261 		key.offset = (u64)-1;
2262 		dirid = key.objectid;
2263 	}
2264 	memmove(name, ptr, total_len);
2265 	name[total_len] = '\0';
2266 	ret = 0;
2267 out:
2268 	btrfs_free_path(path);
2269 	return ret;
2270 }
2271 
2272 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2273 					   void __user *argp)
2274 {
2275 	 struct btrfs_ioctl_ino_lookup_args *args;
2276 	 struct inode *inode;
2277 	int ret = 0;
2278 
2279 	args = memdup_user(argp, sizeof(*args));
2280 	if (IS_ERR(args))
2281 		return PTR_ERR(args);
2282 
2283 	inode = file_inode(file);
2284 
2285 	/*
2286 	 * Unprivileged query to obtain the containing subvolume root id. The
2287 	 * path is reset so it's consistent with btrfs_search_path_in_tree.
2288 	 */
2289 	if (args->treeid == 0)
2290 		args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2291 
2292 	if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2293 		args->name[0] = 0;
2294 		goto out;
2295 	}
2296 
2297 	if (!capable(CAP_SYS_ADMIN)) {
2298 		ret = -EPERM;
2299 		goto out;
2300 	}
2301 
2302 	ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2303 					args->treeid, args->objectid,
2304 					args->name);
2305 
2306 out:
2307 	if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2308 		ret = -EFAULT;
2309 
2310 	kfree(args);
2311 	return ret;
2312 }
2313 
2314 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2315 					     void __user *arg)
2316 {
2317 	struct dentry *parent = file->f_path.dentry;
2318 	struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2319 	struct dentry *dentry;
2320 	struct inode *dir = d_inode(parent);
2321 	struct inode *inode;
2322 	struct btrfs_root *root = BTRFS_I(dir)->root;
2323 	struct btrfs_root *dest = NULL;
2324 	struct btrfs_ioctl_vol_args *vol_args;
2325 	struct btrfs_trans_handle *trans;
2326 	struct btrfs_block_rsv block_rsv;
2327 	u64 root_flags;
2328 	u64 qgroup_reserved;
2329 	int namelen;
2330 	int ret;
2331 	int err = 0;
2332 
2333 	if (!S_ISDIR(dir->i_mode))
2334 		return -ENOTDIR;
2335 
2336 	vol_args = memdup_user(arg, sizeof(*vol_args));
2337 	if (IS_ERR(vol_args))
2338 		return PTR_ERR(vol_args);
2339 
2340 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2341 	namelen = strlen(vol_args->name);
2342 	if (strchr(vol_args->name, '/') ||
2343 	    strncmp(vol_args->name, "..", namelen) == 0) {
2344 		err = -EINVAL;
2345 		goto out;
2346 	}
2347 
2348 	err = mnt_want_write_file(file);
2349 	if (err)
2350 		goto out;
2351 
2352 
2353 	err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2354 	if (err == -EINTR)
2355 		goto out_drop_write;
2356 	dentry = lookup_one_len(vol_args->name, parent, namelen);
2357 	if (IS_ERR(dentry)) {
2358 		err = PTR_ERR(dentry);
2359 		goto out_unlock_dir;
2360 	}
2361 
2362 	if (d_really_is_negative(dentry)) {
2363 		err = -ENOENT;
2364 		goto out_dput;
2365 	}
2366 
2367 	inode = d_inode(dentry);
2368 	dest = BTRFS_I(inode)->root;
2369 	if (!capable(CAP_SYS_ADMIN)) {
2370 		/*
2371 		 * Regular user.  Only allow this with a special mount
2372 		 * option, when the user has write+exec access to the
2373 		 * subvol root, and when rmdir(2) would have been
2374 		 * allowed.
2375 		 *
2376 		 * Note that this is _not_ check that the subvol is
2377 		 * empty or doesn't contain data that we wouldn't
2378 		 * otherwise be able to delete.
2379 		 *
2380 		 * Users who want to delete empty subvols should try
2381 		 * rmdir(2).
2382 		 */
2383 		err = -EPERM;
2384 		if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2385 			goto out_dput;
2386 
2387 		/*
2388 		 * Do not allow deletion if the parent dir is the same
2389 		 * as the dir to be deleted.  That means the ioctl
2390 		 * must be called on the dentry referencing the root
2391 		 * of the subvol, not a random directory contained
2392 		 * within it.
2393 		 */
2394 		err = -EINVAL;
2395 		if (root == dest)
2396 			goto out_dput;
2397 
2398 		err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2399 		if (err)
2400 			goto out_dput;
2401 	}
2402 
2403 	/* check if subvolume may be deleted by a user */
2404 	err = btrfs_may_delete(dir, dentry, 1);
2405 	if (err)
2406 		goto out_dput;
2407 
2408 	if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2409 		err = -EINVAL;
2410 		goto out_dput;
2411 	}
2412 
2413 	inode_lock(inode);
2414 
2415 	/*
2416 	 * Don't allow to delete a subvolume with send in progress. This is
2417 	 * inside the i_mutex so the error handling that has to drop the bit
2418 	 * again is not run concurrently.
2419 	 */
2420 	spin_lock(&dest->root_item_lock);
2421 	root_flags = btrfs_root_flags(&dest->root_item);
2422 	if (dest->send_in_progress == 0) {
2423 		btrfs_set_root_flags(&dest->root_item,
2424 				root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2425 		spin_unlock(&dest->root_item_lock);
2426 	} else {
2427 		spin_unlock(&dest->root_item_lock);
2428 		btrfs_warn(fs_info,
2429 			   "Attempt to delete subvolume %llu during send",
2430 			   dest->root_key.objectid);
2431 		err = -EPERM;
2432 		goto out_unlock_inode;
2433 	}
2434 
2435 	down_write(&fs_info->subvol_sem);
2436 
2437 	err = may_destroy_subvol(dest);
2438 	if (err)
2439 		goto out_up_write;
2440 
2441 	btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2442 	/*
2443 	 * One for dir inode, two for dir entries, two for root
2444 	 * ref/backref.
2445 	 */
2446 	err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2447 					       5, &qgroup_reserved, true);
2448 	if (err)
2449 		goto out_up_write;
2450 
2451 	trans = btrfs_start_transaction(root, 0);
2452 	if (IS_ERR(trans)) {
2453 		err = PTR_ERR(trans);
2454 		goto out_release;
2455 	}
2456 	trans->block_rsv = &block_rsv;
2457 	trans->bytes_reserved = block_rsv.size;
2458 
2459 	btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
2460 
2461 	ret = btrfs_unlink_subvol(trans, root, dir,
2462 				dest->root_key.objectid,
2463 				dentry->d_name.name,
2464 				dentry->d_name.len);
2465 	if (ret) {
2466 		err = ret;
2467 		btrfs_abort_transaction(trans, ret);
2468 		goto out_end_trans;
2469 	}
2470 
2471 	btrfs_record_root_in_trans(trans, dest);
2472 
2473 	memset(&dest->root_item.drop_progress, 0,
2474 		sizeof(dest->root_item.drop_progress));
2475 	dest->root_item.drop_level = 0;
2476 	btrfs_set_root_refs(&dest->root_item, 0);
2477 
2478 	if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2479 		ret = btrfs_insert_orphan_item(trans,
2480 					fs_info->tree_root,
2481 					dest->root_key.objectid);
2482 		if (ret) {
2483 			btrfs_abort_transaction(trans, ret);
2484 			err = ret;
2485 			goto out_end_trans;
2486 		}
2487 	}
2488 
2489 	ret = btrfs_uuid_tree_rem(trans, fs_info, dest->root_item.uuid,
2490 				  BTRFS_UUID_KEY_SUBVOL,
2491 				  dest->root_key.objectid);
2492 	if (ret && ret != -ENOENT) {
2493 		btrfs_abort_transaction(trans, ret);
2494 		err = ret;
2495 		goto out_end_trans;
2496 	}
2497 	if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2498 		ret = btrfs_uuid_tree_rem(trans, fs_info,
2499 					  dest->root_item.received_uuid,
2500 					  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2501 					  dest->root_key.objectid);
2502 		if (ret && ret != -ENOENT) {
2503 			btrfs_abort_transaction(trans, ret);
2504 			err = ret;
2505 			goto out_end_trans;
2506 		}
2507 	}
2508 
2509 out_end_trans:
2510 	trans->block_rsv = NULL;
2511 	trans->bytes_reserved = 0;
2512 	ret = btrfs_end_transaction(trans);
2513 	if (ret && !err)
2514 		err = ret;
2515 	inode->i_flags |= S_DEAD;
2516 out_release:
2517 	btrfs_subvolume_release_metadata(fs_info, &block_rsv);
2518 out_up_write:
2519 	up_write(&fs_info->subvol_sem);
2520 	if (err) {
2521 		spin_lock(&dest->root_item_lock);
2522 		root_flags = btrfs_root_flags(&dest->root_item);
2523 		btrfs_set_root_flags(&dest->root_item,
2524 				root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2525 		spin_unlock(&dest->root_item_lock);
2526 	}
2527 out_unlock_inode:
2528 	inode_unlock(inode);
2529 	if (!err) {
2530 		d_invalidate(dentry);
2531 		btrfs_invalidate_inodes(dest);
2532 		d_delete(dentry);
2533 		ASSERT(dest->send_in_progress == 0);
2534 
2535 		/* the last ref */
2536 		if (dest->ino_cache_inode) {
2537 			iput(dest->ino_cache_inode);
2538 			dest->ino_cache_inode = NULL;
2539 		}
2540 	}
2541 out_dput:
2542 	dput(dentry);
2543 out_unlock_dir:
2544 	inode_unlock(dir);
2545 out_drop_write:
2546 	mnt_drop_write_file(file);
2547 out:
2548 	kfree(vol_args);
2549 	return err;
2550 }
2551 
2552 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2553 {
2554 	struct inode *inode = file_inode(file);
2555 	struct btrfs_root *root = BTRFS_I(inode)->root;
2556 	struct btrfs_ioctl_defrag_range_args *range;
2557 	int ret;
2558 
2559 	ret = mnt_want_write_file(file);
2560 	if (ret)
2561 		return ret;
2562 
2563 	if (btrfs_root_readonly(root)) {
2564 		ret = -EROFS;
2565 		goto out;
2566 	}
2567 
2568 	switch (inode->i_mode & S_IFMT) {
2569 	case S_IFDIR:
2570 		if (!capable(CAP_SYS_ADMIN)) {
2571 			ret = -EPERM;
2572 			goto out;
2573 		}
2574 		ret = btrfs_defrag_root(root);
2575 		break;
2576 	case S_IFREG:
2577 		if (!(file->f_mode & FMODE_WRITE)) {
2578 			ret = -EINVAL;
2579 			goto out;
2580 		}
2581 
2582 		range = kzalloc(sizeof(*range), GFP_KERNEL);
2583 		if (!range) {
2584 			ret = -ENOMEM;
2585 			goto out;
2586 		}
2587 
2588 		if (argp) {
2589 			if (copy_from_user(range, argp,
2590 					   sizeof(*range))) {
2591 				ret = -EFAULT;
2592 				kfree(range);
2593 				goto out;
2594 			}
2595 			/* compression requires us to start the IO */
2596 			if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2597 				range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2598 				range->extent_thresh = (u32)-1;
2599 			}
2600 		} else {
2601 			/* the rest are all set to zero by kzalloc */
2602 			range->len = (u64)-1;
2603 		}
2604 		ret = btrfs_defrag_file(file_inode(file), file,
2605 					range, 0, 0);
2606 		if (ret > 0)
2607 			ret = 0;
2608 		kfree(range);
2609 		break;
2610 	default:
2611 		ret = -EINVAL;
2612 	}
2613 out:
2614 	mnt_drop_write_file(file);
2615 	return ret;
2616 }
2617 
2618 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
2619 {
2620 	struct btrfs_ioctl_vol_args *vol_args;
2621 	int ret;
2622 
2623 	if (!capable(CAP_SYS_ADMIN))
2624 		return -EPERM;
2625 
2626 	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
2627 		return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2628 
2629 	mutex_lock(&fs_info->volume_mutex);
2630 	vol_args = memdup_user(arg, sizeof(*vol_args));
2631 	if (IS_ERR(vol_args)) {
2632 		ret = PTR_ERR(vol_args);
2633 		goto out;
2634 	}
2635 
2636 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2637 	ret = btrfs_init_new_device(fs_info, vol_args->name);
2638 
2639 	if (!ret)
2640 		btrfs_info(fs_info, "disk added %s", vol_args->name);
2641 
2642 	kfree(vol_args);
2643 out:
2644 	mutex_unlock(&fs_info->volume_mutex);
2645 	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2646 	return ret;
2647 }
2648 
2649 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
2650 {
2651 	struct inode *inode = file_inode(file);
2652 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2653 	struct btrfs_ioctl_vol_args_v2 *vol_args;
2654 	int ret;
2655 
2656 	if (!capable(CAP_SYS_ADMIN))
2657 		return -EPERM;
2658 
2659 	ret = mnt_want_write_file(file);
2660 	if (ret)
2661 		return ret;
2662 
2663 	vol_args = memdup_user(arg, sizeof(*vol_args));
2664 	if (IS_ERR(vol_args)) {
2665 		ret = PTR_ERR(vol_args);
2666 		goto err_drop;
2667 	}
2668 
2669 	/* Check for compatibility reject unknown flags */
2670 	if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED)
2671 		return -EOPNOTSUPP;
2672 
2673 	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
2674 		ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2675 		goto out;
2676 	}
2677 
2678 	mutex_lock(&fs_info->volume_mutex);
2679 	if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
2680 		ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
2681 	} else {
2682 		vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2683 		ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2684 	}
2685 	mutex_unlock(&fs_info->volume_mutex);
2686 	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2687 
2688 	if (!ret) {
2689 		if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
2690 			btrfs_info(fs_info, "device deleted: id %llu",
2691 					vol_args->devid);
2692 		else
2693 			btrfs_info(fs_info, "device deleted: %s",
2694 					vol_args->name);
2695 	}
2696 out:
2697 	kfree(vol_args);
2698 err_drop:
2699 	mnt_drop_write_file(file);
2700 	return ret;
2701 }
2702 
2703 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2704 {
2705 	struct inode *inode = file_inode(file);
2706 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2707 	struct btrfs_ioctl_vol_args *vol_args;
2708 	int ret;
2709 
2710 	if (!capable(CAP_SYS_ADMIN))
2711 		return -EPERM;
2712 
2713 	ret = mnt_want_write_file(file);
2714 	if (ret)
2715 		return ret;
2716 
2717 	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
2718 		ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2719 		goto out_drop_write;
2720 	}
2721 
2722 	vol_args = memdup_user(arg, sizeof(*vol_args));
2723 	if (IS_ERR(vol_args)) {
2724 		ret = PTR_ERR(vol_args);
2725 		goto out;
2726 	}
2727 
2728 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2729 	mutex_lock(&fs_info->volume_mutex);
2730 	ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2731 	mutex_unlock(&fs_info->volume_mutex);
2732 
2733 	if (!ret)
2734 		btrfs_info(fs_info, "disk deleted %s", vol_args->name);
2735 	kfree(vol_args);
2736 out:
2737 	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2738 out_drop_write:
2739 	mnt_drop_write_file(file);
2740 
2741 	return ret;
2742 }
2743 
2744 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
2745 				void __user *arg)
2746 {
2747 	struct btrfs_ioctl_fs_info_args *fi_args;
2748 	struct btrfs_device *device;
2749 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2750 	int ret = 0;
2751 
2752 	fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2753 	if (!fi_args)
2754 		return -ENOMEM;
2755 
2756 	mutex_lock(&fs_devices->device_list_mutex);
2757 	fi_args->num_devices = fs_devices->num_devices;
2758 	memcpy(&fi_args->fsid, fs_info->fsid, sizeof(fi_args->fsid));
2759 
2760 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
2761 		if (device->devid > fi_args->max_id)
2762 			fi_args->max_id = device->devid;
2763 	}
2764 	mutex_unlock(&fs_devices->device_list_mutex);
2765 
2766 	fi_args->nodesize = fs_info->nodesize;
2767 	fi_args->sectorsize = fs_info->sectorsize;
2768 	fi_args->clone_alignment = fs_info->sectorsize;
2769 
2770 	if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2771 		ret = -EFAULT;
2772 
2773 	kfree(fi_args);
2774 	return ret;
2775 }
2776 
2777 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
2778 				 void __user *arg)
2779 {
2780 	struct btrfs_ioctl_dev_info_args *di_args;
2781 	struct btrfs_device *dev;
2782 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2783 	int ret = 0;
2784 	char *s_uuid = NULL;
2785 
2786 	di_args = memdup_user(arg, sizeof(*di_args));
2787 	if (IS_ERR(di_args))
2788 		return PTR_ERR(di_args);
2789 
2790 	if (!btrfs_is_empty_uuid(di_args->uuid))
2791 		s_uuid = di_args->uuid;
2792 
2793 	mutex_lock(&fs_devices->device_list_mutex);
2794 	dev = btrfs_find_device(fs_info, di_args->devid, s_uuid, NULL);
2795 
2796 	if (!dev) {
2797 		ret = -ENODEV;
2798 		goto out;
2799 	}
2800 
2801 	di_args->devid = dev->devid;
2802 	di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2803 	di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2804 	memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2805 	if (dev->name) {
2806 		struct rcu_string *name;
2807 
2808 		rcu_read_lock();
2809 		name = rcu_dereference(dev->name);
2810 		strncpy(di_args->path, name->str, sizeof(di_args->path));
2811 		rcu_read_unlock();
2812 		di_args->path[sizeof(di_args->path) - 1] = 0;
2813 	} else {
2814 		di_args->path[0] = '\0';
2815 	}
2816 
2817 out:
2818 	mutex_unlock(&fs_devices->device_list_mutex);
2819 	if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2820 		ret = -EFAULT;
2821 
2822 	kfree(di_args);
2823 	return ret;
2824 }
2825 
2826 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
2827 {
2828 	struct page *page;
2829 
2830 	page = grab_cache_page(inode->i_mapping, index);
2831 	if (!page)
2832 		return ERR_PTR(-ENOMEM);
2833 
2834 	if (!PageUptodate(page)) {
2835 		int ret;
2836 
2837 		ret = btrfs_readpage(NULL, page);
2838 		if (ret)
2839 			return ERR_PTR(ret);
2840 		lock_page(page);
2841 		if (!PageUptodate(page)) {
2842 			unlock_page(page);
2843 			put_page(page);
2844 			return ERR_PTR(-EIO);
2845 		}
2846 		if (page->mapping != inode->i_mapping) {
2847 			unlock_page(page);
2848 			put_page(page);
2849 			return ERR_PTR(-EAGAIN);
2850 		}
2851 	}
2852 
2853 	return page;
2854 }
2855 
2856 static int gather_extent_pages(struct inode *inode, struct page **pages,
2857 			       int num_pages, u64 off)
2858 {
2859 	int i;
2860 	pgoff_t index = off >> PAGE_SHIFT;
2861 
2862 	for (i = 0; i < num_pages; i++) {
2863 again:
2864 		pages[i] = extent_same_get_page(inode, index + i);
2865 		if (IS_ERR(pages[i])) {
2866 			int err = PTR_ERR(pages[i]);
2867 
2868 			if (err == -EAGAIN)
2869 				goto again;
2870 			pages[i] = NULL;
2871 			return err;
2872 		}
2873 	}
2874 	return 0;
2875 }
2876 
2877 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
2878 			     bool retry_range_locking)
2879 {
2880 	/*
2881 	 * Do any pending delalloc/csum calculations on inode, one way or
2882 	 * another, and lock file content.
2883 	 * The locking order is:
2884 	 *
2885 	 *   1) pages
2886 	 *   2) range in the inode's io tree
2887 	 */
2888 	while (1) {
2889 		struct btrfs_ordered_extent *ordered;
2890 		lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2891 		ordered = btrfs_lookup_first_ordered_extent(inode,
2892 							    off + len - 1);
2893 		if ((!ordered ||
2894 		     ordered->file_offset + ordered->len <= off ||
2895 		     ordered->file_offset >= off + len) &&
2896 		    !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2897 				    off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2898 			if (ordered)
2899 				btrfs_put_ordered_extent(ordered);
2900 			break;
2901 		}
2902 		unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2903 		if (ordered)
2904 			btrfs_put_ordered_extent(ordered);
2905 		if (!retry_range_locking)
2906 			return -EAGAIN;
2907 		btrfs_wait_ordered_range(inode, off, len);
2908 	}
2909 	return 0;
2910 }
2911 
2912 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
2913 {
2914 	inode_unlock(inode1);
2915 	inode_unlock(inode2);
2916 }
2917 
2918 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
2919 {
2920 	if (inode1 < inode2)
2921 		swap(inode1, inode2);
2922 
2923 	inode_lock_nested(inode1, I_MUTEX_PARENT);
2924 	inode_lock_nested(inode2, I_MUTEX_CHILD);
2925 }
2926 
2927 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
2928 				      struct inode *inode2, u64 loff2, u64 len)
2929 {
2930 	unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2931 	unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2932 }
2933 
2934 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
2935 				    struct inode *inode2, u64 loff2, u64 len,
2936 				    bool retry_range_locking)
2937 {
2938 	int ret;
2939 
2940 	if (inode1 < inode2) {
2941 		swap(inode1, inode2);
2942 		swap(loff1, loff2);
2943 	}
2944 	ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
2945 	if (ret)
2946 		return ret;
2947 	ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
2948 	if (ret)
2949 		unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
2950 			      loff1 + len - 1);
2951 	return ret;
2952 }
2953 
2954 struct cmp_pages {
2955 	int		num_pages;
2956 	struct page	**src_pages;
2957 	struct page	**dst_pages;
2958 };
2959 
2960 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
2961 {
2962 	int i;
2963 	struct page *pg;
2964 
2965 	for (i = 0; i < cmp->num_pages; i++) {
2966 		pg = cmp->src_pages[i];
2967 		if (pg) {
2968 			unlock_page(pg);
2969 			put_page(pg);
2970 		}
2971 		pg = cmp->dst_pages[i];
2972 		if (pg) {
2973 			unlock_page(pg);
2974 			put_page(pg);
2975 		}
2976 	}
2977 	kfree(cmp->src_pages);
2978 	kfree(cmp->dst_pages);
2979 }
2980 
2981 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
2982 				  struct inode *dst, u64 dst_loff,
2983 				  u64 len, struct cmp_pages *cmp)
2984 {
2985 	int ret;
2986 	int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT;
2987 	struct page **src_pgarr, **dst_pgarr;
2988 
2989 	/*
2990 	 * We must gather up all the pages before we initiate our
2991 	 * extent locking. We use an array for the page pointers. Size
2992 	 * of the array is bounded by len, which is in turn bounded by
2993 	 * BTRFS_MAX_DEDUPE_LEN.
2994 	 */
2995 	src_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
2996 	dst_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
2997 	if (!src_pgarr || !dst_pgarr) {
2998 		kfree(src_pgarr);
2999 		kfree(dst_pgarr);
3000 		return -ENOMEM;
3001 	}
3002 	cmp->num_pages = num_pages;
3003 	cmp->src_pages = src_pgarr;
3004 	cmp->dst_pages = dst_pgarr;
3005 
3006 	/*
3007 	 * If deduping ranges in the same inode, locking rules make it mandatory
3008 	 * to always lock pages in ascending order to avoid deadlocks with
3009 	 * concurrent tasks (such as starting writeback/delalloc).
3010 	 */
3011 	if (src == dst && dst_loff < loff) {
3012 		swap(src_pgarr, dst_pgarr);
3013 		swap(loff, dst_loff);
3014 	}
3015 
3016 	ret = gather_extent_pages(src, src_pgarr, cmp->num_pages, loff);
3017 	if (ret)
3018 		goto out;
3019 
3020 	ret = gather_extent_pages(dst, dst_pgarr, cmp->num_pages, dst_loff);
3021 
3022 out:
3023 	if (ret)
3024 		btrfs_cmp_data_free(cmp);
3025 	return ret;
3026 }
3027 
3028 static int btrfs_cmp_data(u64 len, struct cmp_pages *cmp)
3029 {
3030 	int ret = 0;
3031 	int i;
3032 	struct page *src_page, *dst_page;
3033 	unsigned int cmp_len = PAGE_SIZE;
3034 	void *addr, *dst_addr;
3035 
3036 	i = 0;
3037 	while (len) {
3038 		if (len < PAGE_SIZE)
3039 			cmp_len = len;
3040 
3041 		BUG_ON(i >= cmp->num_pages);
3042 
3043 		src_page = cmp->src_pages[i];
3044 		dst_page = cmp->dst_pages[i];
3045 		ASSERT(PageLocked(src_page));
3046 		ASSERT(PageLocked(dst_page));
3047 
3048 		addr = kmap_atomic(src_page);
3049 		dst_addr = kmap_atomic(dst_page);
3050 
3051 		flush_dcache_page(src_page);
3052 		flush_dcache_page(dst_page);
3053 
3054 		if (memcmp(addr, dst_addr, cmp_len))
3055 			ret = -EBADE;
3056 
3057 		kunmap_atomic(addr);
3058 		kunmap_atomic(dst_addr);
3059 
3060 		if (ret)
3061 			break;
3062 
3063 		len -= cmp_len;
3064 		i++;
3065 	}
3066 
3067 	return ret;
3068 }
3069 
3070 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3071 				     u64 olen)
3072 {
3073 	u64 len = *plen;
3074 	u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3075 
3076 	if (off + olen > inode->i_size || off + olen < off)
3077 		return -EINVAL;
3078 
3079 	/* if we extend to eof, continue to block boundary */
3080 	if (off + len == inode->i_size)
3081 		*plen = len = ALIGN(inode->i_size, bs) - off;
3082 
3083 	/* Check that we are block aligned - btrfs_clone() requires this */
3084 	if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3085 		return -EINVAL;
3086 
3087 	return 0;
3088 }
3089 
3090 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3091 			     struct inode *dst, u64 dst_loff)
3092 {
3093 	int ret;
3094 	u64 len = olen;
3095 	struct cmp_pages cmp;
3096 	bool same_inode = (src == dst);
3097 	u64 same_lock_start = 0;
3098 	u64 same_lock_len = 0;
3099 
3100 	if (len == 0)
3101 		return 0;
3102 
3103 	if (same_inode)
3104 		inode_lock(src);
3105 	else
3106 		btrfs_double_inode_lock(src, dst);
3107 
3108 	ret = extent_same_check_offsets(src, loff, &len, olen);
3109 	if (ret)
3110 		goto out_unlock;
3111 
3112 	ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3113 	if (ret)
3114 		goto out_unlock;
3115 
3116 	if (same_inode) {
3117 		/*
3118 		 * Single inode case wants the same checks, except we
3119 		 * don't want our length pushed out past i_size as
3120 		 * comparing that data range makes no sense.
3121 		 *
3122 		 * extent_same_check_offsets() will do this for an
3123 		 * unaligned length at i_size, so catch it here and
3124 		 * reject the request.
3125 		 *
3126 		 * This effectively means we require aligned extents
3127 		 * for the single-inode case, whereas the other cases
3128 		 * allow an unaligned length so long as it ends at
3129 		 * i_size.
3130 		 */
3131 		if (len != olen) {
3132 			ret = -EINVAL;
3133 			goto out_unlock;
3134 		}
3135 
3136 		/* Check for overlapping ranges */
3137 		if (dst_loff + len > loff && dst_loff < loff + len) {
3138 			ret = -EINVAL;
3139 			goto out_unlock;
3140 		}
3141 
3142 		same_lock_start = min_t(u64, loff, dst_loff);
3143 		same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3144 	}
3145 
3146 	/* don't make the dst file partly checksummed */
3147 	if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3148 	    (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3149 		ret = -EINVAL;
3150 		goto out_unlock;
3151 	}
3152 
3153 again:
3154 	ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp);
3155 	if (ret)
3156 		goto out_unlock;
3157 
3158 	if (same_inode)
3159 		ret = lock_extent_range(src, same_lock_start, same_lock_len,
3160 					false);
3161 	else
3162 		ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3163 					       false);
3164 	/*
3165 	 * If one of the inodes has dirty pages in the respective range or
3166 	 * ordered extents, we need to flush dellaloc and wait for all ordered
3167 	 * extents in the range. We must unlock the pages and the ranges in the
3168 	 * io trees to avoid deadlocks when flushing delalloc (requires locking
3169 	 * pages) and when waiting for ordered extents to complete (they require
3170 	 * range locking).
3171 	 */
3172 	if (ret == -EAGAIN) {
3173 		/*
3174 		 * Ranges in the io trees already unlocked. Now unlock all
3175 		 * pages before waiting for all IO to complete.
3176 		 */
3177 		btrfs_cmp_data_free(&cmp);
3178 		if (same_inode) {
3179 			btrfs_wait_ordered_range(src, same_lock_start,
3180 						 same_lock_len);
3181 		} else {
3182 			btrfs_wait_ordered_range(src, loff, len);
3183 			btrfs_wait_ordered_range(dst, dst_loff, len);
3184 		}
3185 		goto again;
3186 	}
3187 	ASSERT(ret == 0);
3188 	if (WARN_ON(ret)) {
3189 		/* ranges in the io trees already unlocked */
3190 		btrfs_cmp_data_free(&cmp);
3191 		return ret;
3192 	}
3193 
3194 	/* pass original length for comparison so we stay within i_size */
3195 	ret = btrfs_cmp_data(olen, &cmp);
3196 	if (ret == 0)
3197 		ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3198 
3199 	if (same_inode)
3200 		unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3201 			      same_lock_start + same_lock_len - 1);
3202 	else
3203 		btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3204 
3205 	btrfs_cmp_data_free(&cmp);
3206 out_unlock:
3207 	if (same_inode)
3208 		inode_unlock(src);
3209 	else
3210 		btrfs_double_inode_unlock(src, dst);
3211 
3212 	return ret;
3213 }
3214 
3215 #define BTRFS_MAX_DEDUPE_LEN	SZ_16M
3216 
3217 ssize_t btrfs_dedupe_file_range(struct file *src_file, u64 loff, u64 olen,
3218 				struct file *dst_file, u64 dst_loff)
3219 {
3220 	struct inode *src = file_inode(src_file);
3221 	struct inode *dst = file_inode(dst_file);
3222 	u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3223 	ssize_t res;
3224 
3225 	if (olen > BTRFS_MAX_DEDUPE_LEN)
3226 		olen = BTRFS_MAX_DEDUPE_LEN;
3227 
3228 	if (WARN_ON_ONCE(bs < PAGE_SIZE)) {
3229 		/*
3230 		 * Btrfs does not support blocksize < page_size. As a
3231 		 * result, btrfs_cmp_data() won't correctly handle
3232 		 * this situation without an update.
3233 		 */
3234 		return -EINVAL;
3235 	}
3236 
3237 	res = btrfs_extent_same(src, loff, olen, dst, dst_loff);
3238 	if (res)
3239 		return res;
3240 	return olen;
3241 }
3242 
3243 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3244 				     struct inode *inode,
3245 				     u64 endoff,
3246 				     const u64 destoff,
3247 				     const u64 olen,
3248 				     int no_time_update)
3249 {
3250 	struct btrfs_root *root = BTRFS_I(inode)->root;
3251 	int ret;
3252 
3253 	inode_inc_iversion(inode);
3254 	if (!no_time_update)
3255 		inode->i_mtime = inode->i_ctime = current_time(inode);
3256 	/*
3257 	 * We round up to the block size at eof when determining which
3258 	 * extents to clone above, but shouldn't round up the file size.
3259 	 */
3260 	if (endoff > destoff + olen)
3261 		endoff = destoff + olen;
3262 	if (endoff > inode->i_size)
3263 		btrfs_i_size_write(BTRFS_I(inode), endoff);
3264 
3265 	ret = btrfs_update_inode(trans, root, inode);
3266 	if (ret) {
3267 		btrfs_abort_transaction(trans, ret);
3268 		btrfs_end_transaction(trans);
3269 		goto out;
3270 	}
3271 	ret = btrfs_end_transaction(trans);
3272 out:
3273 	return ret;
3274 }
3275 
3276 static void clone_update_extent_map(struct btrfs_inode *inode,
3277 				    const struct btrfs_trans_handle *trans,
3278 				    const struct btrfs_path *path,
3279 				    const u64 hole_offset,
3280 				    const u64 hole_len)
3281 {
3282 	struct extent_map_tree *em_tree = &inode->extent_tree;
3283 	struct extent_map *em;
3284 	int ret;
3285 
3286 	em = alloc_extent_map();
3287 	if (!em) {
3288 		set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3289 		return;
3290 	}
3291 
3292 	if (path) {
3293 		struct btrfs_file_extent_item *fi;
3294 
3295 		fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3296 				    struct btrfs_file_extent_item);
3297 		btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3298 		em->generation = -1;
3299 		if (btrfs_file_extent_type(path->nodes[0], fi) ==
3300 		    BTRFS_FILE_EXTENT_INLINE)
3301 			set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3302 					&inode->runtime_flags);
3303 	} else {
3304 		em->start = hole_offset;
3305 		em->len = hole_len;
3306 		em->ram_bytes = em->len;
3307 		em->orig_start = hole_offset;
3308 		em->block_start = EXTENT_MAP_HOLE;
3309 		em->block_len = 0;
3310 		em->orig_block_len = 0;
3311 		em->compress_type = BTRFS_COMPRESS_NONE;
3312 		em->generation = trans->transid;
3313 	}
3314 
3315 	while (1) {
3316 		write_lock(&em_tree->lock);
3317 		ret = add_extent_mapping(em_tree, em, 1);
3318 		write_unlock(&em_tree->lock);
3319 		if (ret != -EEXIST) {
3320 			free_extent_map(em);
3321 			break;
3322 		}
3323 		btrfs_drop_extent_cache(inode, em->start,
3324 					em->start + em->len - 1, 0);
3325 	}
3326 
3327 	if (ret)
3328 		set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3329 }
3330 
3331 /*
3332  * Make sure we do not end up inserting an inline extent into a file that has
3333  * already other (non-inline) extents. If a file has an inline extent it can
3334  * not have any other extents and the (single) inline extent must start at the
3335  * file offset 0. Failing to respect these rules will lead to file corruption,
3336  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3337  *
3338  * We can have extents that have been already written to disk or we can have
3339  * dirty ranges still in delalloc, in which case the extent maps and items are
3340  * created only when we run delalloc, and the delalloc ranges might fall outside
3341  * the range we are currently locking in the inode's io tree. So we check the
3342  * inode's i_size because of that (i_size updates are done while holding the
3343  * i_mutex, which we are holding here).
3344  * We also check to see if the inode has a size not greater than "datal" but has
3345  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3346  * protected against such concurrent fallocate calls by the i_mutex).
3347  *
3348  * If the file has no extents but a size greater than datal, do not allow the
3349  * copy because we would need turn the inline extent into a non-inline one (even
3350  * with NO_HOLES enabled). If we find our destination inode only has one inline
3351  * extent, just overwrite it with the source inline extent if its size is less
3352  * than the source extent's size, or we could copy the source inline extent's
3353  * data into the destination inode's inline extent if the later is greater then
3354  * the former.
3355  */
3356 static int clone_copy_inline_extent(struct inode *dst,
3357 				    struct btrfs_trans_handle *trans,
3358 				    struct btrfs_path *path,
3359 				    struct btrfs_key *new_key,
3360 				    const u64 drop_start,
3361 				    const u64 datal,
3362 				    const u64 skip,
3363 				    const u64 size,
3364 				    char *inline_data)
3365 {
3366 	struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb);
3367 	struct btrfs_root *root = BTRFS_I(dst)->root;
3368 	const u64 aligned_end = ALIGN(new_key->offset + datal,
3369 				      fs_info->sectorsize);
3370 	int ret;
3371 	struct btrfs_key key;
3372 
3373 	if (new_key->offset > 0)
3374 		return -EOPNOTSUPP;
3375 
3376 	key.objectid = btrfs_ino(BTRFS_I(dst));
3377 	key.type = BTRFS_EXTENT_DATA_KEY;
3378 	key.offset = 0;
3379 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3380 	if (ret < 0) {
3381 		return ret;
3382 	} else if (ret > 0) {
3383 		if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3384 			ret = btrfs_next_leaf(root, path);
3385 			if (ret < 0)
3386 				return ret;
3387 			else if (ret > 0)
3388 				goto copy_inline_extent;
3389 		}
3390 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3391 		if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3392 		    key.type == BTRFS_EXTENT_DATA_KEY) {
3393 			ASSERT(key.offset > 0);
3394 			return -EOPNOTSUPP;
3395 		}
3396 	} else if (i_size_read(dst) <= datal) {
3397 		struct btrfs_file_extent_item *ei;
3398 		u64 ext_len;
3399 
3400 		/*
3401 		 * If the file size is <= datal, make sure there are no other
3402 		 * extents following (can happen do to an fallocate call with
3403 		 * the flag FALLOC_FL_KEEP_SIZE).
3404 		 */
3405 		ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3406 				    struct btrfs_file_extent_item);
3407 		/*
3408 		 * If it's an inline extent, it can not have other extents
3409 		 * following it.
3410 		 */
3411 		if (btrfs_file_extent_type(path->nodes[0], ei) ==
3412 		    BTRFS_FILE_EXTENT_INLINE)
3413 			goto copy_inline_extent;
3414 
3415 		ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3416 		if (ext_len > aligned_end)
3417 			return -EOPNOTSUPP;
3418 
3419 		ret = btrfs_next_item(root, path);
3420 		if (ret < 0) {
3421 			return ret;
3422 		} else if (ret == 0) {
3423 			btrfs_item_key_to_cpu(path->nodes[0], &key,
3424 					      path->slots[0]);
3425 			if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3426 			    key.type == BTRFS_EXTENT_DATA_KEY)
3427 				return -EOPNOTSUPP;
3428 		}
3429 	}
3430 
3431 copy_inline_extent:
3432 	/*
3433 	 * We have no extent items, or we have an extent at offset 0 which may
3434 	 * or may not be inlined. All these cases are dealt the same way.
3435 	 */
3436 	if (i_size_read(dst) > datal) {
3437 		/*
3438 		 * If the destination inode has an inline extent...
3439 		 * This would require copying the data from the source inline
3440 		 * extent into the beginning of the destination's inline extent.
3441 		 * But this is really complex, both extents can be compressed
3442 		 * or just one of them, which would require decompressing and
3443 		 * re-compressing data (which could increase the new compressed
3444 		 * size, not allowing the compressed data to fit anymore in an
3445 		 * inline extent).
3446 		 * So just don't support this case for now (it should be rare,
3447 		 * we are not really saving space when cloning inline extents).
3448 		 */
3449 		return -EOPNOTSUPP;
3450 	}
3451 
3452 	btrfs_release_path(path);
3453 	ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3454 	if (ret)
3455 		return ret;
3456 	ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3457 	if (ret)
3458 		return ret;
3459 
3460 	if (skip) {
3461 		const u32 start = btrfs_file_extent_calc_inline_size(0);
3462 
3463 		memmove(inline_data + start, inline_data + start + skip, datal);
3464 	}
3465 
3466 	write_extent_buffer(path->nodes[0], inline_data,
3467 			    btrfs_item_ptr_offset(path->nodes[0],
3468 						  path->slots[0]),
3469 			    size);
3470 	inode_add_bytes(dst, datal);
3471 
3472 	return 0;
3473 }
3474 
3475 /**
3476  * btrfs_clone() - clone a range from inode file to another
3477  *
3478  * @src: Inode to clone from
3479  * @inode: Inode to clone to
3480  * @off: Offset within source to start clone from
3481  * @olen: Original length, passed by user, of range to clone
3482  * @olen_aligned: Block-aligned value of olen
3483  * @destoff: Offset within @inode to start clone
3484  * @no_time_update: Whether to update mtime/ctime on the target inode
3485  */
3486 static int btrfs_clone(struct inode *src, struct inode *inode,
3487 		       const u64 off, const u64 olen, const u64 olen_aligned,
3488 		       const u64 destoff, int no_time_update)
3489 {
3490 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3491 	struct btrfs_root *root = BTRFS_I(inode)->root;
3492 	struct btrfs_path *path = NULL;
3493 	struct extent_buffer *leaf;
3494 	struct btrfs_trans_handle *trans;
3495 	char *buf = NULL;
3496 	struct btrfs_key key;
3497 	u32 nritems;
3498 	int slot;
3499 	int ret;
3500 	const u64 len = olen_aligned;
3501 	u64 last_dest_end = destoff;
3502 
3503 	ret = -ENOMEM;
3504 	buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
3505 	if (!buf)
3506 		return ret;
3507 
3508 	path = btrfs_alloc_path();
3509 	if (!path) {
3510 		kvfree(buf);
3511 		return ret;
3512 	}
3513 
3514 	path->reada = READA_FORWARD;
3515 	/* clone data */
3516 	key.objectid = btrfs_ino(BTRFS_I(src));
3517 	key.type = BTRFS_EXTENT_DATA_KEY;
3518 	key.offset = off;
3519 
3520 	while (1) {
3521 		u64 next_key_min_offset = key.offset + 1;
3522 
3523 		/*
3524 		 * note the key will change type as we walk through the
3525 		 * tree.
3526 		 */
3527 		path->leave_spinning = 1;
3528 		ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3529 				0, 0);
3530 		if (ret < 0)
3531 			goto out;
3532 		/*
3533 		 * First search, if no extent item that starts at offset off was
3534 		 * found but the previous item is an extent item, it's possible
3535 		 * it might overlap our target range, therefore process it.
3536 		 */
3537 		if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3538 			btrfs_item_key_to_cpu(path->nodes[0], &key,
3539 					      path->slots[0] - 1);
3540 			if (key.type == BTRFS_EXTENT_DATA_KEY)
3541 				path->slots[0]--;
3542 		}
3543 
3544 		nritems = btrfs_header_nritems(path->nodes[0]);
3545 process_slot:
3546 		if (path->slots[0] >= nritems) {
3547 			ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3548 			if (ret < 0)
3549 				goto out;
3550 			if (ret > 0)
3551 				break;
3552 			nritems = btrfs_header_nritems(path->nodes[0]);
3553 		}
3554 		leaf = path->nodes[0];
3555 		slot = path->slots[0];
3556 
3557 		btrfs_item_key_to_cpu(leaf, &key, slot);
3558 		if (key.type > BTRFS_EXTENT_DATA_KEY ||
3559 		    key.objectid != btrfs_ino(BTRFS_I(src)))
3560 			break;
3561 
3562 		if (key.type == BTRFS_EXTENT_DATA_KEY) {
3563 			struct btrfs_file_extent_item *extent;
3564 			int type;
3565 			u32 size;
3566 			struct btrfs_key new_key;
3567 			u64 disko = 0, diskl = 0;
3568 			u64 datao = 0, datal = 0;
3569 			u8 comp;
3570 			u64 drop_start;
3571 
3572 			extent = btrfs_item_ptr(leaf, slot,
3573 						struct btrfs_file_extent_item);
3574 			comp = btrfs_file_extent_compression(leaf, extent);
3575 			type = btrfs_file_extent_type(leaf, extent);
3576 			if (type == BTRFS_FILE_EXTENT_REG ||
3577 			    type == BTRFS_FILE_EXTENT_PREALLOC) {
3578 				disko = btrfs_file_extent_disk_bytenr(leaf,
3579 								      extent);
3580 				diskl = btrfs_file_extent_disk_num_bytes(leaf,
3581 								 extent);
3582 				datao = btrfs_file_extent_offset(leaf, extent);
3583 				datal = btrfs_file_extent_num_bytes(leaf,
3584 								    extent);
3585 			} else if (type == BTRFS_FILE_EXTENT_INLINE) {
3586 				/* take upper bound, may be compressed */
3587 				datal = btrfs_file_extent_ram_bytes(leaf,
3588 								    extent);
3589 			}
3590 
3591 			/*
3592 			 * The first search might have left us at an extent
3593 			 * item that ends before our target range's start, can
3594 			 * happen if we have holes and NO_HOLES feature enabled.
3595 			 */
3596 			if (key.offset + datal <= off) {
3597 				path->slots[0]++;
3598 				goto process_slot;
3599 			} else if (key.offset >= off + len) {
3600 				break;
3601 			}
3602 			next_key_min_offset = key.offset + datal;
3603 			size = btrfs_item_size_nr(leaf, slot);
3604 			read_extent_buffer(leaf, buf,
3605 					   btrfs_item_ptr_offset(leaf, slot),
3606 					   size);
3607 
3608 			btrfs_release_path(path);
3609 			path->leave_spinning = 0;
3610 
3611 			memcpy(&new_key, &key, sizeof(new_key));
3612 			new_key.objectid = btrfs_ino(BTRFS_I(inode));
3613 			if (off <= key.offset)
3614 				new_key.offset = key.offset + destoff - off;
3615 			else
3616 				new_key.offset = destoff;
3617 
3618 			/*
3619 			 * Deal with a hole that doesn't have an extent item
3620 			 * that represents it (NO_HOLES feature enabled).
3621 			 * This hole is either in the middle of the cloning
3622 			 * range or at the beginning (fully overlaps it or
3623 			 * partially overlaps it).
3624 			 */
3625 			if (new_key.offset != last_dest_end)
3626 				drop_start = last_dest_end;
3627 			else
3628 				drop_start = new_key.offset;
3629 
3630 			/*
3631 			 * 1 - adjusting old extent (we may have to split it)
3632 			 * 1 - add new extent
3633 			 * 1 - inode update
3634 			 */
3635 			trans = btrfs_start_transaction(root, 3);
3636 			if (IS_ERR(trans)) {
3637 				ret = PTR_ERR(trans);
3638 				goto out;
3639 			}
3640 
3641 			if (type == BTRFS_FILE_EXTENT_REG ||
3642 			    type == BTRFS_FILE_EXTENT_PREALLOC) {
3643 				/*
3644 				 *    a  | --- range to clone ---|  b
3645 				 * | ------------- extent ------------- |
3646 				 */
3647 
3648 				/* subtract range b */
3649 				if (key.offset + datal > off + len)
3650 					datal = off + len - key.offset;
3651 
3652 				/* subtract range a */
3653 				if (off > key.offset) {
3654 					datao += off - key.offset;
3655 					datal -= off - key.offset;
3656 				}
3657 
3658 				ret = btrfs_drop_extents(trans, root, inode,
3659 							 drop_start,
3660 							 new_key.offset + datal,
3661 							 1);
3662 				if (ret) {
3663 					if (ret != -EOPNOTSUPP)
3664 						btrfs_abort_transaction(trans,
3665 									ret);
3666 					btrfs_end_transaction(trans);
3667 					goto out;
3668 				}
3669 
3670 				ret = btrfs_insert_empty_item(trans, root, path,
3671 							      &new_key, size);
3672 				if (ret) {
3673 					btrfs_abort_transaction(trans, ret);
3674 					btrfs_end_transaction(trans);
3675 					goto out;
3676 				}
3677 
3678 				leaf = path->nodes[0];
3679 				slot = path->slots[0];
3680 				write_extent_buffer(leaf, buf,
3681 					    btrfs_item_ptr_offset(leaf, slot),
3682 					    size);
3683 
3684 				extent = btrfs_item_ptr(leaf, slot,
3685 						struct btrfs_file_extent_item);
3686 
3687 				/* disko == 0 means it's a hole */
3688 				if (!disko)
3689 					datao = 0;
3690 
3691 				btrfs_set_file_extent_offset(leaf, extent,
3692 							     datao);
3693 				btrfs_set_file_extent_num_bytes(leaf, extent,
3694 								datal);
3695 
3696 				if (disko) {
3697 					inode_add_bytes(inode, datal);
3698 					ret = btrfs_inc_extent_ref(trans,
3699 							root,
3700 							disko, diskl, 0,
3701 							root->root_key.objectid,
3702 							btrfs_ino(BTRFS_I(inode)),
3703 							new_key.offset - datao);
3704 					if (ret) {
3705 						btrfs_abort_transaction(trans,
3706 									ret);
3707 						btrfs_end_transaction(trans);
3708 						goto out;
3709 
3710 					}
3711 				}
3712 			} else if (type == BTRFS_FILE_EXTENT_INLINE) {
3713 				u64 skip = 0;
3714 				u64 trim = 0;
3715 
3716 				if (off > key.offset) {
3717 					skip = off - key.offset;
3718 					new_key.offset += skip;
3719 				}
3720 
3721 				if (key.offset + datal > off + len)
3722 					trim = key.offset + datal - (off + len);
3723 
3724 				if (comp && (skip || trim)) {
3725 					ret = -EINVAL;
3726 					btrfs_end_transaction(trans);
3727 					goto out;
3728 				}
3729 				size -= skip + trim;
3730 				datal -= skip + trim;
3731 
3732 				ret = clone_copy_inline_extent(inode,
3733 							       trans, path,
3734 							       &new_key,
3735 							       drop_start,
3736 							       datal,
3737 							       skip, size, buf);
3738 				if (ret) {
3739 					if (ret != -EOPNOTSUPP)
3740 						btrfs_abort_transaction(trans,
3741 									ret);
3742 					btrfs_end_transaction(trans);
3743 					goto out;
3744 				}
3745 				leaf = path->nodes[0];
3746 				slot = path->slots[0];
3747 			}
3748 
3749 			/* If we have an implicit hole (NO_HOLES feature). */
3750 			if (drop_start < new_key.offset)
3751 				clone_update_extent_map(BTRFS_I(inode), trans,
3752 						NULL, drop_start,
3753 						new_key.offset - drop_start);
3754 
3755 			clone_update_extent_map(BTRFS_I(inode), trans,
3756 					path, 0, 0);
3757 
3758 			btrfs_mark_buffer_dirty(leaf);
3759 			btrfs_release_path(path);
3760 
3761 			last_dest_end = ALIGN(new_key.offset + datal,
3762 					      fs_info->sectorsize);
3763 			ret = clone_finish_inode_update(trans, inode,
3764 							last_dest_end,
3765 							destoff, olen,
3766 							no_time_update);
3767 			if (ret)
3768 				goto out;
3769 			if (new_key.offset + datal >= destoff + len)
3770 				break;
3771 		}
3772 		btrfs_release_path(path);
3773 		key.offset = next_key_min_offset;
3774 
3775 		if (fatal_signal_pending(current)) {
3776 			ret = -EINTR;
3777 			goto out;
3778 		}
3779 	}
3780 	ret = 0;
3781 
3782 	if (last_dest_end < destoff + len) {
3783 		/*
3784 		 * We have an implicit hole (NO_HOLES feature is enabled) that
3785 		 * fully or partially overlaps our cloning range at its end.
3786 		 */
3787 		btrfs_release_path(path);
3788 
3789 		/*
3790 		 * 1 - remove extent(s)
3791 		 * 1 - inode update
3792 		 */
3793 		trans = btrfs_start_transaction(root, 2);
3794 		if (IS_ERR(trans)) {
3795 			ret = PTR_ERR(trans);
3796 			goto out;
3797 		}
3798 		ret = btrfs_drop_extents(trans, root, inode,
3799 					 last_dest_end, destoff + len, 1);
3800 		if (ret) {
3801 			if (ret != -EOPNOTSUPP)
3802 				btrfs_abort_transaction(trans, ret);
3803 			btrfs_end_transaction(trans);
3804 			goto out;
3805 		}
3806 		clone_update_extent_map(BTRFS_I(inode), trans, NULL,
3807 				last_dest_end,
3808 				destoff + len - last_dest_end);
3809 		ret = clone_finish_inode_update(trans, inode, destoff + len,
3810 						destoff, olen, no_time_update);
3811 	}
3812 
3813 out:
3814 	btrfs_free_path(path);
3815 	kvfree(buf);
3816 	return ret;
3817 }
3818 
3819 static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
3820 					u64 off, u64 olen, u64 destoff)
3821 {
3822 	struct inode *inode = file_inode(file);
3823 	struct inode *src = file_inode(file_src);
3824 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3825 	struct btrfs_root *root = BTRFS_I(inode)->root;
3826 	int ret;
3827 	u64 len = olen;
3828 	u64 bs = fs_info->sb->s_blocksize;
3829 	int same_inode = src == inode;
3830 
3831 	/*
3832 	 * TODO:
3833 	 * - split compressed inline extents.  annoying: we need to
3834 	 *   decompress into destination's address_space (the file offset
3835 	 *   may change, so source mapping won't do), then recompress (or
3836 	 *   otherwise reinsert) a subrange.
3837 	 *
3838 	 * - split destination inode's inline extents.  The inline extents can
3839 	 *   be either compressed or non-compressed.
3840 	 */
3841 
3842 	if (btrfs_root_readonly(root))
3843 		return -EROFS;
3844 
3845 	if (file_src->f_path.mnt != file->f_path.mnt ||
3846 	    src->i_sb != inode->i_sb)
3847 		return -EXDEV;
3848 
3849 	/* don't make the dst file partly checksummed */
3850 	if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3851 	    (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3852 		return -EINVAL;
3853 
3854 	if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3855 		return -EISDIR;
3856 
3857 	if (!same_inode) {
3858 		btrfs_double_inode_lock(src, inode);
3859 	} else {
3860 		inode_lock(src);
3861 	}
3862 
3863 	/* determine range to clone */
3864 	ret = -EINVAL;
3865 	if (off + len > src->i_size || off + len < off)
3866 		goto out_unlock;
3867 	if (len == 0)
3868 		olen = len = src->i_size - off;
3869 	/* if we extend to eof, continue to block boundary */
3870 	if (off + len == src->i_size)
3871 		len = ALIGN(src->i_size, bs) - off;
3872 
3873 	if (len == 0) {
3874 		ret = 0;
3875 		goto out_unlock;
3876 	}
3877 
3878 	/* verify the end result is block aligned */
3879 	if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3880 	    !IS_ALIGNED(destoff, bs))
3881 		goto out_unlock;
3882 
3883 	/* verify if ranges are overlapped within the same file */
3884 	if (same_inode) {
3885 		if (destoff + len > off && destoff < off + len)
3886 			goto out_unlock;
3887 	}
3888 
3889 	if (destoff > inode->i_size) {
3890 		ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3891 		if (ret)
3892 			goto out_unlock;
3893 	}
3894 
3895 	/*
3896 	 * Lock the target range too. Right after we replace the file extent
3897 	 * items in the fs tree (which now point to the cloned data), we might
3898 	 * have a worker replace them with extent items relative to a write
3899 	 * operation that was issued before this clone operation (i.e. confront
3900 	 * with inode.c:btrfs_finish_ordered_io).
3901 	 */
3902 	if (same_inode) {
3903 		u64 lock_start = min_t(u64, off, destoff);
3904 		u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
3905 
3906 		ret = lock_extent_range(src, lock_start, lock_len, true);
3907 	} else {
3908 		ret = btrfs_double_extent_lock(src, off, inode, destoff, len,
3909 					       true);
3910 	}
3911 	ASSERT(ret == 0);
3912 	if (WARN_ON(ret)) {
3913 		/* ranges in the io trees already unlocked */
3914 		goto out_unlock;
3915 	}
3916 
3917 	ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
3918 
3919 	if (same_inode) {
3920 		u64 lock_start = min_t(u64, off, destoff);
3921 		u64 lock_end = max_t(u64, off, destoff) + len - 1;
3922 
3923 		unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
3924 	} else {
3925 		btrfs_double_extent_unlock(src, off, inode, destoff, len);
3926 	}
3927 	/*
3928 	 * Truncate page cache pages so that future reads will see the cloned
3929 	 * data immediately and not the previous data.
3930 	 */
3931 	truncate_inode_pages_range(&inode->i_data,
3932 				round_down(destoff, PAGE_SIZE),
3933 				round_up(destoff + len, PAGE_SIZE) - 1);
3934 out_unlock:
3935 	if (!same_inode)
3936 		btrfs_double_inode_unlock(src, inode);
3937 	else
3938 		inode_unlock(src);
3939 	return ret;
3940 }
3941 
3942 int btrfs_clone_file_range(struct file *src_file, loff_t off,
3943 		struct file *dst_file, loff_t destoff, u64 len)
3944 {
3945 	return btrfs_clone_files(dst_file, src_file, off, len, destoff);
3946 }
3947 
3948 /*
3949  * there are many ways the trans_start and trans_end ioctls can lead
3950  * to deadlocks.  They should only be used by applications that
3951  * basically own the machine, and have a very in depth understanding
3952  * of all the possible deadlocks and enospc problems.
3953  */
3954 static long btrfs_ioctl_trans_start(struct file *file)
3955 {
3956 	struct inode *inode = file_inode(file);
3957 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3958 	struct btrfs_root *root = BTRFS_I(inode)->root;
3959 	struct btrfs_trans_handle *trans;
3960 	struct btrfs_file_private *private;
3961 	int ret;
3962 	static bool warned = false;
3963 
3964 	ret = -EPERM;
3965 	if (!capable(CAP_SYS_ADMIN))
3966 		goto out;
3967 
3968 	if (!warned) {
3969 		btrfs_warn(fs_info,
3970 			"Userspace transaction mechanism is considered "
3971 			"deprecated and slated to be removed in 4.17. "
3972 			"If you have a valid use case please "
3973 			"speak up on the mailing list");
3974 		WARN_ON(1);
3975 		warned = true;
3976 	}
3977 
3978 	ret = -EINPROGRESS;
3979 	private = file->private_data;
3980 	if (private && private->trans)
3981 		goto out;
3982 	if (!private) {
3983 		private = kzalloc(sizeof(struct btrfs_file_private),
3984 				  GFP_KERNEL);
3985 		if (!private)
3986 			return -ENOMEM;
3987 		file->private_data = private;
3988 	}
3989 
3990 	ret = -EROFS;
3991 	if (btrfs_root_readonly(root))
3992 		goto out;
3993 
3994 	ret = mnt_want_write_file(file);
3995 	if (ret)
3996 		goto out;
3997 
3998 	atomic_inc(&fs_info->open_ioctl_trans);
3999 
4000 	ret = -ENOMEM;
4001 	trans = btrfs_start_ioctl_transaction(root);
4002 	if (IS_ERR(trans))
4003 		goto out_drop;
4004 
4005 	private->trans = trans;
4006 	return 0;
4007 
4008 out_drop:
4009 	atomic_dec(&fs_info->open_ioctl_trans);
4010 	mnt_drop_write_file(file);
4011 out:
4012 	return ret;
4013 }
4014 
4015 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
4016 {
4017 	struct inode *inode = file_inode(file);
4018 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4019 	struct btrfs_root *root = BTRFS_I(inode)->root;
4020 	struct btrfs_root *new_root;
4021 	struct btrfs_dir_item *di;
4022 	struct btrfs_trans_handle *trans;
4023 	struct btrfs_path *path;
4024 	struct btrfs_key location;
4025 	struct btrfs_disk_key disk_key;
4026 	u64 objectid = 0;
4027 	u64 dir_id;
4028 	int ret;
4029 
4030 	if (!capable(CAP_SYS_ADMIN))
4031 		return -EPERM;
4032 
4033 	ret = mnt_want_write_file(file);
4034 	if (ret)
4035 		return ret;
4036 
4037 	if (copy_from_user(&objectid, argp, sizeof(objectid))) {
4038 		ret = -EFAULT;
4039 		goto out;
4040 	}
4041 
4042 	if (!objectid)
4043 		objectid = BTRFS_FS_TREE_OBJECTID;
4044 
4045 	location.objectid = objectid;
4046 	location.type = BTRFS_ROOT_ITEM_KEY;
4047 	location.offset = (u64)-1;
4048 
4049 	new_root = btrfs_read_fs_root_no_name(fs_info, &location);
4050 	if (IS_ERR(new_root)) {
4051 		ret = PTR_ERR(new_root);
4052 		goto out;
4053 	}
4054 	if (!is_fstree(new_root->objectid)) {
4055 		ret = -ENOENT;
4056 		goto out;
4057 	}
4058 
4059 	path = btrfs_alloc_path();
4060 	if (!path) {
4061 		ret = -ENOMEM;
4062 		goto out;
4063 	}
4064 	path->leave_spinning = 1;
4065 
4066 	trans = btrfs_start_transaction(root, 1);
4067 	if (IS_ERR(trans)) {
4068 		btrfs_free_path(path);
4069 		ret = PTR_ERR(trans);
4070 		goto out;
4071 	}
4072 
4073 	dir_id = btrfs_super_root_dir(fs_info->super_copy);
4074 	di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
4075 				   dir_id, "default", 7, 1);
4076 	if (IS_ERR_OR_NULL(di)) {
4077 		btrfs_free_path(path);
4078 		btrfs_end_transaction(trans);
4079 		btrfs_err(fs_info,
4080 			  "Umm, you don't have the default diritem, this isn't going to work");
4081 		ret = -ENOENT;
4082 		goto out;
4083 	}
4084 
4085 	btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
4086 	btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
4087 	btrfs_mark_buffer_dirty(path->nodes[0]);
4088 	btrfs_free_path(path);
4089 
4090 	btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
4091 	btrfs_end_transaction(trans);
4092 out:
4093 	mnt_drop_write_file(file);
4094 	return ret;
4095 }
4096 
4097 void btrfs_get_block_group_info(struct list_head *groups_list,
4098 				struct btrfs_ioctl_space_info *space)
4099 {
4100 	struct btrfs_block_group_cache *block_group;
4101 
4102 	space->total_bytes = 0;
4103 	space->used_bytes = 0;
4104 	space->flags = 0;
4105 	list_for_each_entry(block_group, groups_list, list) {
4106 		space->flags = block_group->flags;
4107 		space->total_bytes += block_group->key.offset;
4108 		space->used_bytes +=
4109 			btrfs_block_group_used(&block_group->item);
4110 	}
4111 }
4112 
4113 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
4114 				   void __user *arg)
4115 {
4116 	struct btrfs_ioctl_space_args space_args;
4117 	struct btrfs_ioctl_space_info space;
4118 	struct btrfs_ioctl_space_info *dest;
4119 	struct btrfs_ioctl_space_info *dest_orig;
4120 	struct btrfs_ioctl_space_info __user *user_dest;
4121 	struct btrfs_space_info *info;
4122 	static const u64 types[] = {
4123 		BTRFS_BLOCK_GROUP_DATA,
4124 		BTRFS_BLOCK_GROUP_SYSTEM,
4125 		BTRFS_BLOCK_GROUP_METADATA,
4126 		BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
4127 	};
4128 	int num_types = 4;
4129 	int alloc_size;
4130 	int ret = 0;
4131 	u64 slot_count = 0;
4132 	int i, c;
4133 
4134 	if (copy_from_user(&space_args,
4135 			   (struct btrfs_ioctl_space_args __user *)arg,
4136 			   sizeof(space_args)))
4137 		return -EFAULT;
4138 
4139 	for (i = 0; i < num_types; i++) {
4140 		struct btrfs_space_info *tmp;
4141 
4142 		info = NULL;
4143 		rcu_read_lock();
4144 		list_for_each_entry_rcu(tmp, &fs_info->space_info,
4145 					list) {
4146 			if (tmp->flags == types[i]) {
4147 				info = tmp;
4148 				break;
4149 			}
4150 		}
4151 		rcu_read_unlock();
4152 
4153 		if (!info)
4154 			continue;
4155 
4156 		down_read(&info->groups_sem);
4157 		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4158 			if (!list_empty(&info->block_groups[c]))
4159 				slot_count++;
4160 		}
4161 		up_read(&info->groups_sem);
4162 	}
4163 
4164 	/*
4165 	 * Global block reserve, exported as a space_info
4166 	 */
4167 	slot_count++;
4168 
4169 	/* space_slots == 0 means they are asking for a count */
4170 	if (space_args.space_slots == 0) {
4171 		space_args.total_spaces = slot_count;
4172 		goto out;
4173 	}
4174 
4175 	slot_count = min_t(u64, space_args.space_slots, slot_count);
4176 
4177 	alloc_size = sizeof(*dest) * slot_count;
4178 
4179 	/* we generally have at most 6 or so space infos, one for each raid
4180 	 * level.  So, a whole page should be more than enough for everyone
4181 	 */
4182 	if (alloc_size > PAGE_SIZE)
4183 		return -ENOMEM;
4184 
4185 	space_args.total_spaces = 0;
4186 	dest = kmalloc(alloc_size, GFP_KERNEL);
4187 	if (!dest)
4188 		return -ENOMEM;
4189 	dest_orig = dest;
4190 
4191 	/* now we have a buffer to copy into */
4192 	for (i = 0; i < num_types; i++) {
4193 		struct btrfs_space_info *tmp;
4194 
4195 		if (!slot_count)
4196 			break;
4197 
4198 		info = NULL;
4199 		rcu_read_lock();
4200 		list_for_each_entry_rcu(tmp, &fs_info->space_info,
4201 					list) {
4202 			if (tmp->flags == types[i]) {
4203 				info = tmp;
4204 				break;
4205 			}
4206 		}
4207 		rcu_read_unlock();
4208 
4209 		if (!info)
4210 			continue;
4211 		down_read(&info->groups_sem);
4212 		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4213 			if (!list_empty(&info->block_groups[c])) {
4214 				btrfs_get_block_group_info(
4215 					&info->block_groups[c], &space);
4216 				memcpy(dest, &space, sizeof(space));
4217 				dest++;
4218 				space_args.total_spaces++;
4219 				slot_count--;
4220 			}
4221 			if (!slot_count)
4222 				break;
4223 		}
4224 		up_read(&info->groups_sem);
4225 	}
4226 
4227 	/*
4228 	 * Add global block reserve
4229 	 */
4230 	if (slot_count) {
4231 		struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4232 
4233 		spin_lock(&block_rsv->lock);
4234 		space.total_bytes = block_rsv->size;
4235 		space.used_bytes = block_rsv->size - block_rsv->reserved;
4236 		spin_unlock(&block_rsv->lock);
4237 		space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4238 		memcpy(dest, &space, sizeof(space));
4239 		space_args.total_spaces++;
4240 	}
4241 
4242 	user_dest = (struct btrfs_ioctl_space_info __user *)
4243 		(arg + sizeof(struct btrfs_ioctl_space_args));
4244 
4245 	if (copy_to_user(user_dest, dest_orig, alloc_size))
4246 		ret = -EFAULT;
4247 
4248 	kfree(dest_orig);
4249 out:
4250 	if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4251 		ret = -EFAULT;
4252 
4253 	return ret;
4254 }
4255 
4256 /*
4257  * there are many ways the trans_start and trans_end ioctls can lead
4258  * to deadlocks.  They should only be used by applications that
4259  * basically own the machine, and have a very in depth understanding
4260  * of all the possible deadlocks and enospc problems.
4261  */
4262 long btrfs_ioctl_trans_end(struct file *file)
4263 {
4264 	struct inode *inode = file_inode(file);
4265 	struct btrfs_root *root = BTRFS_I(inode)->root;
4266 	struct btrfs_file_private *private = file->private_data;
4267 
4268 	if (!private || !private->trans)
4269 		return -EINVAL;
4270 
4271 	btrfs_end_transaction(private->trans);
4272 	private->trans = NULL;
4273 
4274 	atomic_dec(&root->fs_info->open_ioctl_trans);
4275 
4276 	mnt_drop_write_file(file);
4277 	return 0;
4278 }
4279 
4280 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4281 					    void __user *argp)
4282 {
4283 	struct btrfs_trans_handle *trans;
4284 	u64 transid;
4285 	int ret;
4286 
4287 	trans = btrfs_attach_transaction_barrier(root);
4288 	if (IS_ERR(trans)) {
4289 		if (PTR_ERR(trans) != -ENOENT)
4290 			return PTR_ERR(trans);
4291 
4292 		/* No running transaction, don't bother */
4293 		transid = root->fs_info->last_trans_committed;
4294 		goto out;
4295 	}
4296 	transid = trans->transid;
4297 	ret = btrfs_commit_transaction_async(trans, 0);
4298 	if (ret) {
4299 		btrfs_end_transaction(trans);
4300 		return ret;
4301 	}
4302 out:
4303 	if (argp)
4304 		if (copy_to_user(argp, &transid, sizeof(transid)))
4305 			return -EFAULT;
4306 	return 0;
4307 }
4308 
4309 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
4310 					   void __user *argp)
4311 {
4312 	u64 transid;
4313 
4314 	if (argp) {
4315 		if (copy_from_user(&transid, argp, sizeof(transid)))
4316 			return -EFAULT;
4317 	} else {
4318 		transid = 0;  /* current trans */
4319 	}
4320 	return btrfs_wait_for_commit(fs_info, transid);
4321 }
4322 
4323 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4324 {
4325 	struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
4326 	struct btrfs_ioctl_scrub_args *sa;
4327 	int ret;
4328 
4329 	if (!capable(CAP_SYS_ADMIN))
4330 		return -EPERM;
4331 
4332 	sa = memdup_user(arg, sizeof(*sa));
4333 	if (IS_ERR(sa))
4334 		return PTR_ERR(sa);
4335 
4336 	if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4337 		ret = mnt_want_write_file(file);
4338 		if (ret)
4339 			goto out;
4340 	}
4341 
4342 	ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
4343 			      &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4344 			      0);
4345 
4346 	if (copy_to_user(arg, sa, sizeof(*sa)))
4347 		ret = -EFAULT;
4348 
4349 	if (!(sa->flags & BTRFS_SCRUB_READONLY))
4350 		mnt_drop_write_file(file);
4351 out:
4352 	kfree(sa);
4353 	return ret;
4354 }
4355 
4356 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
4357 {
4358 	if (!capable(CAP_SYS_ADMIN))
4359 		return -EPERM;
4360 
4361 	return btrfs_scrub_cancel(fs_info);
4362 }
4363 
4364 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
4365 				       void __user *arg)
4366 {
4367 	struct btrfs_ioctl_scrub_args *sa;
4368 	int ret;
4369 
4370 	if (!capable(CAP_SYS_ADMIN))
4371 		return -EPERM;
4372 
4373 	sa = memdup_user(arg, sizeof(*sa));
4374 	if (IS_ERR(sa))
4375 		return PTR_ERR(sa);
4376 
4377 	ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
4378 
4379 	if (copy_to_user(arg, sa, sizeof(*sa)))
4380 		ret = -EFAULT;
4381 
4382 	kfree(sa);
4383 	return ret;
4384 }
4385 
4386 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
4387 				      void __user *arg)
4388 {
4389 	struct btrfs_ioctl_get_dev_stats *sa;
4390 	int ret;
4391 
4392 	sa = memdup_user(arg, sizeof(*sa));
4393 	if (IS_ERR(sa))
4394 		return PTR_ERR(sa);
4395 
4396 	if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4397 		kfree(sa);
4398 		return -EPERM;
4399 	}
4400 
4401 	ret = btrfs_get_dev_stats(fs_info, sa);
4402 
4403 	if (copy_to_user(arg, sa, sizeof(*sa)))
4404 		ret = -EFAULT;
4405 
4406 	kfree(sa);
4407 	return ret;
4408 }
4409 
4410 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
4411 				    void __user *arg)
4412 {
4413 	struct btrfs_ioctl_dev_replace_args *p;
4414 	int ret;
4415 
4416 	if (!capable(CAP_SYS_ADMIN))
4417 		return -EPERM;
4418 
4419 	p = memdup_user(arg, sizeof(*p));
4420 	if (IS_ERR(p))
4421 		return PTR_ERR(p);
4422 
4423 	switch (p->cmd) {
4424 	case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4425 		if (sb_rdonly(fs_info->sb)) {
4426 			ret = -EROFS;
4427 			goto out;
4428 		}
4429 		if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4430 			ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4431 		} else {
4432 			ret = btrfs_dev_replace_by_ioctl(fs_info, p);
4433 			clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4434 		}
4435 		break;
4436 	case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4437 		btrfs_dev_replace_status(fs_info, p);
4438 		ret = 0;
4439 		break;
4440 	case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4441 		ret = btrfs_dev_replace_cancel(fs_info, p);
4442 		break;
4443 	default:
4444 		ret = -EINVAL;
4445 		break;
4446 	}
4447 
4448 	if (copy_to_user(arg, p, sizeof(*p)))
4449 		ret = -EFAULT;
4450 out:
4451 	kfree(p);
4452 	return ret;
4453 }
4454 
4455 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4456 {
4457 	int ret = 0;
4458 	int i;
4459 	u64 rel_ptr;
4460 	int size;
4461 	struct btrfs_ioctl_ino_path_args *ipa = NULL;
4462 	struct inode_fs_paths *ipath = NULL;
4463 	struct btrfs_path *path;
4464 
4465 	if (!capable(CAP_DAC_READ_SEARCH))
4466 		return -EPERM;
4467 
4468 	path = btrfs_alloc_path();
4469 	if (!path) {
4470 		ret = -ENOMEM;
4471 		goto out;
4472 	}
4473 
4474 	ipa = memdup_user(arg, sizeof(*ipa));
4475 	if (IS_ERR(ipa)) {
4476 		ret = PTR_ERR(ipa);
4477 		ipa = NULL;
4478 		goto out;
4479 	}
4480 
4481 	size = min_t(u32, ipa->size, 4096);
4482 	ipath = init_ipath(size, root, path);
4483 	if (IS_ERR(ipath)) {
4484 		ret = PTR_ERR(ipath);
4485 		ipath = NULL;
4486 		goto out;
4487 	}
4488 
4489 	ret = paths_from_inode(ipa->inum, ipath);
4490 	if (ret < 0)
4491 		goto out;
4492 
4493 	for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4494 		rel_ptr = ipath->fspath->val[i] -
4495 			  (u64)(unsigned long)ipath->fspath->val;
4496 		ipath->fspath->val[i] = rel_ptr;
4497 	}
4498 
4499 	ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
4500 			   ipath->fspath, size);
4501 	if (ret) {
4502 		ret = -EFAULT;
4503 		goto out;
4504 	}
4505 
4506 out:
4507 	btrfs_free_path(path);
4508 	free_ipath(ipath);
4509 	kfree(ipa);
4510 
4511 	return ret;
4512 }
4513 
4514 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4515 {
4516 	struct btrfs_data_container *inodes = ctx;
4517 	const size_t c = 3 * sizeof(u64);
4518 
4519 	if (inodes->bytes_left >= c) {
4520 		inodes->bytes_left -= c;
4521 		inodes->val[inodes->elem_cnt] = inum;
4522 		inodes->val[inodes->elem_cnt + 1] = offset;
4523 		inodes->val[inodes->elem_cnt + 2] = root;
4524 		inodes->elem_cnt += 3;
4525 	} else {
4526 		inodes->bytes_missing += c - inodes->bytes_left;
4527 		inodes->bytes_left = 0;
4528 		inodes->elem_missed += 3;
4529 	}
4530 
4531 	return 0;
4532 }
4533 
4534 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
4535 					void __user *arg, int version)
4536 {
4537 	int ret = 0;
4538 	int size;
4539 	struct btrfs_ioctl_logical_ino_args *loi;
4540 	struct btrfs_data_container *inodes = NULL;
4541 	struct btrfs_path *path = NULL;
4542 	bool ignore_offset;
4543 
4544 	if (!capable(CAP_SYS_ADMIN))
4545 		return -EPERM;
4546 
4547 	loi = memdup_user(arg, sizeof(*loi));
4548 	if (IS_ERR(loi))
4549 		return PTR_ERR(loi);
4550 
4551 	if (version == 1) {
4552 		ignore_offset = false;
4553 		size = min_t(u32, loi->size, SZ_64K);
4554 	} else {
4555 		/* All reserved bits must be 0 for now */
4556 		if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
4557 			ret = -EINVAL;
4558 			goto out_loi;
4559 		}
4560 		/* Only accept flags we have defined so far */
4561 		if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
4562 			ret = -EINVAL;
4563 			goto out_loi;
4564 		}
4565 		ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
4566 		size = min_t(u32, loi->size, SZ_16M);
4567 	}
4568 
4569 	path = btrfs_alloc_path();
4570 	if (!path) {
4571 		ret = -ENOMEM;
4572 		goto out;
4573 	}
4574 
4575 	inodes = init_data_container(size);
4576 	if (IS_ERR(inodes)) {
4577 		ret = PTR_ERR(inodes);
4578 		inodes = NULL;
4579 		goto out;
4580 	}
4581 
4582 	ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
4583 					  build_ino_list, inodes, ignore_offset);
4584 	if (ret == -EINVAL)
4585 		ret = -ENOENT;
4586 	if (ret < 0)
4587 		goto out;
4588 
4589 	ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
4590 			   size);
4591 	if (ret)
4592 		ret = -EFAULT;
4593 
4594 out:
4595 	btrfs_free_path(path);
4596 	kvfree(inodes);
4597 out_loi:
4598 	kfree(loi);
4599 
4600 	return ret;
4601 }
4602 
4603 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4604 			       struct btrfs_ioctl_balance_args *bargs)
4605 {
4606 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4607 
4608 	bargs->flags = bctl->flags;
4609 
4610 	if (atomic_read(&fs_info->balance_running))
4611 		bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4612 	if (atomic_read(&fs_info->balance_pause_req))
4613 		bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4614 	if (atomic_read(&fs_info->balance_cancel_req))
4615 		bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4616 
4617 	memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4618 	memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4619 	memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4620 
4621 	if (lock) {
4622 		spin_lock(&fs_info->balance_lock);
4623 		memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4624 		spin_unlock(&fs_info->balance_lock);
4625 	} else {
4626 		memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4627 	}
4628 }
4629 
4630 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4631 {
4632 	struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4633 	struct btrfs_fs_info *fs_info = root->fs_info;
4634 	struct btrfs_ioctl_balance_args *bargs;
4635 	struct btrfs_balance_control *bctl;
4636 	bool need_unlock; /* for mut. excl. ops lock */
4637 	int ret;
4638 
4639 	if (!capable(CAP_SYS_ADMIN))
4640 		return -EPERM;
4641 
4642 	ret = mnt_want_write_file(file);
4643 	if (ret)
4644 		return ret;
4645 
4646 again:
4647 	if (!test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4648 		mutex_lock(&fs_info->volume_mutex);
4649 		mutex_lock(&fs_info->balance_mutex);
4650 		need_unlock = true;
4651 		goto locked;
4652 	}
4653 
4654 	/*
4655 	 * mut. excl. ops lock is locked.  Three possibilities:
4656 	 *   (1) some other op is running
4657 	 *   (2) balance is running
4658 	 *   (3) balance is paused -- special case (think resume)
4659 	 */
4660 	mutex_lock(&fs_info->balance_mutex);
4661 	if (fs_info->balance_ctl) {
4662 		/* this is either (2) or (3) */
4663 		if (!atomic_read(&fs_info->balance_running)) {
4664 			mutex_unlock(&fs_info->balance_mutex);
4665 			if (!mutex_trylock(&fs_info->volume_mutex))
4666 				goto again;
4667 			mutex_lock(&fs_info->balance_mutex);
4668 
4669 			if (fs_info->balance_ctl &&
4670 			    !atomic_read(&fs_info->balance_running)) {
4671 				/* this is (3) */
4672 				need_unlock = false;
4673 				goto locked;
4674 			}
4675 
4676 			mutex_unlock(&fs_info->balance_mutex);
4677 			mutex_unlock(&fs_info->volume_mutex);
4678 			goto again;
4679 		} else {
4680 			/* this is (2) */
4681 			mutex_unlock(&fs_info->balance_mutex);
4682 			ret = -EINPROGRESS;
4683 			goto out;
4684 		}
4685 	} else {
4686 		/* this is (1) */
4687 		mutex_unlock(&fs_info->balance_mutex);
4688 		ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4689 		goto out;
4690 	}
4691 
4692 locked:
4693 	BUG_ON(!test_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4694 
4695 	if (arg) {
4696 		bargs = memdup_user(arg, sizeof(*bargs));
4697 		if (IS_ERR(bargs)) {
4698 			ret = PTR_ERR(bargs);
4699 			goto out_unlock;
4700 		}
4701 
4702 		if (bargs->flags & BTRFS_BALANCE_RESUME) {
4703 			if (!fs_info->balance_ctl) {
4704 				ret = -ENOTCONN;
4705 				goto out_bargs;
4706 			}
4707 
4708 			bctl = fs_info->balance_ctl;
4709 			spin_lock(&fs_info->balance_lock);
4710 			bctl->flags |= BTRFS_BALANCE_RESUME;
4711 			spin_unlock(&fs_info->balance_lock);
4712 
4713 			goto do_balance;
4714 		}
4715 	} else {
4716 		bargs = NULL;
4717 	}
4718 
4719 	if (fs_info->balance_ctl) {
4720 		ret = -EINPROGRESS;
4721 		goto out_bargs;
4722 	}
4723 
4724 	bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4725 	if (!bctl) {
4726 		ret = -ENOMEM;
4727 		goto out_bargs;
4728 	}
4729 
4730 	bctl->fs_info = fs_info;
4731 	if (arg) {
4732 		memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4733 		memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4734 		memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4735 
4736 		bctl->flags = bargs->flags;
4737 	} else {
4738 		/* balance everything - no filters */
4739 		bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4740 	}
4741 
4742 	if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4743 		ret = -EINVAL;
4744 		goto out_bctl;
4745 	}
4746 
4747 do_balance:
4748 	/*
4749 	 * Ownership of bctl and filesystem flag BTRFS_FS_EXCL_OP
4750 	 * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4751 	 * or, if restriper was paused all the way until unmount, in
4752 	 * free_fs_info.  The flag is cleared in __cancel_balance.
4753 	 */
4754 	need_unlock = false;
4755 
4756 	ret = btrfs_balance(bctl, bargs);
4757 	bctl = NULL;
4758 
4759 	if (arg) {
4760 		if (copy_to_user(arg, bargs, sizeof(*bargs)))
4761 			ret = -EFAULT;
4762 	}
4763 
4764 out_bctl:
4765 	kfree(bctl);
4766 out_bargs:
4767 	kfree(bargs);
4768 out_unlock:
4769 	mutex_unlock(&fs_info->balance_mutex);
4770 	mutex_unlock(&fs_info->volume_mutex);
4771 	if (need_unlock)
4772 		clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4773 out:
4774 	mnt_drop_write_file(file);
4775 	return ret;
4776 }
4777 
4778 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4779 {
4780 	if (!capable(CAP_SYS_ADMIN))
4781 		return -EPERM;
4782 
4783 	switch (cmd) {
4784 	case BTRFS_BALANCE_CTL_PAUSE:
4785 		return btrfs_pause_balance(fs_info);
4786 	case BTRFS_BALANCE_CTL_CANCEL:
4787 		return btrfs_cancel_balance(fs_info);
4788 	}
4789 
4790 	return -EINVAL;
4791 }
4792 
4793 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4794 					 void __user *arg)
4795 {
4796 	struct btrfs_ioctl_balance_args *bargs;
4797 	int ret = 0;
4798 
4799 	if (!capable(CAP_SYS_ADMIN))
4800 		return -EPERM;
4801 
4802 	mutex_lock(&fs_info->balance_mutex);
4803 	if (!fs_info->balance_ctl) {
4804 		ret = -ENOTCONN;
4805 		goto out;
4806 	}
4807 
4808 	bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4809 	if (!bargs) {
4810 		ret = -ENOMEM;
4811 		goto out;
4812 	}
4813 
4814 	update_ioctl_balance_args(fs_info, 1, bargs);
4815 
4816 	if (copy_to_user(arg, bargs, sizeof(*bargs)))
4817 		ret = -EFAULT;
4818 
4819 	kfree(bargs);
4820 out:
4821 	mutex_unlock(&fs_info->balance_mutex);
4822 	return ret;
4823 }
4824 
4825 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4826 {
4827 	struct inode *inode = file_inode(file);
4828 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4829 	struct btrfs_ioctl_quota_ctl_args *sa;
4830 	struct btrfs_trans_handle *trans = NULL;
4831 	int ret;
4832 	int err;
4833 
4834 	if (!capable(CAP_SYS_ADMIN))
4835 		return -EPERM;
4836 
4837 	ret = mnt_want_write_file(file);
4838 	if (ret)
4839 		return ret;
4840 
4841 	sa = memdup_user(arg, sizeof(*sa));
4842 	if (IS_ERR(sa)) {
4843 		ret = PTR_ERR(sa);
4844 		goto drop_write;
4845 	}
4846 
4847 	down_write(&fs_info->subvol_sem);
4848 	trans = btrfs_start_transaction(fs_info->tree_root, 2);
4849 	if (IS_ERR(trans)) {
4850 		ret = PTR_ERR(trans);
4851 		goto out;
4852 	}
4853 
4854 	switch (sa->cmd) {
4855 	case BTRFS_QUOTA_CTL_ENABLE:
4856 		ret = btrfs_quota_enable(trans, fs_info);
4857 		break;
4858 	case BTRFS_QUOTA_CTL_DISABLE:
4859 		ret = btrfs_quota_disable(trans, fs_info);
4860 		break;
4861 	default:
4862 		ret = -EINVAL;
4863 		break;
4864 	}
4865 
4866 	err = btrfs_commit_transaction(trans);
4867 	if (err && !ret)
4868 		ret = err;
4869 out:
4870 	kfree(sa);
4871 	up_write(&fs_info->subvol_sem);
4872 drop_write:
4873 	mnt_drop_write_file(file);
4874 	return ret;
4875 }
4876 
4877 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4878 {
4879 	struct inode *inode = file_inode(file);
4880 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4881 	struct btrfs_root *root = BTRFS_I(inode)->root;
4882 	struct btrfs_ioctl_qgroup_assign_args *sa;
4883 	struct btrfs_trans_handle *trans;
4884 	int ret;
4885 	int err;
4886 
4887 	if (!capable(CAP_SYS_ADMIN))
4888 		return -EPERM;
4889 
4890 	ret = mnt_want_write_file(file);
4891 	if (ret)
4892 		return ret;
4893 
4894 	sa = memdup_user(arg, sizeof(*sa));
4895 	if (IS_ERR(sa)) {
4896 		ret = PTR_ERR(sa);
4897 		goto drop_write;
4898 	}
4899 
4900 	trans = btrfs_join_transaction(root);
4901 	if (IS_ERR(trans)) {
4902 		ret = PTR_ERR(trans);
4903 		goto out;
4904 	}
4905 
4906 	if (sa->assign) {
4907 		ret = btrfs_add_qgroup_relation(trans, fs_info,
4908 						sa->src, sa->dst);
4909 	} else {
4910 		ret = btrfs_del_qgroup_relation(trans, fs_info,
4911 						sa->src, sa->dst);
4912 	}
4913 
4914 	/* update qgroup status and info */
4915 	err = btrfs_run_qgroups(trans, fs_info);
4916 	if (err < 0)
4917 		btrfs_handle_fs_error(fs_info, err,
4918 				      "failed to update qgroup status and info");
4919 	err = btrfs_end_transaction(trans);
4920 	if (err && !ret)
4921 		ret = err;
4922 
4923 out:
4924 	kfree(sa);
4925 drop_write:
4926 	mnt_drop_write_file(file);
4927 	return ret;
4928 }
4929 
4930 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4931 {
4932 	struct inode *inode = file_inode(file);
4933 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4934 	struct btrfs_root *root = BTRFS_I(inode)->root;
4935 	struct btrfs_ioctl_qgroup_create_args *sa;
4936 	struct btrfs_trans_handle *trans;
4937 	int ret;
4938 	int err;
4939 
4940 	if (!capable(CAP_SYS_ADMIN))
4941 		return -EPERM;
4942 
4943 	ret = mnt_want_write_file(file);
4944 	if (ret)
4945 		return ret;
4946 
4947 	sa = memdup_user(arg, sizeof(*sa));
4948 	if (IS_ERR(sa)) {
4949 		ret = PTR_ERR(sa);
4950 		goto drop_write;
4951 	}
4952 
4953 	if (!sa->qgroupid) {
4954 		ret = -EINVAL;
4955 		goto out;
4956 	}
4957 
4958 	trans = btrfs_join_transaction(root);
4959 	if (IS_ERR(trans)) {
4960 		ret = PTR_ERR(trans);
4961 		goto out;
4962 	}
4963 
4964 	if (sa->create) {
4965 		ret = btrfs_create_qgroup(trans, fs_info, sa->qgroupid);
4966 	} else {
4967 		ret = btrfs_remove_qgroup(trans, fs_info, sa->qgroupid);
4968 	}
4969 
4970 	err = btrfs_end_transaction(trans);
4971 	if (err && !ret)
4972 		ret = err;
4973 
4974 out:
4975 	kfree(sa);
4976 drop_write:
4977 	mnt_drop_write_file(file);
4978 	return ret;
4979 }
4980 
4981 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4982 {
4983 	struct inode *inode = file_inode(file);
4984 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4985 	struct btrfs_root *root = BTRFS_I(inode)->root;
4986 	struct btrfs_ioctl_qgroup_limit_args *sa;
4987 	struct btrfs_trans_handle *trans;
4988 	int ret;
4989 	int err;
4990 	u64 qgroupid;
4991 
4992 	if (!capable(CAP_SYS_ADMIN))
4993 		return -EPERM;
4994 
4995 	ret = mnt_want_write_file(file);
4996 	if (ret)
4997 		return ret;
4998 
4999 	sa = memdup_user(arg, sizeof(*sa));
5000 	if (IS_ERR(sa)) {
5001 		ret = PTR_ERR(sa);
5002 		goto drop_write;
5003 	}
5004 
5005 	trans = btrfs_join_transaction(root);
5006 	if (IS_ERR(trans)) {
5007 		ret = PTR_ERR(trans);
5008 		goto out;
5009 	}
5010 
5011 	qgroupid = sa->qgroupid;
5012 	if (!qgroupid) {
5013 		/* take the current subvol as qgroup */
5014 		qgroupid = root->root_key.objectid;
5015 	}
5016 
5017 	ret = btrfs_limit_qgroup(trans, fs_info, qgroupid, &sa->lim);
5018 
5019 	err = btrfs_end_transaction(trans);
5020 	if (err && !ret)
5021 		ret = err;
5022 
5023 out:
5024 	kfree(sa);
5025 drop_write:
5026 	mnt_drop_write_file(file);
5027 	return ret;
5028 }
5029 
5030 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
5031 {
5032 	struct inode *inode = file_inode(file);
5033 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5034 	struct btrfs_ioctl_quota_rescan_args *qsa;
5035 	int ret;
5036 
5037 	if (!capable(CAP_SYS_ADMIN))
5038 		return -EPERM;
5039 
5040 	ret = mnt_want_write_file(file);
5041 	if (ret)
5042 		return ret;
5043 
5044 	qsa = memdup_user(arg, sizeof(*qsa));
5045 	if (IS_ERR(qsa)) {
5046 		ret = PTR_ERR(qsa);
5047 		goto drop_write;
5048 	}
5049 
5050 	if (qsa->flags) {
5051 		ret = -EINVAL;
5052 		goto out;
5053 	}
5054 
5055 	ret = btrfs_qgroup_rescan(fs_info);
5056 
5057 out:
5058 	kfree(qsa);
5059 drop_write:
5060 	mnt_drop_write_file(file);
5061 	return ret;
5062 }
5063 
5064 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
5065 {
5066 	struct inode *inode = file_inode(file);
5067 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5068 	struct btrfs_ioctl_quota_rescan_args *qsa;
5069 	int ret = 0;
5070 
5071 	if (!capable(CAP_SYS_ADMIN))
5072 		return -EPERM;
5073 
5074 	qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
5075 	if (!qsa)
5076 		return -ENOMEM;
5077 
5078 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
5079 		qsa->flags = 1;
5080 		qsa->progress = fs_info->qgroup_rescan_progress.objectid;
5081 	}
5082 
5083 	if (copy_to_user(arg, qsa, sizeof(*qsa)))
5084 		ret = -EFAULT;
5085 
5086 	kfree(qsa);
5087 	return ret;
5088 }
5089 
5090 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
5091 {
5092 	struct inode *inode = file_inode(file);
5093 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5094 
5095 	if (!capable(CAP_SYS_ADMIN))
5096 		return -EPERM;
5097 
5098 	return btrfs_qgroup_wait_for_completion(fs_info, true);
5099 }
5100 
5101 static long _btrfs_ioctl_set_received_subvol(struct file *file,
5102 					    struct btrfs_ioctl_received_subvol_args *sa)
5103 {
5104 	struct inode *inode = file_inode(file);
5105 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5106 	struct btrfs_root *root = BTRFS_I(inode)->root;
5107 	struct btrfs_root_item *root_item = &root->root_item;
5108 	struct btrfs_trans_handle *trans;
5109 	struct timespec ct = current_time(inode);
5110 	int ret = 0;
5111 	int received_uuid_changed;
5112 
5113 	if (!inode_owner_or_capable(inode))
5114 		return -EPERM;
5115 
5116 	ret = mnt_want_write_file(file);
5117 	if (ret < 0)
5118 		return ret;
5119 
5120 	down_write(&fs_info->subvol_sem);
5121 
5122 	if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
5123 		ret = -EINVAL;
5124 		goto out;
5125 	}
5126 
5127 	if (btrfs_root_readonly(root)) {
5128 		ret = -EROFS;
5129 		goto out;
5130 	}
5131 
5132 	/*
5133 	 * 1 - root item
5134 	 * 2 - uuid items (received uuid + subvol uuid)
5135 	 */
5136 	trans = btrfs_start_transaction(root, 3);
5137 	if (IS_ERR(trans)) {
5138 		ret = PTR_ERR(trans);
5139 		trans = NULL;
5140 		goto out;
5141 	}
5142 
5143 	sa->rtransid = trans->transid;
5144 	sa->rtime.sec = ct.tv_sec;
5145 	sa->rtime.nsec = ct.tv_nsec;
5146 
5147 	received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5148 				       BTRFS_UUID_SIZE);
5149 	if (received_uuid_changed &&
5150 	    !btrfs_is_empty_uuid(root_item->received_uuid))
5151 		btrfs_uuid_tree_rem(trans, fs_info, root_item->received_uuid,
5152 				    BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5153 				    root->root_key.objectid);
5154 	memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5155 	btrfs_set_root_stransid(root_item, sa->stransid);
5156 	btrfs_set_root_rtransid(root_item, sa->rtransid);
5157 	btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5158 	btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5159 	btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5160 	btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5161 
5162 	ret = btrfs_update_root(trans, fs_info->tree_root,
5163 				&root->root_key, &root->root_item);
5164 	if (ret < 0) {
5165 		btrfs_end_transaction(trans);
5166 		goto out;
5167 	}
5168 	if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5169 		ret = btrfs_uuid_tree_add(trans, fs_info, sa->uuid,
5170 					  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5171 					  root->root_key.objectid);
5172 		if (ret < 0 && ret != -EEXIST) {
5173 			btrfs_abort_transaction(trans, ret);
5174 			btrfs_end_transaction(trans);
5175 			goto out;
5176 		}
5177 	}
5178 	ret = btrfs_commit_transaction(trans);
5179 out:
5180 	up_write(&fs_info->subvol_sem);
5181 	mnt_drop_write_file(file);
5182 	return ret;
5183 }
5184 
5185 #ifdef CONFIG_64BIT
5186 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5187 						void __user *arg)
5188 {
5189 	struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5190 	struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5191 	int ret = 0;
5192 
5193 	args32 = memdup_user(arg, sizeof(*args32));
5194 	if (IS_ERR(args32))
5195 		return PTR_ERR(args32);
5196 
5197 	args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
5198 	if (!args64) {
5199 		ret = -ENOMEM;
5200 		goto out;
5201 	}
5202 
5203 	memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5204 	args64->stransid = args32->stransid;
5205 	args64->rtransid = args32->rtransid;
5206 	args64->stime.sec = args32->stime.sec;
5207 	args64->stime.nsec = args32->stime.nsec;
5208 	args64->rtime.sec = args32->rtime.sec;
5209 	args64->rtime.nsec = args32->rtime.nsec;
5210 	args64->flags = args32->flags;
5211 
5212 	ret = _btrfs_ioctl_set_received_subvol(file, args64);
5213 	if (ret)
5214 		goto out;
5215 
5216 	memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5217 	args32->stransid = args64->stransid;
5218 	args32->rtransid = args64->rtransid;
5219 	args32->stime.sec = args64->stime.sec;
5220 	args32->stime.nsec = args64->stime.nsec;
5221 	args32->rtime.sec = args64->rtime.sec;
5222 	args32->rtime.nsec = args64->rtime.nsec;
5223 	args32->flags = args64->flags;
5224 
5225 	ret = copy_to_user(arg, args32, sizeof(*args32));
5226 	if (ret)
5227 		ret = -EFAULT;
5228 
5229 out:
5230 	kfree(args32);
5231 	kfree(args64);
5232 	return ret;
5233 }
5234 #endif
5235 
5236 static long btrfs_ioctl_set_received_subvol(struct file *file,
5237 					    void __user *arg)
5238 {
5239 	struct btrfs_ioctl_received_subvol_args *sa = NULL;
5240 	int ret = 0;
5241 
5242 	sa = memdup_user(arg, sizeof(*sa));
5243 	if (IS_ERR(sa))
5244 		return PTR_ERR(sa);
5245 
5246 	ret = _btrfs_ioctl_set_received_subvol(file, sa);
5247 
5248 	if (ret)
5249 		goto out;
5250 
5251 	ret = copy_to_user(arg, sa, sizeof(*sa));
5252 	if (ret)
5253 		ret = -EFAULT;
5254 
5255 out:
5256 	kfree(sa);
5257 	return ret;
5258 }
5259 
5260 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5261 {
5262 	struct inode *inode = file_inode(file);
5263 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5264 	size_t len;
5265 	int ret;
5266 	char label[BTRFS_LABEL_SIZE];
5267 
5268 	spin_lock(&fs_info->super_lock);
5269 	memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5270 	spin_unlock(&fs_info->super_lock);
5271 
5272 	len = strnlen(label, BTRFS_LABEL_SIZE);
5273 
5274 	if (len == BTRFS_LABEL_SIZE) {
5275 		btrfs_warn(fs_info,
5276 			   "label is too long, return the first %zu bytes",
5277 			   --len);
5278 	}
5279 
5280 	ret = copy_to_user(arg, label, len);
5281 
5282 	return ret ? -EFAULT : 0;
5283 }
5284 
5285 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5286 {
5287 	struct inode *inode = file_inode(file);
5288 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5289 	struct btrfs_root *root = BTRFS_I(inode)->root;
5290 	struct btrfs_super_block *super_block = fs_info->super_copy;
5291 	struct btrfs_trans_handle *trans;
5292 	char label[BTRFS_LABEL_SIZE];
5293 	int ret;
5294 
5295 	if (!capable(CAP_SYS_ADMIN))
5296 		return -EPERM;
5297 
5298 	if (copy_from_user(label, arg, sizeof(label)))
5299 		return -EFAULT;
5300 
5301 	if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5302 		btrfs_err(fs_info,
5303 			  "unable to set label with more than %d bytes",
5304 			  BTRFS_LABEL_SIZE - 1);
5305 		return -EINVAL;
5306 	}
5307 
5308 	ret = mnt_want_write_file(file);
5309 	if (ret)
5310 		return ret;
5311 
5312 	trans = btrfs_start_transaction(root, 0);
5313 	if (IS_ERR(trans)) {
5314 		ret = PTR_ERR(trans);
5315 		goto out_unlock;
5316 	}
5317 
5318 	spin_lock(&fs_info->super_lock);
5319 	strcpy(super_block->label, label);
5320 	spin_unlock(&fs_info->super_lock);
5321 	ret = btrfs_commit_transaction(trans);
5322 
5323 out_unlock:
5324 	mnt_drop_write_file(file);
5325 	return ret;
5326 }
5327 
5328 #define INIT_FEATURE_FLAGS(suffix) \
5329 	{ .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5330 	  .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5331 	  .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5332 
5333 int btrfs_ioctl_get_supported_features(void __user *arg)
5334 {
5335 	static const struct btrfs_ioctl_feature_flags features[3] = {
5336 		INIT_FEATURE_FLAGS(SUPP),
5337 		INIT_FEATURE_FLAGS(SAFE_SET),
5338 		INIT_FEATURE_FLAGS(SAFE_CLEAR)
5339 	};
5340 
5341 	if (copy_to_user(arg, &features, sizeof(features)))
5342 		return -EFAULT;
5343 
5344 	return 0;
5345 }
5346 
5347 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5348 {
5349 	struct inode *inode = file_inode(file);
5350 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5351 	struct btrfs_super_block *super_block = fs_info->super_copy;
5352 	struct btrfs_ioctl_feature_flags features;
5353 
5354 	features.compat_flags = btrfs_super_compat_flags(super_block);
5355 	features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5356 	features.incompat_flags = btrfs_super_incompat_flags(super_block);
5357 
5358 	if (copy_to_user(arg, &features, sizeof(features)))
5359 		return -EFAULT;
5360 
5361 	return 0;
5362 }
5363 
5364 static int check_feature_bits(struct btrfs_fs_info *fs_info,
5365 			      enum btrfs_feature_set set,
5366 			      u64 change_mask, u64 flags, u64 supported_flags,
5367 			      u64 safe_set, u64 safe_clear)
5368 {
5369 	const char *type = btrfs_feature_set_names[set];
5370 	char *names;
5371 	u64 disallowed, unsupported;
5372 	u64 set_mask = flags & change_mask;
5373 	u64 clear_mask = ~flags & change_mask;
5374 
5375 	unsupported = set_mask & ~supported_flags;
5376 	if (unsupported) {
5377 		names = btrfs_printable_features(set, unsupported);
5378 		if (names) {
5379 			btrfs_warn(fs_info,
5380 				   "this kernel does not support the %s feature bit%s",
5381 				   names, strchr(names, ',') ? "s" : "");
5382 			kfree(names);
5383 		} else
5384 			btrfs_warn(fs_info,
5385 				   "this kernel does not support %s bits 0x%llx",
5386 				   type, unsupported);
5387 		return -EOPNOTSUPP;
5388 	}
5389 
5390 	disallowed = set_mask & ~safe_set;
5391 	if (disallowed) {
5392 		names = btrfs_printable_features(set, disallowed);
5393 		if (names) {
5394 			btrfs_warn(fs_info,
5395 				   "can't set the %s feature bit%s while mounted",
5396 				   names, strchr(names, ',') ? "s" : "");
5397 			kfree(names);
5398 		} else
5399 			btrfs_warn(fs_info,
5400 				   "can't set %s bits 0x%llx while mounted",
5401 				   type, disallowed);
5402 		return -EPERM;
5403 	}
5404 
5405 	disallowed = clear_mask & ~safe_clear;
5406 	if (disallowed) {
5407 		names = btrfs_printable_features(set, disallowed);
5408 		if (names) {
5409 			btrfs_warn(fs_info,
5410 				   "can't clear the %s feature bit%s while mounted",
5411 				   names, strchr(names, ',') ? "s" : "");
5412 			kfree(names);
5413 		} else
5414 			btrfs_warn(fs_info,
5415 				   "can't clear %s bits 0x%llx while mounted",
5416 				   type, disallowed);
5417 		return -EPERM;
5418 	}
5419 
5420 	return 0;
5421 }
5422 
5423 #define check_feature(fs_info, change_mask, flags, mask_base)	\
5424 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,	\
5425 		   BTRFS_FEATURE_ ## mask_base ## _SUPP,	\
5426 		   BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,	\
5427 		   BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5428 
5429 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5430 {
5431 	struct inode *inode = file_inode(file);
5432 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5433 	struct btrfs_root *root = BTRFS_I(inode)->root;
5434 	struct btrfs_super_block *super_block = fs_info->super_copy;
5435 	struct btrfs_ioctl_feature_flags flags[2];
5436 	struct btrfs_trans_handle *trans;
5437 	u64 newflags;
5438 	int ret;
5439 
5440 	if (!capable(CAP_SYS_ADMIN))
5441 		return -EPERM;
5442 
5443 	if (copy_from_user(flags, arg, sizeof(flags)))
5444 		return -EFAULT;
5445 
5446 	/* Nothing to do */
5447 	if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5448 	    !flags[0].incompat_flags)
5449 		return 0;
5450 
5451 	ret = check_feature(fs_info, flags[0].compat_flags,
5452 			    flags[1].compat_flags, COMPAT);
5453 	if (ret)
5454 		return ret;
5455 
5456 	ret = check_feature(fs_info, flags[0].compat_ro_flags,
5457 			    flags[1].compat_ro_flags, COMPAT_RO);
5458 	if (ret)
5459 		return ret;
5460 
5461 	ret = check_feature(fs_info, flags[0].incompat_flags,
5462 			    flags[1].incompat_flags, INCOMPAT);
5463 	if (ret)
5464 		return ret;
5465 
5466 	ret = mnt_want_write_file(file);
5467 	if (ret)
5468 		return ret;
5469 
5470 	trans = btrfs_start_transaction(root, 0);
5471 	if (IS_ERR(trans)) {
5472 		ret = PTR_ERR(trans);
5473 		goto out_drop_write;
5474 	}
5475 
5476 	spin_lock(&fs_info->super_lock);
5477 	newflags = btrfs_super_compat_flags(super_block);
5478 	newflags |= flags[0].compat_flags & flags[1].compat_flags;
5479 	newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5480 	btrfs_set_super_compat_flags(super_block, newflags);
5481 
5482 	newflags = btrfs_super_compat_ro_flags(super_block);
5483 	newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5484 	newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5485 	btrfs_set_super_compat_ro_flags(super_block, newflags);
5486 
5487 	newflags = btrfs_super_incompat_flags(super_block);
5488 	newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5489 	newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5490 	btrfs_set_super_incompat_flags(super_block, newflags);
5491 	spin_unlock(&fs_info->super_lock);
5492 
5493 	ret = btrfs_commit_transaction(trans);
5494 out_drop_write:
5495 	mnt_drop_write_file(file);
5496 
5497 	return ret;
5498 }
5499 
5500 static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
5501 {
5502 	struct btrfs_ioctl_send_args *arg;
5503 	int ret;
5504 
5505 	if (compat) {
5506 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5507 		struct btrfs_ioctl_send_args_32 args32;
5508 
5509 		ret = copy_from_user(&args32, argp, sizeof(args32));
5510 		if (ret)
5511 			return -EFAULT;
5512 		arg = kzalloc(sizeof(*arg), GFP_KERNEL);
5513 		if (!arg)
5514 			return -ENOMEM;
5515 		arg->send_fd = args32.send_fd;
5516 		arg->clone_sources_count = args32.clone_sources_count;
5517 		arg->clone_sources = compat_ptr(args32.clone_sources);
5518 		arg->parent_root = args32.parent_root;
5519 		arg->flags = args32.flags;
5520 		memcpy(arg->reserved, args32.reserved,
5521 		       sizeof(args32.reserved));
5522 #else
5523 		return -ENOTTY;
5524 #endif
5525 	} else {
5526 		arg = memdup_user(argp, sizeof(*arg));
5527 		if (IS_ERR(arg))
5528 			return PTR_ERR(arg);
5529 	}
5530 	ret = btrfs_ioctl_send(file, arg);
5531 	kfree(arg);
5532 	return ret;
5533 }
5534 
5535 long btrfs_ioctl(struct file *file, unsigned int
5536 		cmd, unsigned long arg)
5537 {
5538 	struct inode *inode = file_inode(file);
5539 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5540 	struct btrfs_root *root = BTRFS_I(inode)->root;
5541 	void __user *argp = (void __user *)arg;
5542 
5543 	switch (cmd) {
5544 	case FS_IOC_GETFLAGS:
5545 		return btrfs_ioctl_getflags(file, argp);
5546 	case FS_IOC_SETFLAGS:
5547 		return btrfs_ioctl_setflags(file, argp);
5548 	case FS_IOC_GETVERSION:
5549 		return btrfs_ioctl_getversion(file, argp);
5550 	case FITRIM:
5551 		return btrfs_ioctl_fitrim(file, argp);
5552 	case BTRFS_IOC_SNAP_CREATE:
5553 		return btrfs_ioctl_snap_create(file, argp, 0);
5554 	case BTRFS_IOC_SNAP_CREATE_V2:
5555 		return btrfs_ioctl_snap_create_v2(file, argp, 0);
5556 	case BTRFS_IOC_SUBVOL_CREATE:
5557 		return btrfs_ioctl_snap_create(file, argp, 1);
5558 	case BTRFS_IOC_SUBVOL_CREATE_V2:
5559 		return btrfs_ioctl_snap_create_v2(file, argp, 1);
5560 	case BTRFS_IOC_SNAP_DESTROY:
5561 		return btrfs_ioctl_snap_destroy(file, argp);
5562 	case BTRFS_IOC_SUBVOL_GETFLAGS:
5563 		return btrfs_ioctl_subvol_getflags(file, argp);
5564 	case BTRFS_IOC_SUBVOL_SETFLAGS:
5565 		return btrfs_ioctl_subvol_setflags(file, argp);
5566 	case BTRFS_IOC_DEFAULT_SUBVOL:
5567 		return btrfs_ioctl_default_subvol(file, argp);
5568 	case BTRFS_IOC_DEFRAG:
5569 		return btrfs_ioctl_defrag(file, NULL);
5570 	case BTRFS_IOC_DEFRAG_RANGE:
5571 		return btrfs_ioctl_defrag(file, argp);
5572 	case BTRFS_IOC_RESIZE:
5573 		return btrfs_ioctl_resize(file, argp);
5574 	case BTRFS_IOC_ADD_DEV:
5575 		return btrfs_ioctl_add_dev(fs_info, argp);
5576 	case BTRFS_IOC_RM_DEV:
5577 		return btrfs_ioctl_rm_dev(file, argp);
5578 	case BTRFS_IOC_RM_DEV_V2:
5579 		return btrfs_ioctl_rm_dev_v2(file, argp);
5580 	case BTRFS_IOC_FS_INFO:
5581 		return btrfs_ioctl_fs_info(fs_info, argp);
5582 	case BTRFS_IOC_DEV_INFO:
5583 		return btrfs_ioctl_dev_info(fs_info, argp);
5584 	case BTRFS_IOC_BALANCE:
5585 		return btrfs_ioctl_balance(file, NULL);
5586 	case BTRFS_IOC_TRANS_START:
5587 		return btrfs_ioctl_trans_start(file);
5588 	case BTRFS_IOC_TRANS_END:
5589 		return btrfs_ioctl_trans_end(file);
5590 	case BTRFS_IOC_TREE_SEARCH:
5591 		return btrfs_ioctl_tree_search(file, argp);
5592 	case BTRFS_IOC_TREE_SEARCH_V2:
5593 		return btrfs_ioctl_tree_search_v2(file, argp);
5594 	case BTRFS_IOC_INO_LOOKUP:
5595 		return btrfs_ioctl_ino_lookup(file, argp);
5596 	case BTRFS_IOC_INO_PATHS:
5597 		return btrfs_ioctl_ino_to_path(root, argp);
5598 	case BTRFS_IOC_LOGICAL_INO:
5599 		return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
5600 	case BTRFS_IOC_LOGICAL_INO_V2:
5601 		return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
5602 	case BTRFS_IOC_SPACE_INFO:
5603 		return btrfs_ioctl_space_info(fs_info, argp);
5604 	case BTRFS_IOC_SYNC: {
5605 		int ret;
5606 
5607 		ret = btrfs_start_delalloc_roots(fs_info, 0, -1);
5608 		if (ret)
5609 			return ret;
5610 		ret = btrfs_sync_fs(inode->i_sb, 1);
5611 		/*
5612 		 * The transaction thread may want to do more work,
5613 		 * namely it pokes the cleaner kthread that will start
5614 		 * processing uncleaned subvols.
5615 		 */
5616 		wake_up_process(fs_info->transaction_kthread);
5617 		return ret;
5618 	}
5619 	case BTRFS_IOC_START_SYNC:
5620 		return btrfs_ioctl_start_sync(root, argp);
5621 	case BTRFS_IOC_WAIT_SYNC:
5622 		return btrfs_ioctl_wait_sync(fs_info, argp);
5623 	case BTRFS_IOC_SCRUB:
5624 		return btrfs_ioctl_scrub(file, argp);
5625 	case BTRFS_IOC_SCRUB_CANCEL:
5626 		return btrfs_ioctl_scrub_cancel(fs_info);
5627 	case BTRFS_IOC_SCRUB_PROGRESS:
5628 		return btrfs_ioctl_scrub_progress(fs_info, argp);
5629 	case BTRFS_IOC_BALANCE_V2:
5630 		return btrfs_ioctl_balance(file, argp);
5631 	case BTRFS_IOC_BALANCE_CTL:
5632 		return btrfs_ioctl_balance_ctl(fs_info, arg);
5633 	case BTRFS_IOC_BALANCE_PROGRESS:
5634 		return btrfs_ioctl_balance_progress(fs_info, argp);
5635 	case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5636 		return btrfs_ioctl_set_received_subvol(file, argp);
5637 #ifdef CONFIG_64BIT
5638 	case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5639 		return btrfs_ioctl_set_received_subvol_32(file, argp);
5640 #endif
5641 	case BTRFS_IOC_SEND:
5642 		return _btrfs_ioctl_send(file, argp, false);
5643 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5644 	case BTRFS_IOC_SEND_32:
5645 		return _btrfs_ioctl_send(file, argp, true);
5646 #endif
5647 	case BTRFS_IOC_GET_DEV_STATS:
5648 		return btrfs_ioctl_get_dev_stats(fs_info, argp);
5649 	case BTRFS_IOC_QUOTA_CTL:
5650 		return btrfs_ioctl_quota_ctl(file, argp);
5651 	case BTRFS_IOC_QGROUP_ASSIGN:
5652 		return btrfs_ioctl_qgroup_assign(file, argp);
5653 	case BTRFS_IOC_QGROUP_CREATE:
5654 		return btrfs_ioctl_qgroup_create(file, argp);
5655 	case BTRFS_IOC_QGROUP_LIMIT:
5656 		return btrfs_ioctl_qgroup_limit(file, argp);
5657 	case BTRFS_IOC_QUOTA_RESCAN:
5658 		return btrfs_ioctl_quota_rescan(file, argp);
5659 	case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5660 		return btrfs_ioctl_quota_rescan_status(file, argp);
5661 	case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5662 		return btrfs_ioctl_quota_rescan_wait(file, argp);
5663 	case BTRFS_IOC_DEV_REPLACE:
5664 		return btrfs_ioctl_dev_replace(fs_info, argp);
5665 	case BTRFS_IOC_GET_FSLABEL:
5666 		return btrfs_ioctl_get_fslabel(file, argp);
5667 	case BTRFS_IOC_SET_FSLABEL:
5668 		return btrfs_ioctl_set_fslabel(file, argp);
5669 	case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5670 		return btrfs_ioctl_get_supported_features(argp);
5671 	case BTRFS_IOC_GET_FEATURES:
5672 		return btrfs_ioctl_get_features(file, argp);
5673 	case BTRFS_IOC_SET_FEATURES:
5674 		return btrfs_ioctl_set_features(file, argp);
5675 	}
5676 
5677 	return -ENOTTY;
5678 }
5679 
5680 #ifdef CONFIG_COMPAT
5681 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5682 {
5683 	/*
5684 	 * These all access 32-bit values anyway so no further
5685 	 * handling is necessary.
5686 	 */
5687 	switch (cmd) {
5688 	case FS_IOC32_GETFLAGS:
5689 		cmd = FS_IOC_GETFLAGS;
5690 		break;
5691 	case FS_IOC32_SETFLAGS:
5692 		cmd = FS_IOC_SETFLAGS;
5693 		break;
5694 	case FS_IOC32_GETVERSION:
5695 		cmd = FS_IOC_GETVERSION;
5696 		break;
5697 	}
5698 
5699 	return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5700 }
5701 #endif
5702