ioctl.c revision 0afb603a
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2007 Oracle.  All rights reserved.
4 */
5
6#include <linux/kernel.h>
7#include <linux/bio.h>
8#include <linux/file.h>
9#include <linux/fs.h>
10#include <linux/fsnotify.h>
11#include <linux/pagemap.h>
12#include <linux/highmem.h>
13#include <linux/time.h>
14#include <linux/string.h>
15#include <linux/backing-dev.h>
16#include <linux/mount.h>
17#include <linux/namei.h>
18#include <linux/writeback.h>
19#include <linux/compat.h>
20#include <linux/security.h>
21#include <linux/xattr.h>
22#include <linux/mm.h>
23#include <linux/slab.h>
24#include <linux/blkdev.h>
25#include <linux/uuid.h>
26#include <linux/btrfs.h>
27#include <linux/uaccess.h>
28#include <linux/iversion.h>
29#include <linux/fileattr.h>
30#include <linux/fsverity.h>
31#include "ctree.h"
32#include "disk-io.h"
33#include "export.h"
34#include "transaction.h"
35#include "btrfs_inode.h"
36#include "print-tree.h"
37#include "volumes.h"
38#include "locking.h"
39#include "backref.h"
40#include "rcu-string.h"
41#include "send.h"
42#include "dev-replace.h"
43#include "props.h"
44#include "sysfs.h"
45#include "qgroup.h"
46#include "tree-log.h"
47#include "compression.h"
48#include "space-info.h"
49#include "delalloc-space.h"
50#include "block-group.h"
51
52#ifdef CONFIG_64BIT
53/* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
54 * structures are incorrect, as the timespec structure from userspace
55 * is 4 bytes too small. We define these alternatives here to teach
56 * the kernel about the 32-bit struct packing.
57 */
58struct btrfs_ioctl_timespec_32 {
59	__u64 sec;
60	__u32 nsec;
61} __attribute__ ((__packed__));
62
63struct btrfs_ioctl_received_subvol_args_32 {
64	char	uuid[BTRFS_UUID_SIZE];	/* in */
65	__u64	stransid;		/* in */
66	__u64	rtransid;		/* out */
67	struct btrfs_ioctl_timespec_32 stime; /* in */
68	struct btrfs_ioctl_timespec_32 rtime; /* out */
69	__u64	flags;			/* in */
70	__u64	reserved[16];		/* in */
71} __attribute__ ((__packed__));
72
73#define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
74				struct btrfs_ioctl_received_subvol_args_32)
75#endif
76
77#if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
78struct btrfs_ioctl_send_args_32 {
79	__s64 send_fd;			/* in */
80	__u64 clone_sources_count;	/* in */
81	compat_uptr_t clone_sources;	/* in */
82	__u64 parent_root;		/* in */
83	__u64 flags;			/* in */
84	__u64 reserved[4];		/* in */
85} __attribute__ ((__packed__));
86
87#define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \
88			       struct btrfs_ioctl_send_args_32)
89#endif
90
91/* Mask out flags that are inappropriate for the given type of inode. */
92static unsigned int btrfs_mask_fsflags_for_type(struct inode *inode,
93		unsigned int flags)
94{
95	if (S_ISDIR(inode->i_mode))
96		return flags;
97	else if (S_ISREG(inode->i_mode))
98		return flags & ~FS_DIRSYNC_FL;
99	else
100		return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
101}
102
103/*
104 * Export internal inode flags to the format expected by the FS_IOC_GETFLAGS
105 * ioctl.
106 */
107static unsigned int btrfs_inode_flags_to_fsflags(struct btrfs_inode *binode)
108{
109	unsigned int iflags = 0;
110	u32 flags = binode->flags;
111	u32 ro_flags = binode->ro_flags;
112
113	if (flags & BTRFS_INODE_SYNC)
114		iflags |= FS_SYNC_FL;
115	if (flags & BTRFS_INODE_IMMUTABLE)
116		iflags |= FS_IMMUTABLE_FL;
117	if (flags & BTRFS_INODE_APPEND)
118		iflags |= FS_APPEND_FL;
119	if (flags & BTRFS_INODE_NODUMP)
120		iflags |= FS_NODUMP_FL;
121	if (flags & BTRFS_INODE_NOATIME)
122		iflags |= FS_NOATIME_FL;
123	if (flags & BTRFS_INODE_DIRSYNC)
124		iflags |= FS_DIRSYNC_FL;
125	if (flags & BTRFS_INODE_NODATACOW)
126		iflags |= FS_NOCOW_FL;
127	if (ro_flags & BTRFS_INODE_RO_VERITY)
128		iflags |= FS_VERITY_FL;
129
130	if (flags & BTRFS_INODE_NOCOMPRESS)
131		iflags |= FS_NOCOMP_FL;
132	else if (flags & BTRFS_INODE_COMPRESS)
133		iflags |= FS_COMPR_FL;
134
135	return iflags;
136}
137
138/*
139 * Update inode->i_flags based on the btrfs internal flags.
140 */
141void btrfs_sync_inode_flags_to_i_flags(struct inode *inode)
142{
143	struct btrfs_inode *binode = BTRFS_I(inode);
144	unsigned int new_fl = 0;
145
146	if (binode->flags & BTRFS_INODE_SYNC)
147		new_fl |= S_SYNC;
148	if (binode->flags & BTRFS_INODE_IMMUTABLE)
149		new_fl |= S_IMMUTABLE;
150	if (binode->flags & BTRFS_INODE_APPEND)
151		new_fl |= S_APPEND;
152	if (binode->flags & BTRFS_INODE_NOATIME)
153		new_fl |= S_NOATIME;
154	if (binode->flags & BTRFS_INODE_DIRSYNC)
155		new_fl |= S_DIRSYNC;
156	if (binode->ro_flags & BTRFS_INODE_RO_VERITY)
157		new_fl |= S_VERITY;
158
159	set_mask_bits(&inode->i_flags,
160		      S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC |
161		      S_VERITY, new_fl);
162}
163
164/*
165 * Check if @flags are a supported and valid set of FS_*_FL flags and that
166 * the old and new flags are not conflicting
167 */
168static int check_fsflags(unsigned int old_flags, unsigned int flags)
169{
170	if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
171		      FS_NOATIME_FL | FS_NODUMP_FL | \
172		      FS_SYNC_FL | FS_DIRSYNC_FL | \
173		      FS_NOCOMP_FL | FS_COMPR_FL |
174		      FS_NOCOW_FL))
175		return -EOPNOTSUPP;
176
177	/* COMPR and NOCOMP on new/old are valid */
178	if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
179		return -EINVAL;
180
181	if ((flags & FS_COMPR_FL) && (flags & FS_NOCOW_FL))
182		return -EINVAL;
183
184	/* NOCOW and compression options are mutually exclusive */
185	if ((old_flags & FS_NOCOW_FL) && (flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
186		return -EINVAL;
187	if ((flags & FS_NOCOW_FL) && (old_flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
188		return -EINVAL;
189
190	return 0;
191}
192
193static int check_fsflags_compatible(struct btrfs_fs_info *fs_info,
194				    unsigned int flags)
195{
196	if (btrfs_is_zoned(fs_info) && (flags & FS_NOCOW_FL))
197		return -EPERM;
198
199	return 0;
200}
201
202/*
203 * Set flags/xflags from the internal inode flags. The remaining items of
204 * fsxattr are zeroed.
205 */
206int btrfs_fileattr_get(struct dentry *dentry, struct fileattr *fa)
207{
208	struct btrfs_inode *binode = BTRFS_I(d_inode(dentry));
209
210	fileattr_fill_flags(fa, btrfs_inode_flags_to_fsflags(binode));
211	return 0;
212}
213
214int btrfs_fileattr_set(struct user_namespace *mnt_userns,
215		       struct dentry *dentry, struct fileattr *fa)
216{
217	struct inode *inode = d_inode(dentry);
218	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
219	struct btrfs_inode *binode = BTRFS_I(inode);
220	struct btrfs_root *root = binode->root;
221	struct btrfs_trans_handle *trans;
222	unsigned int fsflags, old_fsflags;
223	int ret;
224	const char *comp = NULL;
225	u32 binode_flags;
226
227	if (btrfs_root_readonly(root))
228		return -EROFS;
229
230	if (fileattr_has_fsx(fa))
231		return -EOPNOTSUPP;
232
233	fsflags = btrfs_mask_fsflags_for_type(inode, fa->flags);
234	old_fsflags = btrfs_inode_flags_to_fsflags(binode);
235	ret = check_fsflags(old_fsflags, fsflags);
236	if (ret)
237		return ret;
238
239	ret = check_fsflags_compatible(fs_info, fsflags);
240	if (ret)
241		return ret;
242
243	binode_flags = binode->flags;
244	if (fsflags & FS_SYNC_FL)
245		binode_flags |= BTRFS_INODE_SYNC;
246	else
247		binode_flags &= ~BTRFS_INODE_SYNC;
248	if (fsflags & FS_IMMUTABLE_FL)
249		binode_flags |= BTRFS_INODE_IMMUTABLE;
250	else
251		binode_flags &= ~BTRFS_INODE_IMMUTABLE;
252	if (fsflags & FS_APPEND_FL)
253		binode_flags |= BTRFS_INODE_APPEND;
254	else
255		binode_flags &= ~BTRFS_INODE_APPEND;
256	if (fsflags & FS_NODUMP_FL)
257		binode_flags |= BTRFS_INODE_NODUMP;
258	else
259		binode_flags &= ~BTRFS_INODE_NODUMP;
260	if (fsflags & FS_NOATIME_FL)
261		binode_flags |= BTRFS_INODE_NOATIME;
262	else
263		binode_flags &= ~BTRFS_INODE_NOATIME;
264
265	/* If coming from FS_IOC_FSSETXATTR then skip unconverted flags */
266	if (!fa->flags_valid) {
267		/* 1 item for the inode */
268		trans = btrfs_start_transaction(root, 1);
269		if (IS_ERR(trans))
270			return PTR_ERR(trans);
271		goto update_flags;
272	}
273
274	if (fsflags & FS_DIRSYNC_FL)
275		binode_flags |= BTRFS_INODE_DIRSYNC;
276	else
277		binode_flags &= ~BTRFS_INODE_DIRSYNC;
278	if (fsflags & FS_NOCOW_FL) {
279		if (S_ISREG(inode->i_mode)) {
280			/*
281			 * It's safe to turn csums off here, no extents exist.
282			 * Otherwise we want the flag to reflect the real COW
283			 * status of the file and will not set it.
284			 */
285			if (inode->i_size == 0)
286				binode_flags |= BTRFS_INODE_NODATACOW |
287						BTRFS_INODE_NODATASUM;
288		} else {
289			binode_flags |= BTRFS_INODE_NODATACOW;
290		}
291	} else {
292		/*
293		 * Revert back under same assumptions as above
294		 */
295		if (S_ISREG(inode->i_mode)) {
296			if (inode->i_size == 0)
297				binode_flags &= ~(BTRFS_INODE_NODATACOW |
298						  BTRFS_INODE_NODATASUM);
299		} else {
300			binode_flags &= ~BTRFS_INODE_NODATACOW;
301		}
302	}
303
304	/*
305	 * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
306	 * flag may be changed automatically if compression code won't make
307	 * things smaller.
308	 */
309	if (fsflags & FS_NOCOMP_FL) {
310		binode_flags &= ~BTRFS_INODE_COMPRESS;
311		binode_flags |= BTRFS_INODE_NOCOMPRESS;
312	} else if (fsflags & FS_COMPR_FL) {
313
314		if (IS_SWAPFILE(inode))
315			return -ETXTBSY;
316
317		binode_flags |= BTRFS_INODE_COMPRESS;
318		binode_flags &= ~BTRFS_INODE_NOCOMPRESS;
319
320		comp = btrfs_compress_type2str(fs_info->compress_type);
321		if (!comp || comp[0] == 0)
322			comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB);
323	} else {
324		binode_flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
325	}
326
327	/*
328	 * 1 for inode item
329	 * 2 for properties
330	 */
331	trans = btrfs_start_transaction(root, 3);
332	if (IS_ERR(trans))
333		return PTR_ERR(trans);
334
335	if (comp) {
336		ret = btrfs_set_prop(trans, inode, "btrfs.compression", comp,
337				     strlen(comp), 0);
338		if (ret) {
339			btrfs_abort_transaction(trans, ret);
340			goto out_end_trans;
341		}
342	} else {
343		ret = btrfs_set_prop(trans, inode, "btrfs.compression", NULL,
344				     0, 0);
345		if (ret && ret != -ENODATA) {
346			btrfs_abort_transaction(trans, ret);
347			goto out_end_trans;
348		}
349	}
350
351update_flags:
352	binode->flags = binode_flags;
353	btrfs_sync_inode_flags_to_i_flags(inode);
354	inode_inc_iversion(inode);
355	inode->i_ctime = current_time(inode);
356	ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
357
358 out_end_trans:
359	btrfs_end_transaction(trans);
360	return ret;
361}
362
363/*
364 * Start exclusive operation @type, return true on success
365 */
366bool btrfs_exclop_start(struct btrfs_fs_info *fs_info,
367			enum btrfs_exclusive_operation type)
368{
369	bool ret = false;
370
371	spin_lock(&fs_info->super_lock);
372	if (fs_info->exclusive_operation == BTRFS_EXCLOP_NONE) {
373		fs_info->exclusive_operation = type;
374		ret = true;
375	}
376	spin_unlock(&fs_info->super_lock);
377
378	return ret;
379}
380
381/*
382 * Conditionally allow to enter the exclusive operation in case it's compatible
383 * with the running one.  This must be paired with btrfs_exclop_start_unlock and
384 * btrfs_exclop_finish.
385 *
386 * Compatibility:
387 * - the same type is already running
388 * - not BTRFS_EXCLOP_NONE - this is intentionally incompatible and the caller
389 *   must check the condition first that would allow none -> @type
390 */
391bool btrfs_exclop_start_try_lock(struct btrfs_fs_info *fs_info,
392				 enum btrfs_exclusive_operation type)
393{
394	spin_lock(&fs_info->super_lock);
395	if (fs_info->exclusive_operation == type)
396		return true;
397
398	spin_unlock(&fs_info->super_lock);
399	return false;
400}
401
402void btrfs_exclop_start_unlock(struct btrfs_fs_info *fs_info)
403{
404	spin_unlock(&fs_info->super_lock);
405}
406
407void btrfs_exclop_finish(struct btrfs_fs_info *fs_info)
408{
409	spin_lock(&fs_info->super_lock);
410	WRITE_ONCE(fs_info->exclusive_operation, BTRFS_EXCLOP_NONE);
411	spin_unlock(&fs_info->super_lock);
412	sysfs_notify(&fs_info->fs_devices->fsid_kobj, NULL, "exclusive_operation");
413}
414
415static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
416{
417	struct inode *inode = file_inode(file);
418
419	return put_user(inode->i_generation, arg);
420}
421
422static noinline int btrfs_ioctl_fitrim(struct btrfs_fs_info *fs_info,
423					void __user *arg)
424{
425	struct btrfs_device *device;
426	struct request_queue *q;
427	struct fstrim_range range;
428	u64 minlen = ULLONG_MAX;
429	u64 num_devices = 0;
430	int ret;
431
432	if (!capable(CAP_SYS_ADMIN))
433		return -EPERM;
434
435	/*
436	 * btrfs_trim_block_group() depends on space cache, which is not
437	 * available in zoned filesystem. So, disallow fitrim on a zoned
438	 * filesystem for now.
439	 */
440	if (btrfs_is_zoned(fs_info))
441		return -EOPNOTSUPP;
442
443	/*
444	 * If the fs is mounted with nologreplay, which requires it to be
445	 * mounted in RO mode as well, we can not allow discard on free space
446	 * inside block groups, because log trees refer to extents that are not
447	 * pinned in a block group's free space cache (pinning the extents is
448	 * precisely the first phase of replaying a log tree).
449	 */
450	if (btrfs_test_opt(fs_info, NOLOGREPLAY))
451		return -EROFS;
452
453	rcu_read_lock();
454	list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
455				dev_list) {
456		if (!device->bdev)
457			continue;
458		q = bdev_get_queue(device->bdev);
459		if (blk_queue_discard(q)) {
460			num_devices++;
461			minlen = min_t(u64, q->limits.discard_granularity,
462				     minlen);
463		}
464	}
465	rcu_read_unlock();
466
467	if (!num_devices)
468		return -EOPNOTSUPP;
469	if (copy_from_user(&range, arg, sizeof(range)))
470		return -EFAULT;
471
472	/*
473	 * NOTE: Don't truncate the range using super->total_bytes.  Bytenr of
474	 * block group is in the logical address space, which can be any
475	 * sectorsize aligned bytenr in  the range [0, U64_MAX].
476	 */
477	if (range.len < fs_info->sb->s_blocksize)
478		return -EINVAL;
479
480	range.minlen = max(range.minlen, minlen);
481	ret = btrfs_trim_fs(fs_info, &range);
482	if (ret < 0)
483		return ret;
484
485	if (copy_to_user(arg, &range, sizeof(range)))
486		return -EFAULT;
487
488	return 0;
489}
490
491int __pure btrfs_is_empty_uuid(u8 *uuid)
492{
493	int i;
494
495	for (i = 0; i < BTRFS_UUID_SIZE; i++) {
496		if (uuid[i])
497			return 0;
498	}
499	return 1;
500}
501
502static noinline int create_subvol(struct inode *dir,
503				  struct dentry *dentry,
504				  const char *name, int namelen,
505				  struct btrfs_qgroup_inherit *inherit)
506{
507	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
508	struct btrfs_trans_handle *trans;
509	struct btrfs_key key;
510	struct btrfs_root_item *root_item;
511	struct btrfs_inode_item *inode_item;
512	struct extent_buffer *leaf;
513	struct btrfs_root *root = BTRFS_I(dir)->root;
514	struct btrfs_root *new_root;
515	struct btrfs_block_rsv block_rsv;
516	struct timespec64 cur_time = current_time(dir);
517	struct inode *inode;
518	int ret;
519	int err;
520	dev_t anon_dev = 0;
521	u64 objectid;
522	u64 index = 0;
523
524	root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
525	if (!root_item)
526		return -ENOMEM;
527
528	ret = btrfs_get_free_objectid(fs_info->tree_root, &objectid);
529	if (ret)
530		goto fail_free;
531
532	ret = get_anon_bdev(&anon_dev);
533	if (ret < 0)
534		goto fail_free;
535
536	/*
537	 * Don't create subvolume whose level is not zero. Or qgroup will be
538	 * screwed up since it assumes subvolume qgroup's level to be 0.
539	 */
540	if (btrfs_qgroup_level(objectid)) {
541		ret = -ENOSPC;
542		goto fail_free;
543	}
544
545	btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
546	/*
547	 * The same as the snapshot creation, please see the comment
548	 * of create_snapshot().
549	 */
550	ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 8, false);
551	if (ret)
552		goto fail_free;
553
554	trans = btrfs_start_transaction(root, 0);
555	if (IS_ERR(trans)) {
556		ret = PTR_ERR(trans);
557		btrfs_subvolume_release_metadata(root, &block_rsv);
558		goto fail_free;
559	}
560	trans->block_rsv = &block_rsv;
561	trans->bytes_reserved = block_rsv.size;
562
563	ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
564	if (ret)
565		goto fail;
566
567	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
568				      BTRFS_NESTING_NORMAL);
569	if (IS_ERR(leaf)) {
570		ret = PTR_ERR(leaf);
571		goto fail;
572	}
573
574	btrfs_mark_buffer_dirty(leaf);
575
576	inode_item = &root_item->inode;
577	btrfs_set_stack_inode_generation(inode_item, 1);
578	btrfs_set_stack_inode_size(inode_item, 3);
579	btrfs_set_stack_inode_nlink(inode_item, 1);
580	btrfs_set_stack_inode_nbytes(inode_item,
581				     fs_info->nodesize);
582	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
583
584	btrfs_set_root_flags(root_item, 0);
585	btrfs_set_root_limit(root_item, 0);
586	btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
587
588	btrfs_set_root_bytenr(root_item, leaf->start);
589	btrfs_set_root_generation(root_item, trans->transid);
590	btrfs_set_root_level(root_item, 0);
591	btrfs_set_root_refs(root_item, 1);
592	btrfs_set_root_used(root_item, leaf->len);
593	btrfs_set_root_last_snapshot(root_item, 0);
594
595	btrfs_set_root_generation_v2(root_item,
596			btrfs_root_generation(root_item));
597	generate_random_guid(root_item->uuid);
598	btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
599	btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
600	root_item->ctime = root_item->otime;
601	btrfs_set_root_ctransid(root_item, trans->transid);
602	btrfs_set_root_otransid(root_item, trans->transid);
603
604	btrfs_tree_unlock(leaf);
605
606	btrfs_set_root_dirid(root_item, BTRFS_FIRST_FREE_OBJECTID);
607
608	key.objectid = objectid;
609	key.offset = 0;
610	key.type = BTRFS_ROOT_ITEM_KEY;
611	ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
612				root_item);
613	if (ret) {
614		/*
615		 * Since we don't abort the transaction in this case, free the
616		 * tree block so that we don't leak space and leave the
617		 * filesystem in an inconsistent state (an extent item in the
618		 * extent tree without backreferences). Also no need to have
619		 * the tree block locked since it is not in any tree at this
620		 * point, so no other task can find it and use it.
621		 */
622		btrfs_free_tree_block(trans, root, leaf, 0, 1);
623		free_extent_buffer(leaf);
624		goto fail;
625	}
626
627	free_extent_buffer(leaf);
628	leaf = NULL;
629
630	key.offset = (u64)-1;
631	new_root = btrfs_get_new_fs_root(fs_info, objectid, anon_dev);
632	if (IS_ERR(new_root)) {
633		free_anon_bdev(anon_dev);
634		ret = PTR_ERR(new_root);
635		btrfs_abort_transaction(trans, ret);
636		goto fail;
637	}
638	/* Freeing will be done in btrfs_put_root() of new_root */
639	anon_dev = 0;
640
641	ret = btrfs_record_root_in_trans(trans, new_root);
642	if (ret) {
643		btrfs_put_root(new_root);
644		btrfs_abort_transaction(trans, ret);
645		goto fail;
646	}
647
648	ret = btrfs_create_subvol_root(trans, new_root, root);
649	btrfs_put_root(new_root);
650	if (ret) {
651		/* We potentially lose an unused inode item here */
652		btrfs_abort_transaction(trans, ret);
653		goto fail;
654	}
655
656	/*
657	 * insert the directory item
658	 */
659	ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
660	if (ret) {
661		btrfs_abort_transaction(trans, ret);
662		goto fail;
663	}
664
665	ret = btrfs_insert_dir_item(trans, name, namelen, BTRFS_I(dir), &key,
666				    BTRFS_FT_DIR, index);
667	if (ret) {
668		btrfs_abort_transaction(trans, ret);
669		goto fail;
670	}
671
672	btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
673	ret = btrfs_update_inode(trans, root, BTRFS_I(dir));
674	if (ret) {
675		btrfs_abort_transaction(trans, ret);
676		goto fail;
677	}
678
679	ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid,
680				 btrfs_ino(BTRFS_I(dir)), index, name, namelen);
681	if (ret) {
682		btrfs_abort_transaction(trans, ret);
683		goto fail;
684	}
685
686	ret = btrfs_uuid_tree_add(trans, root_item->uuid,
687				  BTRFS_UUID_KEY_SUBVOL, objectid);
688	if (ret)
689		btrfs_abort_transaction(trans, ret);
690
691fail:
692	kfree(root_item);
693	trans->block_rsv = NULL;
694	trans->bytes_reserved = 0;
695	btrfs_subvolume_release_metadata(root, &block_rsv);
696
697	err = btrfs_commit_transaction(trans);
698	if (err && !ret)
699		ret = err;
700
701	if (!ret) {
702		inode = btrfs_lookup_dentry(dir, dentry);
703		if (IS_ERR(inode))
704			return PTR_ERR(inode);
705		d_instantiate(dentry, inode);
706	}
707	return ret;
708
709fail_free:
710	if (anon_dev)
711		free_anon_bdev(anon_dev);
712	kfree(root_item);
713	return ret;
714}
715
716static int create_snapshot(struct btrfs_root *root, struct inode *dir,
717			   struct dentry *dentry, bool readonly,
718			   struct btrfs_qgroup_inherit *inherit)
719{
720	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
721	struct inode *inode;
722	struct btrfs_pending_snapshot *pending_snapshot;
723	struct btrfs_trans_handle *trans;
724	int ret;
725
726	if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
727		return -EINVAL;
728
729	if (atomic_read(&root->nr_swapfiles)) {
730		btrfs_warn(fs_info,
731			   "cannot snapshot subvolume with active swapfile");
732		return -ETXTBSY;
733	}
734
735	pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
736	if (!pending_snapshot)
737		return -ENOMEM;
738
739	ret = get_anon_bdev(&pending_snapshot->anon_dev);
740	if (ret < 0)
741		goto free_pending;
742	pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
743			GFP_KERNEL);
744	pending_snapshot->path = btrfs_alloc_path();
745	if (!pending_snapshot->root_item || !pending_snapshot->path) {
746		ret = -ENOMEM;
747		goto free_pending;
748	}
749
750	btrfs_init_block_rsv(&pending_snapshot->block_rsv,
751			     BTRFS_BLOCK_RSV_TEMP);
752	/*
753	 * 1 - parent dir inode
754	 * 2 - dir entries
755	 * 1 - root item
756	 * 2 - root ref/backref
757	 * 1 - root of snapshot
758	 * 1 - UUID item
759	 */
760	ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
761					&pending_snapshot->block_rsv, 8,
762					false);
763	if (ret)
764		goto free_pending;
765
766	pending_snapshot->dentry = dentry;
767	pending_snapshot->root = root;
768	pending_snapshot->readonly = readonly;
769	pending_snapshot->dir = dir;
770	pending_snapshot->inherit = inherit;
771
772	trans = btrfs_start_transaction(root, 0);
773	if (IS_ERR(trans)) {
774		ret = PTR_ERR(trans);
775		goto fail;
776	}
777
778	spin_lock(&fs_info->trans_lock);
779	list_add(&pending_snapshot->list,
780		 &trans->transaction->pending_snapshots);
781	spin_unlock(&fs_info->trans_lock);
782
783	ret = btrfs_commit_transaction(trans);
784	if (ret)
785		goto fail;
786
787	ret = pending_snapshot->error;
788	if (ret)
789		goto fail;
790
791	ret = btrfs_orphan_cleanup(pending_snapshot->snap);
792	if (ret)
793		goto fail;
794
795	inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
796	if (IS_ERR(inode)) {
797		ret = PTR_ERR(inode);
798		goto fail;
799	}
800
801	d_instantiate(dentry, inode);
802	ret = 0;
803	pending_snapshot->anon_dev = 0;
804fail:
805	/* Prevent double freeing of anon_dev */
806	if (ret && pending_snapshot->snap)
807		pending_snapshot->snap->anon_dev = 0;
808	btrfs_put_root(pending_snapshot->snap);
809	btrfs_subvolume_release_metadata(root, &pending_snapshot->block_rsv);
810free_pending:
811	if (pending_snapshot->anon_dev)
812		free_anon_bdev(pending_snapshot->anon_dev);
813	kfree(pending_snapshot->root_item);
814	btrfs_free_path(pending_snapshot->path);
815	kfree(pending_snapshot);
816
817	return ret;
818}
819
820/*  copy of may_delete in fs/namei.c()
821 *	Check whether we can remove a link victim from directory dir, check
822 *  whether the type of victim is right.
823 *  1. We can't do it if dir is read-only (done in permission())
824 *  2. We should have write and exec permissions on dir
825 *  3. We can't remove anything from append-only dir
826 *  4. We can't do anything with immutable dir (done in permission())
827 *  5. If the sticky bit on dir is set we should either
828 *	a. be owner of dir, or
829 *	b. be owner of victim, or
830 *	c. have CAP_FOWNER capability
831 *  6. If the victim is append-only or immutable we can't do anything with
832 *     links pointing to it.
833 *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
834 *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
835 *  9. We can't remove a root or mountpoint.
836 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
837 *     nfs_async_unlink().
838 */
839
840static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
841{
842	int error;
843
844	if (d_really_is_negative(victim))
845		return -ENOENT;
846
847	BUG_ON(d_inode(victim->d_parent) != dir);
848	audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
849
850	error = inode_permission(&init_user_ns, dir, MAY_WRITE | MAY_EXEC);
851	if (error)
852		return error;
853	if (IS_APPEND(dir))
854		return -EPERM;
855	if (check_sticky(&init_user_ns, dir, d_inode(victim)) ||
856	    IS_APPEND(d_inode(victim)) || IS_IMMUTABLE(d_inode(victim)) ||
857	    IS_SWAPFILE(d_inode(victim)))
858		return -EPERM;
859	if (isdir) {
860		if (!d_is_dir(victim))
861			return -ENOTDIR;
862		if (IS_ROOT(victim))
863			return -EBUSY;
864	} else if (d_is_dir(victim))
865		return -EISDIR;
866	if (IS_DEADDIR(dir))
867		return -ENOENT;
868	if (victim->d_flags & DCACHE_NFSFS_RENAMED)
869		return -EBUSY;
870	return 0;
871}
872
873/* copy of may_create in fs/namei.c() */
874static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
875{
876	if (d_really_is_positive(child))
877		return -EEXIST;
878	if (IS_DEADDIR(dir))
879		return -ENOENT;
880	return inode_permission(&init_user_ns, dir, MAY_WRITE | MAY_EXEC);
881}
882
883/*
884 * Create a new subvolume below @parent.  This is largely modeled after
885 * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
886 * inside this filesystem so it's quite a bit simpler.
887 */
888static noinline int btrfs_mksubvol(const struct path *parent,
889				   const char *name, int namelen,
890				   struct btrfs_root *snap_src,
891				   bool readonly,
892				   struct btrfs_qgroup_inherit *inherit)
893{
894	struct inode *dir = d_inode(parent->dentry);
895	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
896	struct dentry *dentry;
897	int error;
898
899	error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
900	if (error == -EINTR)
901		return error;
902
903	dentry = lookup_one_len(name, parent->dentry, namelen);
904	error = PTR_ERR(dentry);
905	if (IS_ERR(dentry))
906		goto out_unlock;
907
908	error = btrfs_may_create(dir, dentry);
909	if (error)
910		goto out_dput;
911
912	/*
913	 * even if this name doesn't exist, we may get hash collisions.
914	 * check for them now when we can safely fail
915	 */
916	error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
917					       dir->i_ino, name,
918					       namelen);
919	if (error)
920		goto out_dput;
921
922	down_read(&fs_info->subvol_sem);
923
924	if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
925		goto out_up_read;
926
927	if (snap_src)
928		error = create_snapshot(snap_src, dir, dentry, readonly, inherit);
929	else
930		error = create_subvol(dir, dentry, name, namelen, inherit);
931
932	if (!error)
933		fsnotify_mkdir(dir, dentry);
934out_up_read:
935	up_read(&fs_info->subvol_sem);
936out_dput:
937	dput(dentry);
938out_unlock:
939	btrfs_inode_unlock(dir, 0);
940	return error;
941}
942
943static noinline int btrfs_mksnapshot(const struct path *parent,
944				   const char *name, int namelen,
945				   struct btrfs_root *root,
946				   bool readonly,
947				   struct btrfs_qgroup_inherit *inherit)
948{
949	int ret;
950	bool snapshot_force_cow = false;
951
952	/*
953	 * Force new buffered writes to reserve space even when NOCOW is
954	 * possible. This is to avoid later writeback (running dealloc) to
955	 * fallback to COW mode and unexpectedly fail with ENOSPC.
956	 */
957	btrfs_drew_read_lock(&root->snapshot_lock);
958
959	ret = btrfs_start_delalloc_snapshot(root, false);
960	if (ret)
961		goto out;
962
963	/*
964	 * All previous writes have started writeback in NOCOW mode, so now
965	 * we force future writes to fallback to COW mode during snapshot
966	 * creation.
967	 */
968	atomic_inc(&root->snapshot_force_cow);
969	snapshot_force_cow = true;
970
971	btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
972
973	ret = btrfs_mksubvol(parent, name, namelen,
974			     root, readonly, inherit);
975out:
976	if (snapshot_force_cow)
977		atomic_dec(&root->snapshot_force_cow);
978	btrfs_drew_read_unlock(&root->snapshot_lock);
979	return ret;
980}
981
982/*
983 * When we're defragging a range, we don't want to kick it off again
984 * if it is really just waiting for delalloc to send it down.
985 * If we find a nice big extent or delalloc range for the bytes in the
986 * file you want to defrag, we return 0 to let you know to skip this
987 * part of the file
988 */
989static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
990{
991	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
992	struct extent_map *em = NULL;
993	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
994	u64 end;
995
996	read_lock(&em_tree->lock);
997	em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
998	read_unlock(&em_tree->lock);
999
1000	if (em) {
1001		end = extent_map_end(em);
1002		free_extent_map(em);
1003		if (end - offset > thresh)
1004			return 0;
1005	}
1006	/* if we already have a nice delalloc here, just stop */
1007	thresh /= 2;
1008	end = count_range_bits(io_tree, &offset, offset + thresh,
1009			       thresh, EXTENT_DELALLOC, 1);
1010	if (end >= thresh)
1011		return 0;
1012	return 1;
1013}
1014
1015/*
1016 * helper function to walk through a file and find extents
1017 * newer than a specific transid, and smaller than thresh.
1018 *
1019 * This is used by the defragging code to find new and small
1020 * extents
1021 */
1022static int find_new_extents(struct btrfs_root *root,
1023			    struct inode *inode, u64 newer_than,
1024			    u64 *off, u32 thresh)
1025{
1026	struct btrfs_path *path;
1027	struct btrfs_key min_key;
1028	struct extent_buffer *leaf;
1029	struct btrfs_file_extent_item *extent;
1030	int type;
1031	int ret;
1032	u64 ino = btrfs_ino(BTRFS_I(inode));
1033
1034	path = btrfs_alloc_path();
1035	if (!path)
1036		return -ENOMEM;
1037
1038	min_key.objectid = ino;
1039	min_key.type = BTRFS_EXTENT_DATA_KEY;
1040	min_key.offset = *off;
1041
1042	while (1) {
1043		ret = btrfs_search_forward(root, &min_key, path, newer_than);
1044		if (ret != 0)
1045			goto none;
1046process_slot:
1047		if (min_key.objectid != ino)
1048			goto none;
1049		if (min_key.type != BTRFS_EXTENT_DATA_KEY)
1050			goto none;
1051
1052		leaf = path->nodes[0];
1053		extent = btrfs_item_ptr(leaf, path->slots[0],
1054					struct btrfs_file_extent_item);
1055
1056		type = btrfs_file_extent_type(leaf, extent);
1057		if (type == BTRFS_FILE_EXTENT_REG &&
1058		    btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
1059		    check_defrag_in_cache(inode, min_key.offset, thresh)) {
1060			*off = min_key.offset;
1061			btrfs_free_path(path);
1062			return 0;
1063		}
1064
1065		path->slots[0]++;
1066		if (path->slots[0] < btrfs_header_nritems(leaf)) {
1067			btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
1068			goto process_slot;
1069		}
1070
1071		if (min_key.offset == (u64)-1)
1072			goto none;
1073
1074		min_key.offset++;
1075		btrfs_release_path(path);
1076	}
1077none:
1078	btrfs_free_path(path);
1079	return -ENOENT;
1080}
1081
1082static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1083{
1084	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1085	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1086	struct extent_map *em;
1087	u64 len = PAGE_SIZE;
1088
1089	/*
1090	 * hopefully we have this extent in the tree already, try without
1091	 * the full extent lock
1092	 */
1093	read_lock(&em_tree->lock);
1094	em = lookup_extent_mapping(em_tree, start, len);
1095	read_unlock(&em_tree->lock);
1096
1097	if (!em) {
1098		struct extent_state *cached = NULL;
1099		u64 end = start + len - 1;
1100
1101		/* get the big lock and read metadata off disk */
1102		lock_extent_bits(io_tree, start, end, &cached);
1103		em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
1104		unlock_extent_cached(io_tree, start, end, &cached);
1105
1106		if (IS_ERR(em))
1107			return NULL;
1108	}
1109
1110	return em;
1111}
1112
1113static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1114{
1115	struct extent_map *next;
1116	bool ret = true;
1117
1118	/* this is the last extent */
1119	if (em->start + em->len >= i_size_read(inode))
1120		return false;
1121
1122	next = defrag_lookup_extent(inode, em->start + em->len);
1123	if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1124		ret = false;
1125	else if ((em->block_start + em->block_len == next->block_start) &&
1126		 (em->block_len > SZ_128K && next->block_len > SZ_128K))
1127		ret = false;
1128
1129	free_extent_map(next);
1130	return ret;
1131}
1132
1133static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1134			       u64 *last_len, u64 *skip, u64 *defrag_end,
1135			       int compress)
1136{
1137	struct extent_map *em;
1138	int ret = 1;
1139	bool next_mergeable = true;
1140	bool prev_mergeable = true;
1141
1142	/*
1143	 * make sure that once we start defragging an extent, we keep on
1144	 * defragging it
1145	 */
1146	if (start < *defrag_end)
1147		return 1;
1148
1149	*skip = 0;
1150
1151	em = defrag_lookup_extent(inode, start);
1152	if (!em)
1153		return 0;
1154
1155	/* this will cover holes, and inline extents */
1156	if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1157		ret = 0;
1158		goto out;
1159	}
1160
1161	if (!*defrag_end)
1162		prev_mergeable = false;
1163
1164	next_mergeable = defrag_check_next_extent(inode, em);
1165	/*
1166	 * we hit a real extent, if it is big or the next extent is not a
1167	 * real extent, don't bother defragging it
1168	 */
1169	if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1170	    (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1171		ret = 0;
1172out:
1173	/*
1174	 * last_len ends up being a counter of how many bytes we've defragged.
1175	 * every time we choose not to defrag an extent, we reset *last_len
1176	 * so that the next tiny extent will force a defrag.
1177	 *
1178	 * The end result of this is that tiny extents before a single big
1179	 * extent will force at least part of that big extent to be defragged.
1180	 */
1181	if (ret) {
1182		*defrag_end = extent_map_end(em);
1183	} else {
1184		*last_len = 0;
1185		*skip = extent_map_end(em);
1186		*defrag_end = 0;
1187	}
1188
1189	free_extent_map(em);
1190	return ret;
1191}
1192
1193/*
1194 * it doesn't do much good to defrag one or two pages
1195 * at a time.  This pulls in a nice chunk of pages
1196 * to COW and defrag.
1197 *
1198 * It also makes sure the delalloc code has enough
1199 * dirty data to avoid making new small extents as part
1200 * of the defrag
1201 *
1202 * It's a good idea to start RA on this range
1203 * before calling this.
1204 */
1205static int cluster_pages_for_defrag(struct inode *inode,
1206				    struct page **pages,
1207				    unsigned long start_index,
1208				    unsigned long num_pages)
1209{
1210	unsigned long file_end;
1211	u64 isize = i_size_read(inode);
1212	u64 page_start;
1213	u64 page_end;
1214	u64 page_cnt;
1215	u64 start = (u64)start_index << PAGE_SHIFT;
1216	u64 search_start;
1217	int ret;
1218	int i;
1219	int i_done;
1220	struct btrfs_ordered_extent *ordered;
1221	struct extent_state *cached_state = NULL;
1222	struct extent_io_tree *tree;
1223	struct extent_changeset *data_reserved = NULL;
1224	gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1225
1226	file_end = (isize - 1) >> PAGE_SHIFT;
1227	if (!isize || start_index > file_end)
1228		return 0;
1229
1230	page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1231
1232	ret = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
1233			start, page_cnt << PAGE_SHIFT);
1234	if (ret)
1235		return ret;
1236	i_done = 0;
1237	tree = &BTRFS_I(inode)->io_tree;
1238
1239	/* step one, lock all the pages */
1240	for (i = 0; i < page_cnt; i++) {
1241		struct page *page;
1242again:
1243		page = find_or_create_page(inode->i_mapping,
1244					   start_index + i, mask);
1245		if (!page)
1246			break;
1247
1248		ret = set_page_extent_mapped(page);
1249		if (ret < 0) {
1250			unlock_page(page);
1251			put_page(page);
1252			break;
1253		}
1254
1255		page_start = page_offset(page);
1256		page_end = page_start + PAGE_SIZE - 1;
1257		while (1) {
1258			lock_extent_bits(tree, page_start, page_end,
1259					 &cached_state);
1260			ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode),
1261							      page_start);
1262			unlock_extent_cached(tree, page_start, page_end,
1263					     &cached_state);
1264			if (!ordered)
1265				break;
1266
1267			unlock_page(page);
1268			btrfs_start_ordered_extent(ordered, 1);
1269			btrfs_put_ordered_extent(ordered);
1270			lock_page(page);
1271			/*
1272			 * we unlocked the page above, so we need check if
1273			 * it was released or not.
1274			 */
1275			if (page->mapping != inode->i_mapping) {
1276				unlock_page(page);
1277				put_page(page);
1278				goto again;
1279			}
1280		}
1281
1282		if (!PageUptodate(page)) {
1283			btrfs_readpage(NULL, page);
1284			lock_page(page);
1285			if (!PageUptodate(page)) {
1286				unlock_page(page);
1287				put_page(page);
1288				ret = -EIO;
1289				break;
1290			}
1291		}
1292
1293		if (page->mapping != inode->i_mapping) {
1294			unlock_page(page);
1295			put_page(page);
1296			goto again;
1297		}
1298
1299		pages[i] = page;
1300		i_done++;
1301	}
1302	if (!i_done || ret)
1303		goto out;
1304
1305	if (!(inode->i_sb->s_flags & SB_ACTIVE))
1306		goto out;
1307
1308	/*
1309	 * so now we have a nice long stream of locked
1310	 * and up to date pages, lets wait on them
1311	 */
1312	for (i = 0; i < i_done; i++)
1313		wait_on_page_writeback(pages[i]);
1314
1315	page_start = page_offset(pages[0]);
1316	page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1317
1318	lock_extent_bits(&BTRFS_I(inode)->io_tree,
1319			 page_start, page_end - 1, &cached_state);
1320
1321	/*
1322	 * When defragmenting we skip ranges that have holes or inline extents,
1323	 * (check should_defrag_range()), to avoid unnecessary IO and wasting
1324	 * space. At btrfs_defrag_file(), we check if a range should be defragged
1325	 * before locking the inode and then, if it should, we trigger a sync
1326	 * page cache readahead - we lock the inode only after that to avoid
1327	 * blocking for too long other tasks that possibly want to operate on
1328	 * other file ranges. But before we were able to get the inode lock,
1329	 * some other task may have punched a hole in the range, or we may have
1330	 * now an inline extent, in which case we should not defrag. So check
1331	 * for that here, where we have the inode and the range locked, and bail
1332	 * out if that happened.
1333	 */
1334	search_start = page_start;
1335	while (search_start < page_end) {
1336		struct extent_map *em;
1337
1338		em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, search_start,
1339				      page_end - search_start);
1340		if (IS_ERR(em)) {
1341			ret = PTR_ERR(em);
1342			goto out_unlock_range;
1343		}
1344		if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1345			free_extent_map(em);
1346			/* Ok, 0 means we did not defrag anything */
1347			ret = 0;
1348			goto out_unlock_range;
1349		}
1350		search_start = extent_map_end(em);
1351		free_extent_map(em);
1352	}
1353
1354	clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1355			  page_end - 1, EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
1356			  EXTENT_DEFRAG, 0, 0, &cached_state);
1357
1358	if (i_done != page_cnt) {
1359		spin_lock(&BTRFS_I(inode)->lock);
1360		btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1361		spin_unlock(&BTRFS_I(inode)->lock);
1362		btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
1363				start, (page_cnt - i_done) << PAGE_SHIFT, true);
1364	}
1365
1366
1367	set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1368			  &cached_state);
1369
1370	unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1371			     page_start, page_end - 1, &cached_state);
1372
1373	for (i = 0; i < i_done; i++) {
1374		clear_page_dirty_for_io(pages[i]);
1375		ClearPageChecked(pages[i]);
1376		set_page_dirty(pages[i]);
1377		unlock_page(pages[i]);
1378		put_page(pages[i]);
1379	}
1380	btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1381	extent_changeset_free(data_reserved);
1382	return i_done;
1383
1384out_unlock_range:
1385	unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1386			     page_start, page_end - 1, &cached_state);
1387out:
1388	for (i = 0; i < i_done; i++) {
1389		unlock_page(pages[i]);
1390		put_page(pages[i]);
1391	}
1392	btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
1393			start, page_cnt << PAGE_SHIFT, true);
1394	btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1395	extent_changeset_free(data_reserved);
1396	return ret;
1397
1398}
1399
1400int btrfs_defrag_file(struct inode *inode, struct file *file,
1401		      struct btrfs_ioctl_defrag_range_args *range,
1402		      u64 newer_than, unsigned long max_to_defrag)
1403{
1404	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1405	struct btrfs_root *root = BTRFS_I(inode)->root;
1406	struct file_ra_state *ra = NULL;
1407	unsigned long last_index;
1408	u64 isize = i_size_read(inode);
1409	u64 last_len = 0;
1410	u64 skip = 0;
1411	u64 defrag_end = 0;
1412	u64 newer_off = range->start;
1413	unsigned long i;
1414	unsigned long ra_index = 0;
1415	int ret;
1416	int defrag_count = 0;
1417	int compress_type = BTRFS_COMPRESS_ZLIB;
1418	u32 extent_thresh = range->extent_thresh;
1419	unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1420	unsigned long cluster = max_cluster;
1421	u64 new_align = ~((u64)SZ_128K - 1);
1422	struct page **pages = NULL;
1423	bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1424
1425	if (isize == 0)
1426		return 0;
1427
1428	if (range->start >= isize)
1429		return -EINVAL;
1430
1431	if (do_compress) {
1432		if (range->compress_type >= BTRFS_NR_COMPRESS_TYPES)
1433			return -EINVAL;
1434		if (range->compress_type)
1435			compress_type = range->compress_type;
1436	}
1437
1438	if (extent_thresh == 0)
1439		extent_thresh = SZ_256K;
1440
1441	/*
1442	 * If we were not given a file, allocate a readahead context. As
1443	 * readahead is just an optimization, defrag will work without it so
1444	 * we don't error out.
1445	 */
1446	if (!file) {
1447		ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1448		if (ra)
1449			file_ra_state_init(ra, inode->i_mapping);
1450	} else {
1451		ra = &file->f_ra;
1452	}
1453
1454	pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1455	if (!pages) {
1456		ret = -ENOMEM;
1457		goto out_ra;
1458	}
1459
1460	/* find the last page to defrag */
1461	if (range->start + range->len > range->start) {
1462		last_index = min_t(u64, isize - 1,
1463			 range->start + range->len - 1) >> PAGE_SHIFT;
1464	} else {
1465		last_index = (isize - 1) >> PAGE_SHIFT;
1466	}
1467
1468	if (newer_than) {
1469		ret = find_new_extents(root, inode, newer_than,
1470				       &newer_off, SZ_64K);
1471		if (!ret) {
1472			range->start = newer_off;
1473			/*
1474			 * we always align our defrag to help keep
1475			 * the extents in the file evenly spaced
1476			 */
1477			i = (newer_off & new_align) >> PAGE_SHIFT;
1478		} else
1479			goto out_ra;
1480	} else {
1481		i = range->start >> PAGE_SHIFT;
1482	}
1483	if (!max_to_defrag)
1484		max_to_defrag = last_index - i + 1;
1485
1486	/*
1487	 * make writeback starts from i, so the defrag range can be
1488	 * written sequentially.
1489	 */
1490	if (i < inode->i_mapping->writeback_index)
1491		inode->i_mapping->writeback_index = i;
1492
1493	while (i <= last_index && defrag_count < max_to_defrag &&
1494	       (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1495		/*
1496		 * make sure we stop running if someone unmounts
1497		 * the FS
1498		 */
1499		if (!(inode->i_sb->s_flags & SB_ACTIVE))
1500			break;
1501
1502		if (btrfs_defrag_cancelled(fs_info)) {
1503			btrfs_debug(fs_info, "defrag_file cancelled");
1504			ret = -EAGAIN;
1505			goto error;
1506		}
1507
1508		if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1509					 extent_thresh, &last_len, &skip,
1510					 &defrag_end, do_compress)){
1511			unsigned long next;
1512			/*
1513			 * the should_defrag function tells us how much to skip
1514			 * bump our counter by the suggested amount
1515			 */
1516			next = DIV_ROUND_UP(skip, PAGE_SIZE);
1517			i = max(i + 1, next);
1518			continue;
1519		}
1520
1521		if (!newer_than) {
1522			cluster = (PAGE_ALIGN(defrag_end) >>
1523				   PAGE_SHIFT) - i;
1524			cluster = min(cluster, max_cluster);
1525		} else {
1526			cluster = max_cluster;
1527		}
1528
1529		if (i + cluster > ra_index) {
1530			ra_index = max(i, ra_index);
1531			if (ra)
1532				page_cache_sync_readahead(inode->i_mapping, ra,
1533						file, ra_index, cluster);
1534			ra_index += cluster;
1535		}
1536
1537		btrfs_inode_lock(inode, 0);
1538		if (IS_SWAPFILE(inode)) {
1539			ret = -ETXTBSY;
1540		} else {
1541			if (do_compress)
1542				BTRFS_I(inode)->defrag_compress = compress_type;
1543			ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1544		}
1545		if (ret < 0) {
1546			btrfs_inode_unlock(inode, 0);
1547			goto out_ra;
1548		}
1549
1550		defrag_count += ret;
1551		balance_dirty_pages_ratelimited(inode->i_mapping);
1552		btrfs_inode_unlock(inode, 0);
1553
1554		if (newer_than) {
1555			if (newer_off == (u64)-1)
1556				break;
1557
1558			if (ret > 0)
1559				i += ret;
1560
1561			newer_off = max(newer_off + 1,
1562					(u64)i << PAGE_SHIFT);
1563
1564			ret = find_new_extents(root, inode, newer_than,
1565					       &newer_off, SZ_64K);
1566			if (!ret) {
1567				range->start = newer_off;
1568				i = (newer_off & new_align) >> PAGE_SHIFT;
1569			} else {
1570				break;
1571			}
1572		} else {
1573			if (ret > 0) {
1574				i += ret;
1575				last_len += ret << PAGE_SHIFT;
1576			} else {
1577				i++;
1578				last_len = 0;
1579			}
1580		}
1581	}
1582
1583	ret = defrag_count;
1584error:
1585	if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1586		filemap_flush(inode->i_mapping);
1587		if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1588			     &BTRFS_I(inode)->runtime_flags))
1589			filemap_flush(inode->i_mapping);
1590	}
1591
1592	if (range->compress_type == BTRFS_COMPRESS_LZO) {
1593		btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1594	} else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1595		btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1596	}
1597
1598out_ra:
1599	if (do_compress) {
1600		btrfs_inode_lock(inode, 0);
1601		BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1602		btrfs_inode_unlock(inode, 0);
1603	}
1604	if (!file)
1605		kfree(ra);
1606	kfree(pages);
1607	return ret;
1608}
1609
1610/*
1611 * Try to start exclusive operation @type or cancel it if it's running.
1612 *
1613 * Return:
1614 *   0        - normal mode, newly claimed op started
1615 *  >0        - normal mode, something else is running,
1616 *              return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS to user space
1617 * ECANCELED  - cancel mode, successful cancel
1618 * ENOTCONN   - cancel mode, operation not running anymore
1619 */
1620static int exclop_start_or_cancel_reloc(struct btrfs_fs_info *fs_info,
1621			enum btrfs_exclusive_operation type, bool cancel)
1622{
1623	if (!cancel) {
1624		/* Start normal op */
1625		if (!btrfs_exclop_start(fs_info, type))
1626			return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1627		/* Exclusive operation is now claimed */
1628		return 0;
1629	}
1630
1631	/* Cancel running op */
1632	if (btrfs_exclop_start_try_lock(fs_info, type)) {
1633		/*
1634		 * This blocks any exclop finish from setting it to NONE, so we
1635		 * request cancellation. Either it runs and we will wait for it,
1636		 * or it has finished and no waiting will happen.
1637		 */
1638		atomic_inc(&fs_info->reloc_cancel_req);
1639		btrfs_exclop_start_unlock(fs_info);
1640
1641		if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
1642			wait_on_bit(&fs_info->flags, BTRFS_FS_RELOC_RUNNING,
1643				    TASK_INTERRUPTIBLE);
1644
1645		return -ECANCELED;
1646	}
1647
1648	/* Something else is running or none */
1649	return -ENOTCONN;
1650}
1651
1652static noinline int btrfs_ioctl_resize(struct file *file,
1653					void __user *arg)
1654{
1655	struct inode *inode = file_inode(file);
1656	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1657	u64 new_size;
1658	u64 old_size;
1659	u64 devid = 1;
1660	struct btrfs_root *root = BTRFS_I(inode)->root;
1661	struct btrfs_ioctl_vol_args *vol_args;
1662	struct btrfs_trans_handle *trans;
1663	struct btrfs_device *device = NULL;
1664	char *sizestr;
1665	char *retptr;
1666	char *devstr = NULL;
1667	int ret = 0;
1668	int mod = 0;
1669	bool cancel;
1670
1671	if (!capable(CAP_SYS_ADMIN))
1672		return -EPERM;
1673
1674	ret = mnt_want_write_file(file);
1675	if (ret)
1676		return ret;
1677
1678	/*
1679	 * Read the arguments before checking exclusivity to be able to
1680	 * distinguish regular resize and cancel
1681	 */
1682	vol_args = memdup_user(arg, sizeof(*vol_args));
1683	if (IS_ERR(vol_args)) {
1684		ret = PTR_ERR(vol_args);
1685		goto out_drop;
1686	}
1687	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1688	sizestr = vol_args->name;
1689	cancel = (strcmp("cancel", sizestr) == 0);
1690	ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_RESIZE, cancel);
1691	if (ret)
1692		goto out_free;
1693	/* Exclusive operation is now claimed */
1694
1695	devstr = strchr(sizestr, ':');
1696	if (devstr) {
1697		sizestr = devstr + 1;
1698		*devstr = '\0';
1699		devstr = vol_args->name;
1700		ret = kstrtoull(devstr, 10, &devid);
1701		if (ret)
1702			goto out_finish;
1703		if (!devid) {
1704			ret = -EINVAL;
1705			goto out_finish;
1706		}
1707		btrfs_info(fs_info, "resizing devid %llu", devid);
1708	}
1709
1710	device = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL);
1711	if (!device) {
1712		btrfs_info(fs_info, "resizer unable to find device %llu",
1713			   devid);
1714		ret = -ENODEV;
1715		goto out_finish;
1716	}
1717
1718	if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1719		btrfs_info(fs_info,
1720			   "resizer unable to apply on readonly device %llu",
1721		       devid);
1722		ret = -EPERM;
1723		goto out_finish;
1724	}
1725
1726	if (!strcmp(sizestr, "max"))
1727		new_size = device->bdev->bd_inode->i_size;
1728	else {
1729		if (sizestr[0] == '-') {
1730			mod = -1;
1731			sizestr++;
1732		} else if (sizestr[0] == '+') {
1733			mod = 1;
1734			sizestr++;
1735		}
1736		new_size = memparse(sizestr, &retptr);
1737		if (*retptr != '\0' || new_size == 0) {
1738			ret = -EINVAL;
1739			goto out_finish;
1740		}
1741	}
1742
1743	if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1744		ret = -EPERM;
1745		goto out_finish;
1746	}
1747
1748	old_size = btrfs_device_get_total_bytes(device);
1749
1750	if (mod < 0) {
1751		if (new_size > old_size) {
1752			ret = -EINVAL;
1753			goto out_finish;
1754		}
1755		new_size = old_size - new_size;
1756	} else if (mod > 0) {
1757		if (new_size > ULLONG_MAX - old_size) {
1758			ret = -ERANGE;
1759			goto out_finish;
1760		}
1761		new_size = old_size + new_size;
1762	}
1763
1764	if (new_size < SZ_256M) {
1765		ret = -EINVAL;
1766		goto out_finish;
1767	}
1768	if (new_size > device->bdev->bd_inode->i_size) {
1769		ret = -EFBIG;
1770		goto out_finish;
1771	}
1772
1773	new_size = round_down(new_size, fs_info->sectorsize);
1774
1775	if (new_size > old_size) {
1776		trans = btrfs_start_transaction(root, 0);
1777		if (IS_ERR(trans)) {
1778			ret = PTR_ERR(trans);
1779			goto out_finish;
1780		}
1781		ret = btrfs_grow_device(trans, device, new_size);
1782		btrfs_commit_transaction(trans);
1783	} else if (new_size < old_size) {
1784		ret = btrfs_shrink_device(device, new_size);
1785	} /* equal, nothing need to do */
1786
1787	if (ret == 0 && new_size != old_size)
1788		btrfs_info_in_rcu(fs_info,
1789			"resize device %s (devid %llu) from %llu to %llu",
1790			rcu_str_deref(device->name), device->devid,
1791			old_size, new_size);
1792out_finish:
1793	btrfs_exclop_finish(fs_info);
1794out_free:
1795	kfree(vol_args);
1796out_drop:
1797	mnt_drop_write_file(file);
1798	return ret;
1799}
1800
1801static noinline int __btrfs_ioctl_snap_create(struct file *file,
1802				const char *name, unsigned long fd, int subvol,
1803				bool readonly,
1804				struct btrfs_qgroup_inherit *inherit)
1805{
1806	int namelen;
1807	int ret = 0;
1808
1809	if (!S_ISDIR(file_inode(file)->i_mode))
1810		return -ENOTDIR;
1811
1812	ret = mnt_want_write_file(file);
1813	if (ret)
1814		goto out;
1815
1816	namelen = strlen(name);
1817	if (strchr(name, '/')) {
1818		ret = -EINVAL;
1819		goto out_drop_write;
1820	}
1821
1822	if (name[0] == '.' &&
1823	   (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1824		ret = -EEXIST;
1825		goto out_drop_write;
1826	}
1827
1828	if (subvol) {
1829		ret = btrfs_mksubvol(&file->f_path, name, namelen,
1830				     NULL, readonly, inherit);
1831	} else {
1832		struct fd src = fdget(fd);
1833		struct inode *src_inode;
1834		if (!src.file) {
1835			ret = -EINVAL;
1836			goto out_drop_write;
1837		}
1838
1839		src_inode = file_inode(src.file);
1840		if (src_inode->i_sb != file_inode(file)->i_sb) {
1841			btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1842				   "Snapshot src from another FS");
1843			ret = -EXDEV;
1844		} else if (!inode_owner_or_capable(&init_user_ns, src_inode)) {
1845			/*
1846			 * Subvolume creation is not restricted, but snapshots
1847			 * are limited to own subvolumes only
1848			 */
1849			ret = -EPERM;
1850		} else {
1851			ret = btrfs_mksnapshot(&file->f_path, name, namelen,
1852					     BTRFS_I(src_inode)->root,
1853					     readonly, inherit);
1854		}
1855		fdput(src);
1856	}
1857out_drop_write:
1858	mnt_drop_write_file(file);
1859out:
1860	return ret;
1861}
1862
1863static noinline int btrfs_ioctl_snap_create(struct file *file,
1864					    void __user *arg, int subvol)
1865{
1866	struct btrfs_ioctl_vol_args *vol_args;
1867	int ret;
1868
1869	if (!S_ISDIR(file_inode(file)->i_mode))
1870		return -ENOTDIR;
1871
1872	vol_args = memdup_user(arg, sizeof(*vol_args));
1873	if (IS_ERR(vol_args))
1874		return PTR_ERR(vol_args);
1875	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1876
1877	ret = __btrfs_ioctl_snap_create(file, vol_args->name, vol_args->fd,
1878					subvol, false, NULL);
1879
1880	kfree(vol_args);
1881	return ret;
1882}
1883
1884static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1885					       void __user *arg, int subvol)
1886{
1887	struct btrfs_ioctl_vol_args_v2 *vol_args;
1888	int ret;
1889	bool readonly = false;
1890	struct btrfs_qgroup_inherit *inherit = NULL;
1891
1892	if (!S_ISDIR(file_inode(file)->i_mode))
1893		return -ENOTDIR;
1894
1895	vol_args = memdup_user(arg, sizeof(*vol_args));
1896	if (IS_ERR(vol_args))
1897		return PTR_ERR(vol_args);
1898	vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1899
1900	if (vol_args->flags & ~BTRFS_SUBVOL_CREATE_ARGS_MASK) {
1901		ret = -EOPNOTSUPP;
1902		goto free_args;
1903	}
1904
1905	if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1906		readonly = true;
1907	if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1908		u64 nums;
1909
1910		if (vol_args->size < sizeof(*inherit) ||
1911		    vol_args->size > PAGE_SIZE) {
1912			ret = -EINVAL;
1913			goto free_args;
1914		}
1915		inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1916		if (IS_ERR(inherit)) {
1917			ret = PTR_ERR(inherit);
1918			goto free_args;
1919		}
1920
1921		if (inherit->num_qgroups > PAGE_SIZE ||
1922		    inherit->num_ref_copies > PAGE_SIZE ||
1923		    inherit->num_excl_copies > PAGE_SIZE) {
1924			ret = -EINVAL;
1925			goto free_inherit;
1926		}
1927
1928		nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
1929		       2 * inherit->num_excl_copies;
1930		if (vol_args->size != struct_size(inherit, qgroups, nums)) {
1931			ret = -EINVAL;
1932			goto free_inherit;
1933		}
1934	}
1935
1936	ret = __btrfs_ioctl_snap_create(file, vol_args->name, vol_args->fd,
1937					subvol, readonly, inherit);
1938	if (ret)
1939		goto free_inherit;
1940free_inherit:
1941	kfree(inherit);
1942free_args:
1943	kfree(vol_args);
1944	return ret;
1945}
1946
1947static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1948						void __user *arg)
1949{
1950	struct inode *inode = file_inode(file);
1951	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1952	struct btrfs_root *root = BTRFS_I(inode)->root;
1953	int ret = 0;
1954	u64 flags = 0;
1955
1956	if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1957		return -EINVAL;
1958
1959	down_read(&fs_info->subvol_sem);
1960	if (btrfs_root_readonly(root))
1961		flags |= BTRFS_SUBVOL_RDONLY;
1962	up_read(&fs_info->subvol_sem);
1963
1964	if (copy_to_user(arg, &flags, sizeof(flags)))
1965		ret = -EFAULT;
1966
1967	return ret;
1968}
1969
1970static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1971					      void __user *arg)
1972{
1973	struct inode *inode = file_inode(file);
1974	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1975	struct btrfs_root *root = BTRFS_I(inode)->root;
1976	struct btrfs_trans_handle *trans;
1977	u64 root_flags;
1978	u64 flags;
1979	int ret = 0;
1980
1981	if (!inode_owner_or_capable(&init_user_ns, inode))
1982		return -EPERM;
1983
1984	ret = mnt_want_write_file(file);
1985	if (ret)
1986		goto out;
1987
1988	if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1989		ret = -EINVAL;
1990		goto out_drop_write;
1991	}
1992
1993	if (copy_from_user(&flags, arg, sizeof(flags))) {
1994		ret = -EFAULT;
1995		goto out_drop_write;
1996	}
1997
1998	if (flags & ~BTRFS_SUBVOL_RDONLY) {
1999		ret = -EOPNOTSUPP;
2000		goto out_drop_write;
2001	}
2002
2003	down_write(&fs_info->subvol_sem);
2004
2005	/* nothing to do */
2006	if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
2007		goto out_drop_sem;
2008
2009	root_flags = btrfs_root_flags(&root->root_item);
2010	if (flags & BTRFS_SUBVOL_RDONLY) {
2011		btrfs_set_root_flags(&root->root_item,
2012				     root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
2013	} else {
2014		/*
2015		 * Block RO -> RW transition if this subvolume is involved in
2016		 * send
2017		 */
2018		spin_lock(&root->root_item_lock);
2019		if (root->send_in_progress == 0) {
2020			btrfs_set_root_flags(&root->root_item,
2021				     root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
2022			spin_unlock(&root->root_item_lock);
2023		} else {
2024			spin_unlock(&root->root_item_lock);
2025			btrfs_warn(fs_info,
2026				   "Attempt to set subvolume %llu read-write during send",
2027				   root->root_key.objectid);
2028			ret = -EPERM;
2029			goto out_drop_sem;
2030		}
2031	}
2032
2033	trans = btrfs_start_transaction(root, 1);
2034	if (IS_ERR(trans)) {
2035		ret = PTR_ERR(trans);
2036		goto out_reset;
2037	}
2038
2039	ret = btrfs_update_root(trans, fs_info->tree_root,
2040				&root->root_key, &root->root_item);
2041	if (ret < 0) {
2042		btrfs_end_transaction(trans);
2043		goto out_reset;
2044	}
2045
2046	ret = btrfs_commit_transaction(trans);
2047
2048out_reset:
2049	if (ret)
2050		btrfs_set_root_flags(&root->root_item, root_flags);
2051out_drop_sem:
2052	up_write(&fs_info->subvol_sem);
2053out_drop_write:
2054	mnt_drop_write_file(file);
2055out:
2056	return ret;
2057}
2058
2059static noinline int key_in_sk(struct btrfs_key *key,
2060			      struct btrfs_ioctl_search_key *sk)
2061{
2062	struct btrfs_key test;
2063	int ret;
2064
2065	test.objectid = sk->min_objectid;
2066	test.type = sk->min_type;
2067	test.offset = sk->min_offset;
2068
2069	ret = btrfs_comp_cpu_keys(key, &test);
2070	if (ret < 0)
2071		return 0;
2072
2073	test.objectid = sk->max_objectid;
2074	test.type = sk->max_type;
2075	test.offset = sk->max_offset;
2076
2077	ret = btrfs_comp_cpu_keys(key, &test);
2078	if (ret > 0)
2079		return 0;
2080	return 1;
2081}
2082
2083static noinline int copy_to_sk(struct btrfs_path *path,
2084			       struct btrfs_key *key,
2085			       struct btrfs_ioctl_search_key *sk,
2086			       size_t *buf_size,
2087			       char __user *ubuf,
2088			       unsigned long *sk_offset,
2089			       int *num_found)
2090{
2091	u64 found_transid;
2092	struct extent_buffer *leaf;
2093	struct btrfs_ioctl_search_header sh;
2094	struct btrfs_key test;
2095	unsigned long item_off;
2096	unsigned long item_len;
2097	int nritems;
2098	int i;
2099	int slot;
2100	int ret = 0;
2101
2102	leaf = path->nodes[0];
2103	slot = path->slots[0];
2104	nritems = btrfs_header_nritems(leaf);
2105
2106	if (btrfs_header_generation(leaf) > sk->max_transid) {
2107		i = nritems;
2108		goto advance_key;
2109	}
2110	found_transid = btrfs_header_generation(leaf);
2111
2112	for (i = slot; i < nritems; i++) {
2113		item_off = btrfs_item_ptr_offset(leaf, i);
2114		item_len = btrfs_item_size_nr(leaf, i);
2115
2116		btrfs_item_key_to_cpu(leaf, key, i);
2117		if (!key_in_sk(key, sk))
2118			continue;
2119
2120		if (sizeof(sh) + item_len > *buf_size) {
2121			if (*num_found) {
2122				ret = 1;
2123				goto out;
2124			}
2125
2126			/*
2127			 * return one empty item back for v1, which does not
2128			 * handle -EOVERFLOW
2129			 */
2130
2131			*buf_size = sizeof(sh) + item_len;
2132			item_len = 0;
2133			ret = -EOVERFLOW;
2134		}
2135
2136		if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2137			ret = 1;
2138			goto out;
2139		}
2140
2141		sh.objectid = key->objectid;
2142		sh.offset = key->offset;
2143		sh.type = key->type;
2144		sh.len = item_len;
2145		sh.transid = found_transid;
2146
2147		/*
2148		 * Copy search result header. If we fault then loop again so we
2149		 * can fault in the pages and -EFAULT there if there's a
2150		 * problem. Otherwise we'll fault and then copy the buffer in
2151		 * properly this next time through
2152		 */
2153		if (copy_to_user_nofault(ubuf + *sk_offset, &sh, sizeof(sh))) {
2154			ret = 0;
2155			goto out;
2156		}
2157
2158		*sk_offset += sizeof(sh);
2159
2160		if (item_len) {
2161			char __user *up = ubuf + *sk_offset;
2162			/*
2163			 * Copy the item, same behavior as above, but reset the
2164			 * * sk_offset so we copy the full thing again.
2165			 */
2166			if (read_extent_buffer_to_user_nofault(leaf, up,
2167						item_off, item_len)) {
2168				ret = 0;
2169				*sk_offset -= sizeof(sh);
2170				goto out;
2171			}
2172
2173			*sk_offset += item_len;
2174		}
2175		(*num_found)++;
2176
2177		if (ret) /* -EOVERFLOW from above */
2178			goto out;
2179
2180		if (*num_found >= sk->nr_items) {
2181			ret = 1;
2182			goto out;
2183		}
2184	}
2185advance_key:
2186	ret = 0;
2187	test.objectid = sk->max_objectid;
2188	test.type = sk->max_type;
2189	test.offset = sk->max_offset;
2190	if (btrfs_comp_cpu_keys(key, &test) >= 0)
2191		ret = 1;
2192	else if (key->offset < (u64)-1)
2193		key->offset++;
2194	else if (key->type < (u8)-1) {
2195		key->offset = 0;
2196		key->type++;
2197	} else if (key->objectid < (u64)-1) {
2198		key->offset = 0;
2199		key->type = 0;
2200		key->objectid++;
2201	} else
2202		ret = 1;
2203out:
2204	/*
2205	 *  0: all items from this leaf copied, continue with next
2206	 *  1: * more items can be copied, but unused buffer is too small
2207	 *     * all items were found
2208	 *     Either way, it will stops the loop which iterates to the next
2209	 *     leaf
2210	 *  -EOVERFLOW: item was to large for buffer
2211	 *  -EFAULT: could not copy extent buffer back to userspace
2212	 */
2213	return ret;
2214}
2215
2216static noinline int search_ioctl(struct inode *inode,
2217				 struct btrfs_ioctl_search_key *sk,
2218				 size_t *buf_size,
2219				 char __user *ubuf)
2220{
2221	struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2222	struct btrfs_root *root;
2223	struct btrfs_key key;
2224	struct btrfs_path *path;
2225	int ret;
2226	int num_found = 0;
2227	unsigned long sk_offset = 0;
2228
2229	if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2230		*buf_size = sizeof(struct btrfs_ioctl_search_header);
2231		return -EOVERFLOW;
2232	}
2233
2234	path = btrfs_alloc_path();
2235	if (!path)
2236		return -ENOMEM;
2237
2238	if (sk->tree_id == 0) {
2239		/* search the root of the inode that was passed */
2240		root = btrfs_grab_root(BTRFS_I(inode)->root);
2241	} else {
2242		root = btrfs_get_fs_root(info, sk->tree_id, true);
2243		if (IS_ERR(root)) {
2244			btrfs_free_path(path);
2245			return PTR_ERR(root);
2246		}
2247	}
2248
2249	key.objectid = sk->min_objectid;
2250	key.type = sk->min_type;
2251	key.offset = sk->min_offset;
2252
2253	while (1) {
2254		ret = fault_in_pages_writeable(ubuf + sk_offset,
2255					       *buf_size - sk_offset);
2256		if (ret)
2257			break;
2258
2259		ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2260		if (ret != 0) {
2261			if (ret > 0)
2262				ret = 0;
2263			goto err;
2264		}
2265		ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2266				 &sk_offset, &num_found);
2267		btrfs_release_path(path);
2268		if (ret)
2269			break;
2270
2271	}
2272	if (ret > 0)
2273		ret = 0;
2274err:
2275	sk->nr_items = num_found;
2276	btrfs_put_root(root);
2277	btrfs_free_path(path);
2278	return ret;
2279}
2280
2281static noinline int btrfs_ioctl_tree_search(struct file *file,
2282					   void __user *argp)
2283{
2284	struct btrfs_ioctl_search_args __user *uargs;
2285	struct btrfs_ioctl_search_key sk;
2286	struct inode *inode;
2287	int ret;
2288	size_t buf_size;
2289
2290	if (!capable(CAP_SYS_ADMIN))
2291		return -EPERM;
2292
2293	uargs = (struct btrfs_ioctl_search_args __user *)argp;
2294
2295	if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2296		return -EFAULT;
2297
2298	buf_size = sizeof(uargs->buf);
2299
2300	inode = file_inode(file);
2301	ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2302
2303	/*
2304	 * In the origin implementation an overflow is handled by returning a
2305	 * search header with a len of zero, so reset ret.
2306	 */
2307	if (ret == -EOVERFLOW)
2308		ret = 0;
2309
2310	if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2311		ret = -EFAULT;
2312	return ret;
2313}
2314
2315static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2316					       void __user *argp)
2317{
2318	struct btrfs_ioctl_search_args_v2 __user *uarg;
2319	struct btrfs_ioctl_search_args_v2 args;
2320	struct inode *inode;
2321	int ret;
2322	size_t buf_size;
2323	const size_t buf_limit = SZ_16M;
2324
2325	if (!capable(CAP_SYS_ADMIN))
2326		return -EPERM;
2327
2328	/* copy search header and buffer size */
2329	uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2330	if (copy_from_user(&args, uarg, sizeof(args)))
2331		return -EFAULT;
2332
2333	buf_size = args.buf_size;
2334
2335	/* limit result size to 16MB */
2336	if (buf_size > buf_limit)
2337		buf_size = buf_limit;
2338
2339	inode = file_inode(file);
2340	ret = search_ioctl(inode, &args.key, &buf_size,
2341			   (char __user *)(&uarg->buf[0]));
2342	if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2343		ret = -EFAULT;
2344	else if (ret == -EOVERFLOW &&
2345		copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2346		ret = -EFAULT;
2347
2348	return ret;
2349}
2350
2351/*
2352 * Search INODE_REFs to identify path name of 'dirid' directory
2353 * in a 'tree_id' tree. and sets path name to 'name'.
2354 */
2355static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2356				u64 tree_id, u64 dirid, char *name)
2357{
2358	struct btrfs_root *root;
2359	struct btrfs_key key;
2360	char *ptr;
2361	int ret = -1;
2362	int slot;
2363	int len;
2364	int total_len = 0;
2365	struct btrfs_inode_ref *iref;
2366	struct extent_buffer *l;
2367	struct btrfs_path *path;
2368
2369	if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2370		name[0]='\0';
2371		return 0;
2372	}
2373
2374	path = btrfs_alloc_path();
2375	if (!path)
2376		return -ENOMEM;
2377
2378	ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2379
2380	root = btrfs_get_fs_root(info, tree_id, true);
2381	if (IS_ERR(root)) {
2382		ret = PTR_ERR(root);
2383		root = NULL;
2384		goto out;
2385	}
2386
2387	key.objectid = dirid;
2388	key.type = BTRFS_INODE_REF_KEY;
2389	key.offset = (u64)-1;
2390
2391	while (1) {
2392		ret = btrfs_search_backwards(root, &key, path);
2393		if (ret < 0)
2394			goto out;
2395		else if (ret > 0) {
2396			ret = -ENOENT;
2397			goto out;
2398		}
2399
2400		l = path->nodes[0];
2401		slot = path->slots[0];
2402
2403		iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2404		len = btrfs_inode_ref_name_len(l, iref);
2405		ptr -= len + 1;
2406		total_len += len + 1;
2407		if (ptr < name) {
2408			ret = -ENAMETOOLONG;
2409			goto out;
2410		}
2411
2412		*(ptr + len) = '/';
2413		read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2414
2415		if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2416			break;
2417
2418		btrfs_release_path(path);
2419		key.objectid = key.offset;
2420		key.offset = (u64)-1;
2421		dirid = key.objectid;
2422	}
2423	memmove(name, ptr, total_len);
2424	name[total_len] = '\0';
2425	ret = 0;
2426out:
2427	btrfs_put_root(root);
2428	btrfs_free_path(path);
2429	return ret;
2430}
2431
2432static int btrfs_search_path_in_tree_user(struct inode *inode,
2433				struct btrfs_ioctl_ino_lookup_user_args *args)
2434{
2435	struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2436	struct super_block *sb = inode->i_sb;
2437	struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2438	u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2439	u64 dirid = args->dirid;
2440	unsigned long item_off;
2441	unsigned long item_len;
2442	struct btrfs_inode_ref *iref;
2443	struct btrfs_root_ref *rref;
2444	struct btrfs_root *root = NULL;
2445	struct btrfs_path *path;
2446	struct btrfs_key key, key2;
2447	struct extent_buffer *leaf;
2448	struct inode *temp_inode;
2449	char *ptr;
2450	int slot;
2451	int len;
2452	int total_len = 0;
2453	int ret;
2454
2455	path = btrfs_alloc_path();
2456	if (!path)
2457		return -ENOMEM;
2458
2459	/*
2460	 * If the bottom subvolume does not exist directly under upper_limit,
2461	 * construct the path in from the bottom up.
2462	 */
2463	if (dirid != upper_limit.objectid) {
2464		ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2465
2466		root = btrfs_get_fs_root(fs_info, treeid, true);
2467		if (IS_ERR(root)) {
2468			ret = PTR_ERR(root);
2469			goto out;
2470		}
2471
2472		key.objectid = dirid;
2473		key.type = BTRFS_INODE_REF_KEY;
2474		key.offset = (u64)-1;
2475		while (1) {
2476			ret = btrfs_search_backwards(root, &key, path);
2477			if (ret < 0)
2478				goto out_put;
2479			else if (ret > 0) {
2480				ret = -ENOENT;
2481				goto out_put;
2482			}
2483
2484			leaf = path->nodes[0];
2485			slot = path->slots[0];
2486
2487			iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2488			len = btrfs_inode_ref_name_len(leaf, iref);
2489			ptr -= len + 1;
2490			total_len += len + 1;
2491			if (ptr < args->path) {
2492				ret = -ENAMETOOLONG;
2493				goto out_put;
2494			}
2495
2496			*(ptr + len) = '/';
2497			read_extent_buffer(leaf, ptr,
2498					(unsigned long)(iref + 1), len);
2499
2500			/* Check the read+exec permission of this directory */
2501			ret = btrfs_previous_item(root, path, dirid,
2502						  BTRFS_INODE_ITEM_KEY);
2503			if (ret < 0) {
2504				goto out_put;
2505			} else if (ret > 0) {
2506				ret = -ENOENT;
2507				goto out_put;
2508			}
2509
2510			leaf = path->nodes[0];
2511			slot = path->slots[0];
2512			btrfs_item_key_to_cpu(leaf, &key2, slot);
2513			if (key2.objectid != dirid) {
2514				ret = -ENOENT;
2515				goto out_put;
2516			}
2517
2518			temp_inode = btrfs_iget(sb, key2.objectid, root);
2519			if (IS_ERR(temp_inode)) {
2520				ret = PTR_ERR(temp_inode);
2521				goto out_put;
2522			}
2523			ret = inode_permission(&init_user_ns, temp_inode,
2524					       MAY_READ | MAY_EXEC);
2525			iput(temp_inode);
2526			if (ret) {
2527				ret = -EACCES;
2528				goto out_put;
2529			}
2530
2531			if (key.offset == upper_limit.objectid)
2532				break;
2533			if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2534				ret = -EACCES;
2535				goto out_put;
2536			}
2537
2538			btrfs_release_path(path);
2539			key.objectid = key.offset;
2540			key.offset = (u64)-1;
2541			dirid = key.objectid;
2542		}
2543
2544		memmove(args->path, ptr, total_len);
2545		args->path[total_len] = '\0';
2546		btrfs_put_root(root);
2547		root = NULL;
2548		btrfs_release_path(path);
2549	}
2550
2551	/* Get the bottom subvolume's name from ROOT_REF */
2552	key.objectid = treeid;
2553	key.type = BTRFS_ROOT_REF_KEY;
2554	key.offset = args->treeid;
2555	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2556	if (ret < 0) {
2557		goto out;
2558	} else if (ret > 0) {
2559		ret = -ENOENT;
2560		goto out;
2561	}
2562
2563	leaf = path->nodes[0];
2564	slot = path->slots[0];
2565	btrfs_item_key_to_cpu(leaf, &key, slot);
2566
2567	item_off = btrfs_item_ptr_offset(leaf, slot);
2568	item_len = btrfs_item_size_nr(leaf, slot);
2569	/* Check if dirid in ROOT_REF corresponds to passed dirid */
2570	rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2571	if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2572		ret = -EINVAL;
2573		goto out;
2574	}
2575
2576	/* Copy subvolume's name */
2577	item_off += sizeof(struct btrfs_root_ref);
2578	item_len -= sizeof(struct btrfs_root_ref);
2579	read_extent_buffer(leaf, args->name, item_off, item_len);
2580	args->name[item_len] = 0;
2581
2582out_put:
2583	btrfs_put_root(root);
2584out:
2585	btrfs_free_path(path);
2586	return ret;
2587}
2588
2589static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2590					   void __user *argp)
2591{
2592	struct btrfs_ioctl_ino_lookup_args *args;
2593	struct inode *inode;
2594	int ret = 0;
2595
2596	args = memdup_user(argp, sizeof(*args));
2597	if (IS_ERR(args))
2598		return PTR_ERR(args);
2599
2600	inode = file_inode(file);
2601
2602	/*
2603	 * Unprivileged query to obtain the containing subvolume root id. The
2604	 * path is reset so it's consistent with btrfs_search_path_in_tree.
2605	 */
2606	if (args->treeid == 0)
2607		args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2608
2609	if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2610		args->name[0] = 0;
2611		goto out;
2612	}
2613
2614	if (!capable(CAP_SYS_ADMIN)) {
2615		ret = -EPERM;
2616		goto out;
2617	}
2618
2619	ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2620					args->treeid, args->objectid,
2621					args->name);
2622
2623out:
2624	if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2625		ret = -EFAULT;
2626
2627	kfree(args);
2628	return ret;
2629}
2630
2631/*
2632 * Version of ino_lookup ioctl (unprivileged)
2633 *
2634 * The main differences from ino_lookup ioctl are:
2635 *
2636 *   1. Read + Exec permission will be checked using inode_permission() during
2637 *      path construction. -EACCES will be returned in case of failure.
2638 *   2. Path construction will be stopped at the inode number which corresponds
2639 *      to the fd with which this ioctl is called. If constructed path does not
2640 *      exist under fd's inode, -EACCES will be returned.
2641 *   3. The name of bottom subvolume is also searched and filled.
2642 */
2643static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2644{
2645	struct btrfs_ioctl_ino_lookup_user_args *args;
2646	struct inode *inode;
2647	int ret;
2648
2649	args = memdup_user(argp, sizeof(*args));
2650	if (IS_ERR(args))
2651		return PTR_ERR(args);
2652
2653	inode = file_inode(file);
2654
2655	if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2656	    BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2657		/*
2658		 * The subvolume does not exist under fd with which this is
2659		 * called
2660		 */
2661		kfree(args);
2662		return -EACCES;
2663	}
2664
2665	ret = btrfs_search_path_in_tree_user(inode, args);
2666
2667	if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2668		ret = -EFAULT;
2669
2670	kfree(args);
2671	return ret;
2672}
2673
2674/* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2675static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp)
2676{
2677	struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2678	struct btrfs_fs_info *fs_info;
2679	struct btrfs_root *root;
2680	struct btrfs_path *path;
2681	struct btrfs_key key;
2682	struct btrfs_root_item *root_item;
2683	struct btrfs_root_ref *rref;
2684	struct extent_buffer *leaf;
2685	unsigned long item_off;
2686	unsigned long item_len;
2687	struct inode *inode;
2688	int slot;
2689	int ret = 0;
2690
2691	path = btrfs_alloc_path();
2692	if (!path)
2693		return -ENOMEM;
2694
2695	subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2696	if (!subvol_info) {
2697		btrfs_free_path(path);
2698		return -ENOMEM;
2699	}
2700
2701	inode = file_inode(file);
2702	fs_info = BTRFS_I(inode)->root->fs_info;
2703
2704	/* Get root_item of inode's subvolume */
2705	key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2706	root = btrfs_get_fs_root(fs_info, key.objectid, true);
2707	if (IS_ERR(root)) {
2708		ret = PTR_ERR(root);
2709		goto out_free;
2710	}
2711	root_item = &root->root_item;
2712
2713	subvol_info->treeid = key.objectid;
2714
2715	subvol_info->generation = btrfs_root_generation(root_item);
2716	subvol_info->flags = btrfs_root_flags(root_item);
2717
2718	memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2719	memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2720						    BTRFS_UUID_SIZE);
2721	memcpy(subvol_info->received_uuid, root_item->received_uuid,
2722						    BTRFS_UUID_SIZE);
2723
2724	subvol_info->ctransid = btrfs_root_ctransid(root_item);
2725	subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2726	subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2727
2728	subvol_info->otransid = btrfs_root_otransid(root_item);
2729	subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2730	subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2731
2732	subvol_info->stransid = btrfs_root_stransid(root_item);
2733	subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2734	subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2735
2736	subvol_info->rtransid = btrfs_root_rtransid(root_item);
2737	subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2738	subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2739
2740	if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2741		/* Search root tree for ROOT_BACKREF of this subvolume */
2742		key.type = BTRFS_ROOT_BACKREF_KEY;
2743		key.offset = 0;
2744		ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2745		if (ret < 0) {
2746			goto out;
2747		} else if (path->slots[0] >=
2748			   btrfs_header_nritems(path->nodes[0])) {
2749			ret = btrfs_next_leaf(fs_info->tree_root, path);
2750			if (ret < 0) {
2751				goto out;
2752			} else if (ret > 0) {
2753				ret = -EUCLEAN;
2754				goto out;
2755			}
2756		}
2757
2758		leaf = path->nodes[0];
2759		slot = path->slots[0];
2760		btrfs_item_key_to_cpu(leaf, &key, slot);
2761		if (key.objectid == subvol_info->treeid &&
2762		    key.type == BTRFS_ROOT_BACKREF_KEY) {
2763			subvol_info->parent_id = key.offset;
2764
2765			rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2766			subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2767
2768			item_off = btrfs_item_ptr_offset(leaf, slot)
2769					+ sizeof(struct btrfs_root_ref);
2770			item_len = btrfs_item_size_nr(leaf, slot)
2771					- sizeof(struct btrfs_root_ref);
2772			read_extent_buffer(leaf, subvol_info->name,
2773					   item_off, item_len);
2774		} else {
2775			ret = -ENOENT;
2776			goto out;
2777		}
2778	}
2779
2780	if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2781		ret = -EFAULT;
2782
2783out:
2784	btrfs_put_root(root);
2785out_free:
2786	btrfs_free_path(path);
2787	kfree(subvol_info);
2788	return ret;
2789}
2790
2791/*
2792 * Return ROOT_REF information of the subvolume containing this inode
2793 * except the subvolume name.
2794 */
2795static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp)
2796{
2797	struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2798	struct btrfs_root_ref *rref;
2799	struct btrfs_root *root;
2800	struct btrfs_path *path;
2801	struct btrfs_key key;
2802	struct extent_buffer *leaf;
2803	struct inode *inode;
2804	u64 objectid;
2805	int slot;
2806	int ret;
2807	u8 found;
2808
2809	path = btrfs_alloc_path();
2810	if (!path)
2811		return -ENOMEM;
2812
2813	rootrefs = memdup_user(argp, sizeof(*rootrefs));
2814	if (IS_ERR(rootrefs)) {
2815		btrfs_free_path(path);
2816		return PTR_ERR(rootrefs);
2817	}
2818
2819	inode = file_inode(file);
2820	root = BTRFS_I(inode)->root->fs_info->tree_root;
2821	objectid = BTRFS_I(inode)->root->root_key.objectid;
2822
2823	key.objectid = objectid;
2824	key.type = BTRFS_ROOT_REF_KEY;
2825	key.offset = rootrefs->min_treeid;
2826	found = 0;
2827
2828	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2829	if (ret < 0) {
2830		goto out;
2831	} else if (path->slots[0] >=
2832		   btrfs_header_nritems(path->nodes[0])) {
2833		ret = btrfs_next_leaf(root, path);
2834		if (ret < 0) {
2835			goto out;
2836		} else if (ret > 0) {
2837			ret = -EUCLEAN;
2838			goto out;
2839		}
2840	}
2841	while (1) {
2842		leaf = path->nodes[0];
2843		slot = path->slots[0];
2844
2845		btrfs_item_key_to_cpu(leaf, &key, slot);
2846		if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2847			ret = 0;
2848			goto out;
2849		}
2850
2851		if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2852			ret = -EOVERFLOW;
2853			goto out;
2854		}
2855
2856		rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2857		rootrefs->rootref[found].treeid = key.offset;
2858		rootrefs->rootref[found].dirid =
2859				  btrfs_root_ref_dirid(leaf, rref);
2860		found++;
2861
2862		ret = btrfs_next_item(root, path);
2863		if (ret < 0) {
2864			goto out;
2865		} else if (ret > 0) {
2866			ret = -EUCLEAN;
2867			goto out;
2868		}
2869	}
2870
2871out:
2872	if (!ret || ret == -EOVERFLOW) {
2873		rootrefs->num_items = found;
2874		/* update min_treeid for next search */
2875		if (found)
2876			rootrefs->min_treeid =
2877				rootrefs->rootref[found - 1].treeid + 1;
2878		if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2879			ret = -EFAULT;
2880	}
2881
2882	kfree(rootrefs);
2883	btrfs_free_path(path);
2884
2885	return ret;
2886}
2887
2888static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2889					     void __user *arg,
2890					     bool destroy_v2)
2891{
2892	struct dentry *parent = file->f_path.dentry;
2893	struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2894	struct dentry *dentry;
2895	struct inode *dir = d_inode(parent);
2896	struct inode *inode;
2897	struct btrfs_root *root = BTRFS_I(dir)->root;
2898	struct btrfs_root *dest = NULL;
2899	struct btrfs_ioctl_vol_args *vol_args = NULL;
2900	struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL;
2901	char *subvol_name, *subvol_name_ptr = NULL;
2902	int subvol_namelen;
2903	int err = 0;
2904	bool destroy_parent = false;
2905
2906	if (destroy_v2) {
2907		vol_args2 = memdup_user(arg, sizeof(*vol_args2));
2908		if (IS_ERR(vol_args2))
2909			return PTR_ERR(vol_args2);
2910
2911		if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) {
2912			err = -EOPNOTSUPP;
2913			goto out;
2914		}
2915
2916		/*
2917		 * If SPEC_BY_ID is not set, we are looking for the subvolume by
2918		 * name, same as v1 currently does.
2919		 */
2920		if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) {
2921			vol_args2->name[BTRFS_SUBVOL_NAME_MAX] = 0;
2922			subvol_name = vol_args2->name;
2923
2924			err = mnt_want_write_file(file);
2925			if (err)
2926				goto out;
2927		} else {
2928			if (vol_args2->subvolid < BTRFS_FIRST_FREE_OBJECTID) {
2929				err = -EINVAL;
2930				goto out;
2931			}
2932
2933			err = mnt_want_write_file(file);
2934			if (err)
2935				goto out;
2936
2937			dentry = btrfs_get_dentry(fs_info->sb,
2938					BTRFS_FIRST_FREE_OBJECTID,
2939					vol_args2->subvolid, 0, 0);
2940			if (IS_ERR(dentry)) {
2941				err = PTR_ERR(dentry);
2942				goto out_drop_write;
2943			}
2944
2945			/*
2946			 * Change the default parent since the subvolume being
2947			 * deleted can be outside of the current mount point.
2948			 */
2949			parent = btrfs_get_parent(dentry);
2950
2951			/*
2952			 * At this point dentry->d_name can point to '/' if the
2953			 * subvolume we want to destroy is outsite of the
2954			 * current mount point, so we need to release the
2955			 * current dentry and execute the lookup to return a new
2956			 * one with ->d_name pointing to the
2957			 * <mount point>/subvol_name.
2958			 */
2959			dput(dentry);
2960			if (IS_ERR(parent)) {
2961				err = PTR_ERR(parent);
2962				goto out_drop_write;
2963			}
2964			dir = d_inode(parent);
2965
2966			/*
2967			 * If v2 was used with SPEC_BY_ID, a new parent was
2968			 * allocated since the subvolume can be outside of the
2969			 * current mount point. Later on we need to release this
2970			 * new parent dentry.
2971			 */
2972			destroy_parent = true;
2973
2974			subvol_name_ptr = btrfs_get_subvol_name_from_objectid(
2975						fs_info, vol_args2->subvolid);
2976			if (IS_ERR(subvol_name_ptr)) {
2977				err = PTR_ERR(subvol_name_ptr);
2978				goto free_parent;
2979			}
2980			/* subvol_name_ptr is already nul terminated */
2981			subvol_name = (char *)kbasename(subvol_name_ptr);
2982		}
2983	} else {
2984		vol_args = memdup_user(arg, sizeof(*vol_args));
2985		if (IS_ERR(vol_args))
2986			return PTR_ERR(vol_args);
2987
2988		vol_args->name[BTRFS_PATH_NAME_MAX] = 0;
2989		subvol_name = vol_args->name;
2990
2991		err = mnt_want_write_file(file);
2992		if (err)
2993			goto out;
2994	}
2995
2996	subvol_namelen = strlen(subvol_name);
2997
2998	if (strchr(subvol_name, '/') ||
2999	    strncmp(subvol_name, "..", subvol_namelen) == 0) {
3000		err = -EINVAL;
3001		goto free_subvol_name;
3002	}
3003
3004	if (!S_ISDIR(dir->i_mode)) {
3005		err = -ENOTDIR;
3006		goto free_subvol_name;
3007	}
3008
3009	err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
3010	if (err == -EINTR)
3011		goto free_subvol_name;
3012	dentry = lookup_one_len(subvol_name, parent, subvol_namelen);
3013	if (IS_ERR(dentry)) {
3014		err = PTR_ERR(dentry);
3015		goto out_unlock_dir;
3016	}
3017
3018	if (d_really_is_negative(dentry)) {
3019		err = -ENOENT;
3020		goto out_dput;
3021	}
3022
3023	inode = d_inode(dentry);
3024	dest = BTRFS_I(inode)->root;
3025	if (!capable(CAP_SYS_ADMIN)) {
3026		/*
3027		 * Regular user.  Only allow this with a special mount
3028		 * option, when the user has write+exec access to the
3029		 * subvol root, and when rmdir(2) would have been
3030		 * allowed.
3031		 *
3032		 * Note that this is _not_ check that the subvol is
3033		 * empty or doesn't contain data that we wouldn't
3034		 * otherwise be able to delete.
3035		 *
3036		 * Users who want to delete empty subvols should try
3037		 * rmdir(2).
3038		 */
3039		err = -EPERM;
3040		if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
3041			goto out_dput;
3042
3043		/*
3044		 * Do not allow deletion if the parent dir is the same
3045		 * as the dir to be deleted.  That means the ioctl
3046		 * must be called on the dentry referencing the root
3047		 * of the subvol, not a random directory contained
3048		 * within it.
3049		 */
3050		err = -EINVAL;
3051		if (root == dest)
3052			goto out_dput;
3053
3054		err = inode_permission(&init_user_ns, inode,
3055				       MAY_WRITE | MAY_EXEC);
3056		if (err)
3057			goto out_dput;
3058	}
3059
3060	/* check if subvolume may be deleted by a user */
3061	err = btrfs_may_delete(dir, dentry, 1);
3062	if (err)
3063		goto out_dput;
3064
3065	if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
3066		err = -EINVAL;
3067		goto out_dput;
3068	}
3069
3070	btrfs_inode_lock(inode, 0);
3071	err = btrfs_delete_subvolume(dir, dentry);
3072	btrfs_inode_unlock(inode, 0);
3073	if (!err) {
3074		fsnotify_rmdir(dir, dentry);
3075		d_delete(dentry);
3076	}
3077
3078out_dput:
3079	dput(dentry);
3080out_unlock_dir:
3081	btrfs_inode_unlock(dir, 0);
3082free_subvol_name:
3083	kfree(subvol_name_ptr);
3084free_parent:
3085	if (destroy_parent)
3086		dput(parent);
3087out_drop_write:
3088	mnt_drop_write_file(file);
3089out:
3090	kfree(vol_args2);
3091	kfree(vol_args);
3092	return err;
3093}
3094
3095static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
3096{
3097	struct inode *inode = file_inode(file);
3098	struct btrfs_root *root = BTRFS_I(inode)->root;
3099	struct btrfs_ioctl_defrag_range_args *range;
3100	int ret;
3101
3102	ret = mnt_want_write_file(file);
3103	if (ret)
3104		return ret;
3105
3106	if (btrfs_root_readonly(root)) {
3107		ret = -EROFS;
3108		goto out;
3109	}
3110
3111	/* Subpage defrag will be supported in later commits */
3112	if (root->fs_info->sectorsize < PAGE_SIZE) {
3113		ret = -ENOTTY;
3114		goto out;
3115	}
3116
3117	switch (inode->i_mode & S_IFMT) {
3118	case S_IFDIR:
3119		if (!capable(CAP_SYS_ADMIN)) {
3120			ret = -EPERM;
3121			goto out;
3122		}
3123		ret = btrfs_defrag_root(root);
3124		break;
3125	case S_IFREG:
3126		/*
3127		 * Note that this does not check the file descriptor for write
3128		 * access. This prevents defragmenting executables that are
3129		 * running and allows defrag on files open in read-only mode.
3130		 */
3131		if (!capable(CAP_SYS_ADMIN) &&
3132		    inode_permission(&init_user_ns, inode, MAY_WRITE)) {
3133			ret = -EPERM;
3134			goto out;
3135		}
3136
3137		range = kzalloc(sizeof(*range), GFP_KERNEL);
3138		if (!range) {
3139			ret = -ENOMEM;
3140			goto out;
3141		}
3142
3143		if (argp) {
3144			if (copy_from_user(range, argp,
3145					   sizeof(*range))) {
3146				ret = -EFAULT;
3147				kfree(range);
3148				goto out;
3149			}
3150			/* compression requires us to start the IO */
3151			if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
3152				range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
3153				range->extent_thresh = (u32)-1;
3154			}
3155		} else {
3156			/* the rest are all set to zero by kzalloc */
3157			range->len = (u64)-1;
3158		}
3159		ret = btrfs_defrag_file(file_inode(file), file,
3160					range, BTRFS_OLDEST_GENERATION, 0);
3161		if (ret > 0)
3162			ret = 0;
3163		kfree(range);
3164		break;
3165	default:
3166		ret = -EINVAL;
3167	}
3168out:
3169	mnt_drop_write_file(file);
3170	return ret;
3171}
3172
3173static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
3174{
3175	struct btrfs_ioctl_vol_args *vol_args;
3176	int ret;
3177
3178	if (!capable(CAP_SYS_ADMIN))
3179		return -EPERM;
3180
3181	if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_ADD))
3182		return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3183
3184	vol_args = memdup_user(arg, sizeof(*vol_args));
3185	if (IS_ERR(vol_args)) {
3186		ret = PTR_ERR(vol_args);
3187		goto out;
3188	}
3189
3190	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3191	ret = btrfs_init_new_device(fs_info, vol_args->name);
3192
3193	if (!ret)
3194		btrfs_info(fs_info, "disk added %s", vol_args->name);
3195
3196	kfree(vol_args);
3197out:
3198	btrfs_exclop_finish(fs_info);
3199	return ret;
3200}
3201
3202static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3203{
3204	struct inode *inode = file_inode(file);
3205	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3206	struct btrfs_ioctl_vol_args_v2 *vol_args;
3207	int ret;
3208	bool cancel = false;
3209
3210	if (!capable(CAP_SYS_ADMIN))
3211		return -EPERM;
3212
3213	ret = mnt_want_write_file(file);
3214	if (ret)
3215		return ret;
3216
3217	vol_args = memdup_user(arg, sizeof(*vol_args));
3218	if (IS_ERR(vol_args)) {
3219		ret = PTR_ERR(vol_args);
3220		goto err_drop;
3221	}
3222
3223	if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) {
3224		ret = -EOPNOTSUPP;
3225		goto out;
3226	}
3227	vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3228	if (!(vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) &&
3229	    strcmp("cancel", vol_args->name) == 0)
3230		cancel = true;
3231
3232	ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
3233					   cancel);
3234	if (ret)
3235		goto out;
3236	/* Exclusive operation is now claimed */
3237
3238	if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3239		ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
3240	else
3241		ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3242
3243	btrfs_exclop_finish(fs_info);
3244
3245	if (!ret) {
3246		if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3247			btrfs_info(fs_info, "device deleted: id %llu",
3248					vol_args->devid);
3249		else
3250			btrfs_info(fs_info, "device deleted: %s",
3251					vol_args->name);
3252	}
3253out:
3254	kfree(vol_args);
3255err_drop:
3256	mnt_drop_write_file(file);
3257	return ret;
3258}
3259
3260static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3261{
3262	struct inode *inode = file_inode(file);
3263	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3264	struct btrfs_ioctl_vol_args *vol_args;
3265	int ret;
3266	bool cancel;
3267
3268	if (!capable(CAP_SYS_ADMIN))
3269		return -EPERM;
3270
3271	ret = mnt_want_write_file(file);
3272	if (ret)
3273		return ret;
3274
3275	vol_args = memdup_user(arg, sizeof(*vol_args));
3276	if (IS_ERR(vol_args)) {
3277		ret = PTR_ERR(vol_args);
3278		goto out_drop_write;
3279	}
3280	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3281	cancel = (strcmp("cancel", vol_args->name) == 0);
3282
3283	ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
3284					   cancel);
3285	if (ret == 0) {
3286		ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3287		if (!ret)
3288			btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3289		btrfs_exclop_finish(fs_info);
3290	}
3291
3292	kfree(vol_args);
3293out_drop_write:
3294	mnt_drop_write_file(file);
3295
3296	return ret;
3297}
3298
3299static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3300				void __user *arg)
3301{
3302	struct btrfs_ioctl_fs_info_args *fi_args;
3303	struct btrfs_device *device;
3304	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3305	u64 flags_in;
3306	int ret = 0;
3307
3308	fi_args = memdup_user(arg, sizeof(*fi_args));
3309	if (IS_ERR(fi_args))
3310		return PTR_ERR(fi_args);
3311
3312	flags_in = fi_args->flags;
3313	memset(fi_args, 0, sizeof(*fi_args));
3314
3315	rcu_read_lock();
3316	fi_args->num_devices = fs_devices->num_devices;
3317
3318	list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3319		if (device->devid > fi_args->max_id)
3320			fi_args->max_id = device->devid;
3321	}
3322	rcu_read_unlock();
3323
3324	memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
3325	fi_args->nodesize = fs_info->nodesize;
3326	fi_args->sectorsize = fs_info->sectorsize;
3327	fi_args->clone_alignment = fs_info->sectorsize;
3328
3329	if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) {
3330		fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy);
3331		fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy);
3332		fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO;
3333	}
3334
3335	if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) {
3336		fi_args->generation = fs_info->generation;
3337		fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION;
3338	}
3339
3340	if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) {
3341		memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid,
3342		       sizeof(fi_args->metadata_uuid));
3343		fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID;
3344	}
3345
3346	if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3347		ret = -EFAULT;
3348
3349	kfree(fi_args);
3350	return ret;
3351}
3352
3353static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3354				 void __user *arg)
3355{
3356	struct btrfs_ioctl_dev_info_args *di_args;
3357	struct btrfs_device *dev;
3358	int ret = 0;
3359	char *s_uuid = NULL;
3360
3361	di_args = memdup_user(arg, sizeof(*di_args));
3362	if (IS_ERR(di_args))
3363		return PTR_ERR(di_args);
3364
3365	if (!btrfs_is_empty_uuid(di_args->uuid))
3366		s_uuid = di_args->uuid;
3367
3368	rcu_read_lock();
3369	dev = btrfs_find_device(fs_info->fs_devices, di_args->devid, s_uuid,
3370				NULL);
3371
3372	if (!dev) {
3373		ret = -ENODEV;
3374		goto out;
3375	}
3376
3377	di_args->devid = dev->devid;
3378	di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3379	di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3380	memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3381	if (dev->name) {
3382		strncpy(di_args->path, rcu_str_deref(dev->name),
3383				sizeof(di_args->path) - 1);
3384		di_args->path[sizeof(di_args->path) - 1] = 0;
3385	} else {
3386		di_args->path[0] = '\0';
3387	}
3388
3389out:
3390	rcu_read_unlock();
3391	if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3392		ret = -EFAULT;
3393
3394	kfree(di_args);
3395	return ret;
3396}
3397
3398static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3399{
3400	struct inode *inode = file_inode(file);
3401	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3402	struct btrfs_root *root = BTRFS_I(inode)->root;
3403	struct btrfs_root *new_root;
3404	struct btrfs_dir_item *di;
3405	struct btrfs_trans_handle *trans;
3406	struct btrfs_path *path = NULL;
3407	struct btrfs_disk_key disk_key;
3408	u64 objectid = 0;
3409	u64 dir_id;
3410	int ret;
3411
3412	if (!capable(CAP_SYS_ADMIN))
3413		return -EPERM;
3414
3415	ret = mnt_want_write_file(file);
3416	if (ret)
3417		return ret;
3418
3419	if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3420		ret = -EFAULT;
3421		goto out;
3422	}
3423
3424	if (!objectid)
3425		objectid = BTRFS_FS_TREE_OBJECTID;
3426
3427	new_root = btrfs_get_fs_root(fs_info, objectid, true);
3428	if (IS_ERR(new_root)) {
3429		ret = PTR_ERR(new_root);
3430		goto out;
3431	}
3432	if (!is_fstree(new_root->root_key.objectid)) {
3433		ret = -ENOENT;
3434		goto out_free;
3435	}
3436
3437	path = btrfs_alloc_path();
3438	if (!path) {
3439		ret = -ENOMEM;
3440		goto out_free;
3441	}
3442
3443	trans = btrfs_start_transaction(root, 1);
3444	if (IS_ERR(trans)) {
3445		ret = PTR_ERR(trans);
3446		goto out_free;
3447	}
3448
3449	dir_id = btrfs_super_root_dir(fs_info->super_copy);
3450	di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
3451				   dir_id, "default", 7, 1);
3452	if (IS_ERR_OR_NULL(di)) {
3453		btrfs_release_path(path);
3454		btrfs_end_transaction(trans);
3455		btrfs_err(fs_info,
3456			  "Umm, you don't have the default diritem, this isn't going to work");
3457		ret = -ENOENT;
3458		goto out_free;
3459	}
3460
3461	btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3462	btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3463	btrfs_mark_buffer_dirty(path->nodes[0]);
3464	btrfs_release_path(path);
3465
3466	btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
3467	btrfs_end_transaction(trans);
3468out_free:
3469	btrfs_put_root(new_root);
3470	btrfs_free_path(path);
3471out:
3472	mnt_drop_write_file(file);
3473	return ret;
3474}
3475
3476static void get_block_group_info(struct list_head *groups_list,
3477				 struct btrfs_ioctl_space_info *space)
3478{
3479	struct btrfs_block_group *block_group;
3480
3481	space->total_bytes = 0;
3482	space->used_bytes = 0;
3483	space->flags = 0;
3484	list_for_each_entry(block_group, groups_list, list) {
3485		space->flags = block_group->flags;
3486		space->total_bytes += block_group->length;
3487		space->used_bytes += block_group->used;
3488	}
3489}
3490
3491static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
3492				   void __user *arg)
3493{
3494	struct btrfs_ioctl_space_args space_args;
3495	struct btrfs_ioctl_space_info space;
3496	struct btrfs_ioctl_space_info *dest;
3497	struct btrfs_ioctl_space_info *dest_orig;
3498	struct btrfs_ioctl_space_info __user *user_dest;
3499	struct btrfs_space_info *info;
3500	static const u64 types[] = {
3501		BTRFS_BLOCK_GROUP_DATA,
3502		BTRFS_BLOCK_GROUP_SYSTEM,
3503		BTRFS_BLOCK_GROUP_METADATA,
3504		BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
3505	};
3506	int num_types = 4;
3507	int alloc_size;
3508	int ret = 0;
3509	u64 slot_count = 0;
3510	int i, c;
3511
3512	if (copy_from_user(&space_args,
3513			   (struct btrfs_ioctl_space_args __user *)arg,
3514			   sizeof(space_args)))
3515		return -EFAULT;
3516
3517	for (i = 0; i < num_types; i++) {
3518		struct btrfs_space_info *tmp;
3519
3520		info = NULL;
3521		list_for_each_entry(tmp, &fs_info->space_info, list) {
3522			if (tmp->flags == types[i]) {
3523				info = tmp;
3524				break;
3525			}
3526		}
3527
3528		if (!info)
3529			continue;
3530
3531		down_read(&info->groups_sem);
3532		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3533			if (!list_empty(&info->block_groups[c]))
3534				slot_count++;
3535		}
3536		up_read(&info->groups_sem);
3537	}
3538
3539	/*
3540	 * Global block reserve, exported as a space_info
3541	 */
3542	slot_count++;
3543
3544	/* space_slots == 0 means they are asking for a count */
3545	if (space_args.space_slots == 0) {
3546		space_args.total_spaces = slot_count;
3547		goto out;
3548	}
3549
3550	slot_count = min_t(u64, space_args.space_slots, slot_count);
3551
3552	alloc_size = sizeof(*dest) * slot_count;
3553
3554	/* we generally have at most 6 or so space infos, one for each raid
3555	 * level.  So, a whole page should be more than enough for everyone
3556	 */
3557	if (alloc_size > PAGE_SIZE)
3558		return -ENOMEM;
3559
3560	space_args.total_spaces = 0;
3561	dest = kmalloc(alloc_size, GFP_KERNEL);
3562	if (!dest)
3563		return -ENOMEM;
3564	dest_orig = dest;
3565
3566	/* now we have a buffer to copy into */
3567	for (i = 0; i < num_types; i++) {
3568		struct btrfs_space_info *tmp;
3569
3570		if (!slot_count)
3571			break;
3572
3573		info = NULL;
3574		list_for_each_entry(tmp, &fs_info->space_info, list) {
3575			if (tmp->flags == types[i]) {
3576				info = tmp;
3577				break;
3578			}
3579		}
3580
3581		if (!info)
3582			continue;
3583		down_read(&info->groups_sem);
3584		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3585			if (!list_empty(&info->block_groups[c])) {
3586				get_block_group_info(&info->block_groups[c],
3587						     &space);
3588				memcpy(dest, &space, sizeof(space));
3589				dest++;
3590				space_args.total_spaces++;
3591				slot_count--;
3592			}
3593			if (!slot_count)
3594				break;
3595		}
3596		up_read(&info->groups_sem);
3597	}
3598
3599	/*
3600	 * Add global block reserve
3601	 */
3602	if (slot_count) {
3603		struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3604
3605		spin_lock(&block_rsv->lock);
3606		space.total_bytes = block_rsv->size;
3607		space.used_bytes = block_rsv->size - block_rsv->reserved;
3608		spin_unlock(&block_rsv->lock);
3609		space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3610		memcpy(dest, &space, sizeof(space));
3611		space_args.total_spaces++;
3612	}
3613
3614	user_dest = (struct btrfs_ioctl_space_info __user *)
3615		(arg + sizeof(struct btrfs_ioctl_space_args));
3616
3617	if (copy_to_user(user_dest, dest_orig, alloc_size))
3618		ret = -EFAULT;
3619
3620	kfree(dest_orig);
3621out:
3622	if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3623		ret = -EFAULT;
3624
3625	return ret;
3626}
3627
3628static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3629					    void __user *argp)
3630{
3631	struct btrfs_trans_handle *trans;
3632	u64 transid;
3633	int ret;
3634
3635	trans = btrfs_attach_transaction_barrier(root);
3636	if (IS_ERR(trans)) {
3637		if (PTR_ERR(trans) != -ENOENT)
3638			return PTR_ERR(trans);
3639
3640		/* No running transaction, don't bother */
3641		transid = root->fs_info->last_trans_committed;
3642		goto out;
3643	}
3644	transid = trans->transid;
3645	ret = btrfs_commit_transaction_async(trans);
3646	if (ret) {
3647		btrfs_end_transaction(trans);
3648		return ret;
3649	}
3650out:
3651	if (argp)
3652		if (copy_to_user(argp, &transid, sizeof(transid)))
3653			return -EFAULT;
3654	return 0;
3655}
3656
3657static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
3658					   void __user *argp)
3659{
3660	u64 transid;
3661
3662	if (argp) {
3663		if (copy_from_user(&transid, argp, sizeof(transid)))
3664			return -EFAULT;
3665	} else {
3666		transid = 0;  /* current trans */
3667	}
3668	return btrfs_wait_for_commit(fs_info, transid);
3669}
3670
3671static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3672{
3673	struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
3674	struct btrfs_ioctl_scrub_args *sa;
3675	int ret;
3676
3677	if (!capable(CAP_SYS_ADMIN))
3678		return -EPERM;
3679
3680	sa = memdup_user(arg, sizeof(*sa));
3681	if (IS_ERR(sa))
3682		return PTR_ERR(sa);
3683
3684	if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3685		ret = mnt_want_write_file(file);
3686		if (ret)
3687			goto out;
3688	}
3689
3690	ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
3691			      &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3692			      0);
3693
3694	/*
3695	 * Copy scrub args to user space even if btrfs_scrub_dev() returned an
3696	 * error. This is important as it allows user space to know how much
3697	 * progress scrub has done. For example, if scrub is canceled we get
3698	 * -ECANCELED from btrfs_scrub_dev() and return that error back to user
3699	 * space. Later user space can inspect the progress from the structure
3700	 * btrfs_ioctl_scrub_args and resume scrub from where it left off
3701	 * previously (btrfs-progs does this).
3702	 * If we fail to copy the btrfs_ioctl_scrub_args structure to user space
3703	 * then return -EFAULT to signal the structure was not copied or it may
3704	 * be corrupt and unreliable due to a partial copy.
3705	 */
3706	if (copy_to_user(arg, sa, sizeof(*sa)))
3707		ret = -EFAULT;
3708
3709	if (!(sa->flags & BTRFS_SCRUB_READONLY))
3710		mnt_drop_write_file(file);
3711out:
3712	kfree(sa);
3713	return ret;
3714}
3715
3716static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
3717{
3718	if (!capable(CAP_SYS_ADMIN))
3719		return -EPERM;
3720
3721	return btrfs_scrub_cancel(fs_info);
3722}
3723
3724static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
3725				       void __user *arg)
3726{
3727	struct btrfs_ioctl_scrub_args *sa;
3728	int ret;
3729
3730	if (!capable(CAP_SYS_ADMIN))
3731		return -EPERM;
3732
3733	sa = memdup_user(arg, sizeof(*sa));
3734	if (IS_ERR(sa))
3735		return PTR_ERR(sa);
3736
3737	ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
3738
3739	if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3740		ret = -EFAULT;
3741
3742	kfree(sa);
3743	return ret;
3744}
3745
3746static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
3747				      void __user *arg)
3748{
3749	struct btrfs_ioctl_get_dev_stats *sa;
3750	int ret;
3751
3752	sa = memdup_user(arg, sizeof(*sa));
3753	if (IS_ERR(sa))
3754		return PTR_ERR(sa);
3755
3756	if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3757		kfree(sa);
3758		return -EPERM;
3759	}
3760
3761	ret = btrfs_get_dev_stats(fs_info, sa);
3762
3763	if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3764		ret = -EFAULT;
3765
3766	kfree(sa);
3767	return ret;
3768}
3769
3770static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
3771				    void __user *arg)
3772{
3773	struct btrfs_ioctl_dev_replace_args *p;
3774	int ret;
3775
3776	if (!capable(CAP_SYS_ADMIN))
3777		return -EPERM;
3778
3779	p = memdup_user(arg, sizeof(*p));
3780	if (IS_ERR(p))
3781		return PTR_ERR(p);
3782
3783	switch (p->cmd) {
3784	case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3785		if (sb_rdonly(fs_info->sb)) {
3786			ret = -EROFS;
3787			goto out;
3788		}
3789		if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
3790			ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3791		} else {
3792			ret = btrfs_dev_replace_by_ioctl(fs_info, p);
3793			btrfs_exclop_finish(fs_info);
3794		}
3795		break;
3796	case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3797		btrfs_dev_replace_status(fs_info, p);
3798		ret = 0;
3799		break;
3800	case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3801		p->result = btrfs_dev_replace_cancel(fs_info);
3802		ret = 0;
3803		break;
3804	default:
3805		ret = -EINVAL;
3806		break;
3807	}
3808
3809	if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
3810		ret = -EFAULT;
3811out:
3812	kfree(p);
3813	return ret;
3814}
3815
3816static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3817{
3818	int ret = 0;
3819	int i;
3820	u64 rel_ptr;
3821	int size;
3822	struct btrfs_ioctl_ino_path_args *ipa = NULL;
3823	struct inode_fs_paths *ipath = NULL;
3824	struct btrfs_path *path;
3825
3826	if (!capable(CAP_DAC_READ_SEARCH))
3827		return -EPERM;
3828
3829	path = btrfs_alloc_path();
3830	if (!path) {
3831		ret = -ENOMEM;
3832		goto out;
3833	}
3834
3835	ipa = memdup_user(arg, sizeof(*ipa));
3836	if (IS_ERR(ipa)) {
3837		ret = PTR_ERR(ipa);
3838		ipa = NULL;
3839		goto out;
3840	}
3841
3842	size = min_t(u32, ipa->size, 4096);
3843	ipath = init_ipath(size, root, path);
3844	if (IS_ERR(ipath)) {
3845		ret = PTR_ERR(ipath);
3846		ipath = NULL;
3847		goto out;
3848	}
3849
3850	ret = paths_from_inode(ipa->inum, ipath);
3851	if (ret < 0)
3852		goto out;
3853
3854	for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3855		rel_ptr = ipath->fspath->val[i] -
3856			  (u64)(unsigned long)ipath->fspath->val;
3857		ipath->fspath->val[i] = rel_ptr;
3858	}
3859
3860	ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
3861			   ipath->fspath, size);
3862	if (ret) {
3863		ret = -EFAULT;
3864		goto out;
3865	}
3866
3867out:
3868	btrfs_free_path(path);
3869	free_ipath(ipath);
3870	kfree(ipa);
3871
3872	return ret;
3873}
3874
3875static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3876{
3877	struct btrfs_data_container *inodes = ctx;
3878	const size_t c = 3 * sizeof(u64);
3879
3880	if (inodes->bytes_left >= c) {
3881		inodes->bytes_left -= c;
3882		inodes->val[inodes->elem_cnt] = inum;
3883		inodes->val[inodes->elem_cnt + 1] = offset;
3884		inodes->val[inodes->elem_cnt + 2] = root;
3885		inodes->elem_cnt += 3;
3886	} else {
3887		inodes->bytes_missing += c - inodes->bytes_left;
3888		inodes->bytes_left = 0;
3889		inodes->elem_missed += 3;
3890	}
3891
3892	return 0;
3893}
3894
3895static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
3896					void __user *arg, int version)
3897{
3898	int ret = 0;
3899	int size;
3900	struct btrfs_ioctl_logical_ino_args *loi;
3901	struct btrfs_data_container *inodes = NULL;
3902	struct btrfs_path *path = NULL;
3903	bool ignore_offset;
3904
3905	if (!capable(CAP_SYS_ADMIN))
3906		return -EPERM;
3907
3908	loi = memdup_user(arg, sizeof(*loi));
3909	if (IS_ERR(loi))
3910		return PTR_ERR(loi);
3911
3912	if (version == 1) {
3913		ignore_offset = false;
3914		size = min_t(u32, loi->size, SZ_64K);
3915	} else {
3916		/* All reserved bits must be 0 for now */
3917		if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
3918			ret = -EINVAL;
3919			goto out_loi;
3920		}
3921		/* Only accept flags we have defined so far */
3922		if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
3923			ret = -EINVAL;
3924			goto out_loi;
3925		}
3926		ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
3927		size = min_t(u32, loi->size, SZ_16M);
3928	}
3929
3930	path = btrfs_alloc_path();
3931	if (!path) {
3932		ret = -ENOMEM;
3933		goto out;
3934	}
3935
3936	inodes = init_data_container(size);
3937	if (IS_ERR(inodes)) {
3938		ret = PTR_ERR(inodes);
3939		inodes = NULL;
3940		goto out;
3941	}
3942
3943	ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
3944					  build_ino_list, inodes, ignore_offset);
3945	if (ret == -EINVAL)
3946		ret = -ENOENT;
3947	if (ret < 0)
3948		goto out;
3949
3950	ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
3951			   size);
3952	if (ret)
3953		ret = -EFAULT;
3954
3955out:
3956	btrfs_free_path(path);
3957	kvfree(inodes);
3958out_loi:
3959	kfree(loi);
3960
3961	return ret;
3962}
3963
3964void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
3965			       struct btrfs_ioctl_balance_args *bargs)
3966{
3967	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3968
3969	bargs->flags = bctl->flags;
3970
3971	if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
3972		bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3973	if (atomic_read(&fs_info->balance_pause_req))
3974		bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3975	if (atomic_read(&fs_info->balance_cancel_req))
3976		bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3977
3978	memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3979	memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3980	memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3981
3982	spin_lock(&fs_info->balance_lock);
3983	memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3984	spin_unlock(&fs_info->balance_lock);
3985}
3986
3987static long btrfs_ioctl_balance(struct file *file, void __user *arg)
3988{
3989	struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3990	struct btrfs_fs_info *fs_info = root->fs_info;
3991	struct btrfs_ioctl_balance_args *bargs;
3992	struct btrfs_balance_control *bctl;
3993	bool need_unlock; /* for mut. excl. ops lock */
3994	int ret;
3995
3996	if (!capable(CAP_SYS_ADMIN))
3997		return -EPERM;
3998
3999	ret = mnt_want_write_file(file);
4000	if (ret)
4001		return ret;
4002
4003again:
4004	if (btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
4005		mutex_lock(&fs_info->balance_mutex);
4006		need_unlock = true;
4007		goto locked;
4008	}
4009
4010	/*
4011	 * mut. excl. ops lock is locked.  Three possibilities:
4012	 *   (1) some other op is running
4013	 *   (2) balance is running
4014	 *   (3) balance is paused -- special case (think resume)
4015	 */
4016	mutex_lock(&fs_info->balance_mutex);
4017	if (fs_info->balance_ctl) {
4018		/* this is either (2) or (3) */
4019		if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4020			mutex_unlock(&fs_info->balance_mutex);
4021			/*
4022			 * Lock released to allow other waiters to continue,
4023			 * we'll reexamine the status again.
4024			 */
4025			mutex_lock(&fs_info->balance_mutex);
4026
4027			if (fs_info->balance_ctl &&
4028			    !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4029				/* this is (3) */
4030				need_unlock = false;
4031				goto locked;
4032			}
4033
4034			mutex_unlock(&fs_info->balance_mutex);
4035			goto again;
4036		} else {
4037			/* this is (2) */
4038			mutex_unlock(&fs_info->balance_mutex);
4039			ret = -EINPROGRESS;
4040			goto out;
4041		}
4042	} else {
4043		/* this is (1) */
4044		mutex_unlock(&fs_info->balance_mutex);
4045		ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4046		goto out;
4047	}
4048
4049locked:
4050
4051	if (arg) {
4052		bargs = memdup_user(arg, sizeof(*bargs));
4053		if (IS_ERR(bargs)) {
4054			ret = PTR_ERR(bargs);
4055			goto out_unlock;
4056		}
4057
4058		if (bargs->flags & BTRFS_BALANCE_RESUME) {
4059			if (!fs_info->balance_ctl) {
4060				ret = -ENOTCONN;
4061				goto out_bargs;
4062			}
4063
4064			bctl = fs_info->balance_ctl;
4065			spin_lock(&fs_info->balance_lock);
4066			bctl->flags |= BTRFS_BALANCE_RESUME;
4067			spin_unlock(&fs_info->balance_lock);
4068
4069			goto do_balance;
4070		}
4071	} else {
4072		bargs = NULL;
4073	}
4074
4075	if (fs_info->balance_ctl) {
4076		ret = -EINPROGRESS;
4077		goto out_bargs;
4078	}
4079
4080	bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4081	if (!bctl) {
4082		ret = -ENOMEM;
4083		goto out_bargs;
4084	}
4085
4086	if (arg) {
4087		memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4088		memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4089		memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4090
4091		bctl->flags = bargs->flags;
4092	} else {
4093		/* balance everything - no filters */
4094		bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4095	}
4096
4097	if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4098		ret = -EINVAL;
4099		goto out_bctl;
4100	}
4101
4102do_balance:
4103	/*
4104	 * Ownership of bctl and exclusive operation goes to btrfs_balance.
4105	 * bctl is freed in reset_balance_state, or, if restriper was paused
4106	 * all the way until unmount, in free_fs_info.  The flag should be
4107	 * cleared after reset_balance_state.
4108	 */
4109	need_unlock = false;
4110
4111	ret = btrfs_balance(fs_info, bctl, bargs);
4112	bctl = NULL;
4113
4114	if ((ret == 0 || ret == -ECANCELED) && arg) {
4115		if (copy_to_user(arg, bargs, sizeof(*bargs)))
4116			ret = -EFAULT;
4117	}
4118
4119out_bctl:
4120	kfree(bctl);
4121out_bargs:
4122	kfree(bargs);
4123out_unlock:
4124	mutex_unlock(&fs_info->balance_mutex);
4125	if (need_unlock)
4126		btrfs_exclop_finish(fs_info);
4127out:
4128	mnt_drop_write_file(file);
4129	return ret;
4130}
4131
4132static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4133{
4134	if (!capable(CAP_SYS_ADMIN))
4135		return -EPERM;
4136
4137	switch (cmd) {
4138	case BTRFS_BALANCE_CTL_PAUSE:
4139		return btrfs_pause_balance(fs_info);
4140	case BTRFS_BALANCE_CTL_CANCEL:
4141		return btrfs_cancel_balance(fs_info);
4142	}
4143
4144	return -EINVAL;
4145}
4146
4147static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4148					 void __user *arg)
4149{
4150	struct btrfs_ioctl_balance_args *bargs;
4151	int ret = 0;
4152
4153	if (!capable(CAP_SYS_ADMIN))
4154		return -EPERM;
4155
4156	mutex_lock(&fs_info->balance_mutex);
4157	if (!fs_info->balance_ctl) {
4158		ret = -ENOTCONN;
4159		goto out;
4160	}
4161
4162	bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4163	if (!bargs) {
4164		ret = -ENOMEM;
4165		goto out;
4166	}
4167
4168	btrfs_update_ioctl_balance_args(fs_info, bargs);
4169
4170	if (copy_to_user(arg, bargs, sizeof(*bargs)))
4171		ret = -EFAULT;
4172
4173	kfree(bargs);
4174out:
4175	mutex_unlock(&fs_info->balance_mutex);
4176	return ret;
4177}
4178
4179static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4180{
4181	struct inode *inode = file_inode(file);
4182	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4183	struct btrfs_ioctl_quota_ctl_args *sa;
4184	int ret;
4185
4186	if (!capable(CAP_SYS_ADMIN))
4187		return -EPERM;
4188
4189	ret = mnt_want_write_file(file);
4190	if (ret)
4191		return ret;
4192
4193	sa = memdup_user(arg, sizeof(*sa));
4194	if (IS_ERR(sa)) {
4195		ret = PTR_ERR(sa);
4196		goto drop_write;
4197	}
4198
4199	down_write(&fs_info->subvol_sem);
4200
4201	switch (sa->cmd) {
4202	case BTRFS_QUOTA_CTL_ENABLE:
4203		ret = btrfs_quota_enable(fs_info);
4204		break;
4205	case BTRFS_QUOTA_CTL_DISABLE:
4206		ret = btrfs_quota_disable(fs_info);
4207		break;
4208	default:
4209		ret = -EINVAL;
4210		break;
4211	}
4212
4213	kfree(sa);
4214	up_write(&fs_info->subvol_sem);
4215drop_write:
4216	mnt_drop_write_file(file);
4217	return ret;
4218}
4219
4220static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4221{
4222	struct inode *inode = file_inode(file);
4223	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4224	struct btrfs_root *root = BTRFS_I(inode)->root;
4225	struct btrfs_ioctl_qgroup_assign_args *sa;
4226	struct btrfs_trans_handle *trans;
4227	int ret;
4228	int err;
4229
4230	if (!capable(CAP_SYS_ADMIN))
4231		return -EPERM;
4232
4233	ret = mnt_want_write_file(file);
4234	if (ret)
4235		return ret;
4236
4237	sa = memdup_user(arg, sizeof(*sa));
4238	if (IS_ERR(sa)) {
4239		ret = PTR_ERR(sa);
4240		goto drop_write;
4241	}
4242
4243	trans = btrfs_join_transaction(root);
4244	if (IS_ERR(trans)) {
4245		ret = PTR_ERR(trans);
4246		goto out;
4247	}
4248
4249	if (sa->assign) {
4250		ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
4251	} else {
4252		ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
4253	}
4254
4255	/* update qgroup status and info */
4256	err = btrfs_run_qgroups(trans);
4257	if (err < 0)
4258		btrfs_handle_fs_error(fs_info, err,
4259				      "failed to update qgroup status and info");
4260	err = btrfs_end_transaction(trans);
4261	if (err && !ret)
4262		ret = err;
4263
4264out:
4265	kfree(sa);
4266drop_write:
4267	mnt_drop_write_file(file);
4268	return ret;
4269}
4270
4271static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4272{
4273	struct inode *inode = file_inode(file);
4274	struct btrfs_root *root = BTRFS_I(inode)->root;
4275	struct btrfs_ioctl_qgroup_create_args *sa;
4276	struct btrfs_trans_handle *trans;
4277	int ret;
4278	int err;
4279
4280	if (!capable(CAP_SYS_ADMIN))
4281		return -EPERM;
4282
4283	ret = mnt_want_write_file(file);
4284	if (ret)
4285		return ret;
4286
4287	sa = memdup_user(arg, sizeof(*sa));
4288	if (IS_ERR(sa)) {
4289		ret = PTR_ERR(sa);
4290		goto drop_write;
4291	}
4292
4293	if (!sa->qgroupid) {
4294		ret = -EINVAL;
4295		goto out;
4296	}
4297
4298	trans = btrfs_join_transaction(root);
4299	if (IS_ERR(trans)) {
4300		ret = PTR_ERR(trans);
4301		goto out;
4302	}
4303
4304	if (sa->create) {
4305		ret = btrfs_create_qgroup(trans, sa->qgroupid);
4306	} else {
4307		ret = btrfs_remove_qgroup(trans, sa->qgroupid);
4308	}
4309
4310	err = btrfs_end_transaction(trans);
4311	if (err && !ret)
4312		ret = err;
4313
4314out:
4315	kfree(sa);
4316drop_write:
4317	mnt_drop_write_file(file);
4318	return ret;
4319}
4320
4321static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4322{
4323	struct inode *inode = file_inode(file);
4324	struct btrfs_root *root = BTRFS_I(inode)->root;
4325	struct btrfs_ioctl_qgroup_limit_args *sa;
4326	struct btrfs_trans_handle *trans;
4327	int ret;
4328	int err;
4329	u64 qgroupid;
4330
4331	if (!capable(CAP_SYS_ADMIN))
4332		return -EPERM;
4333
4334	ret = mnt_want_write_file(file);
4335	if (ret)
4336		return ret;
4337
4338	sa = memdup_user(arg, sizeof(*sa));
4339	if (IS_ERR(sa)) {
4340		ret = PTR_ERR(sa);
4341		goto drop_write;
4342	}
4343
4344	trans = btrfs_join_transaction(root);
4345	if (IS_ERR(trans)) {
4346		ret = PTR_ERR(trans);
4347		goto out;
4348	}
4349
4350	qgroupid = sa->qgroupid;
4351	if (!qgroupid) {
4352		/* take the current subvol as qgroup */
4353		qgroupid = root->root_key.objectid;
4354	}
4355
4356	ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
4357
4358	err = btrfs_end_transaction(trans);
4359	if (err && !ret)
4360		ret = err;
4361
4362out:
4363	kfree(sa);
4364drop_write:
4365	mnt_drop_write_file(file);
4366	return ret;
4367}
4368
4369static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4370{
4371	struct inode *inode = file_inode(file);
4372	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4373	struct btrfs_ioctl_quota_rescan_args *qsa;
4374	int ret;
4375
4376	if (!capable(CAP_SYS_ADMIN))
4377		return -EPERM;
4378
4379	ret = mnt_want_write_file(file);
4380	if (ret)
4381		return ret;
4382
4383	qsa = memdup_user(arg, sizeof(*qsa));
4384	if (IS_ERR(qsa)) {
4385		ret = PTR_ERR(qsa);
4386		goto drop_write;
4387	}
4388
4389	if (qsa->flags) {
4390		ret = -EINVAL;
4391		goto out;
4392	}
4393
4394	ret = btrfs_qgroup_rescan(fs_info);
4395
4396out:
4397	kfree(qsa);
4398drop_write:
4399	mnt_drop_write_file(file);
4400	return ret;
4401}
4402
4403static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info,
4404						void __user *arg)
4405{
4406	struct btrfs_ioctl_quota_rescan_args qsa = {0};
4407	int ret = 0;
4408
4409	if (!capable(CAP_SYS_ADMIN))
4410		return -EPERM;
4411
4412	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4413		qsa.flags = 1;
4414		qsa.progress = fs_info->qgroup_rescan_progress.objectid;
4415	}
4416
4417	if (copy_to_user(arg, &qsa, sizeof(qsa)))
4418		ret = -EFAULT;
4419
4420	return ret;
4421}
4422
4423static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info,
4424						void __user *arg)
4425{
4426	if (!capable(CAP_SYS_ADMIN))
4427		return -EPERM;
4428
4429	return btrfs_qgroup_wait_for_completion(fs_info, true);
4430}
4431
4432static long _btrfs_ioctl_set_received_subvol(struct file *file,
4433					    struct btrfs_ioctl_received_subvol_args *sa)
4434{
4435	struct inode *inode = file_inode(file);
4436	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4437	struct btrfs_root *root = BTRFS_I(inode)->root;
4438	struct btrfs_root_item *root_item = &root->root_item;
4439	struct btrfs_trans_handle *trans;
4440	struct timespec64 ct = current_time(inode);
4441	int ret = 0;
4442	int received_uuid_changed;
4443
4444	if (!inode_owner_or_capable(&init_user_ns, inode))
4445		return -EPERM;
4446
4447	ret = mnt_want_write_file(file);
4448	if (ret < 0)
4449		return ret;
4450
4451	down_write(&fs_info->subvol_sem);
4452
4453	if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
4454		ret = -EINVAL;
4455		goto out;
4456	}
4457
4458	if (btrfs_root_readonly(root)) {
4459		ret = -EROFS;
4460		goto out;
4461	}
4462
4463	/*
4464	 * 1 - root item
4465	 * 2 - uuid items (received uuid + subvol uuid)
4466	 */
4467	trans = btrfs_start_transaction(root, 3);
4468	if (IS_ERR(trans)) {
4469		ret = PTR_ERR(trans);
4470		trans = NULL;
4471		goto out;
4472	}
4473
4474	sa->rtransid = trans->transid;
4475	sa->rtime.sec = ct.tv_sec;
4476	sa->rtime.nsec = ct.tv_nsec;
4477
4478	received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4479				       BTRFS_UUID_SIZE);
4480	if (received_uuid_changed &&
4481	    !btrfs_is_empty_uuid(root_item->received_uuid)) {
4482		ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
4483					  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4484					  root->root_key.objectid);
4485		if (ret && ret != -ENOENT) {
4486		        btrfs_abort_transaction(trans, ret);
4487		        btrfs_end_transaction(trans);
4488		        goto out;
4489		}
4490	}
4491	memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4492	btrfs_set_root_stransid(root_item, sa->stransid);
4493	btrfs_set_root_rtransid(root_item, sa->rtransid);
4494	btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4495	btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4496	btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4497	btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4498
4499	ret = btrfs_update_root(trans, fs_info->tree_root,
4500				&root->root_key, &root->root_item);
4501	if (ret < 0) {
4502		btrfs_end_transaction(trans);
4503		goto out;
4504	}
4505	if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4506		ret = btrfs_uuid_tree_add(trans, sa->uuid,
4507					  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4508					  root->root_key.objectid);
4509		if (ret < 0 && ret != -EEXIST) {
4510			btrfs_abort_transaction(trans, ret);
4511			btrfs_end_transaction(trans);
4512			goto out;
4513		}
4514	}
4515	ret = btrfs_commit_transaction(trans);
4516out:
4517	up_write(&fs_info->subvol_sem);
4518	mnt_drop_write_file(file);
4519	return ret;
4520}
4521
4522#ifdef CONFIG_64BIT
4523static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4524						void __user *arg)
4525{
4526	struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4527	struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4528	int ret = 0;
4529
4530	args32 = memdup_user(arg, sizeof(*args32));
4531	if (IS_ERR(args32))
4532		return PTR_ERR(args32);
4533
4534	args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
4535	if (!args64) {
4536		ret = -ENOMEM;
4537		goto out;
4538	}
4539
4540	memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4541	args64->stransid = args32->stransid;
4542	args64->rtransid = args32->rtransid;
4543	args64->stime.sec = args32->stime.sec;
4544	args64->stime.nsec = args32->stime.nsec;
4545	args64->rtime.sec = args32->rtime.sec;
4546	args64->rtime.nsec = args32->rtime.nsec;
4547	args64->flags = args32->flags;
4548
4549	ret = _btrfs_ioctl_set_received_subvol(file, args64);
4550	if (ret)
4551		goto out;
4552
4553	memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4554	args32->stransid = args64->stransid;
4555	args32->rtransid = args64->rtransid;
4556	args32->stime.sec = args64->stime.sec;
4557	args32->stime.nsec = args64->stime.nsec;
4558	args32->rtime.sec = args64->rtime.sec;
4559	args32->rtime.nsec = args64->rtime.nsec;
4560	args32->flags = args64->flags;
4561
4562	ret = copy_to_user(arg, args32, sizeof(*args32));
4563	if (ret)
4564		ret = -EFAULT;
4565
4566out:
4567	kfree(args32);
4568	kfree(args64);
4569	return ret;
4570}
4571#endif
4572
4573static long btrfs_ioctl_set_received_subvol(struct file *file,
4574					    void __user *arg)
4575{
4576	struct btrfs_ioctl_received_subvol_args *sa = NULL;
4577	int ret = 0;
4578
4579	sa = memdup_user(arg, sizeof(*sa));
4580	if (IS_ERR(sa))
4581		return PTR_ERR(sa);
4582
4583	ret = _btrfs_ioctl_set_received_subvol(file, sa);
4584
4585	if (ret)
4586		goto out;
4587
4588	ret = copy_to_user(arg, sa, sizeof(*sa));
4589	if (ret)
4590		ret = -EFAULT;
4591
4592out:
4593	kfree(sa);
4594	return ret;
4595}
4596
4597static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info,
4598					void __user *arg)
4599{
4600	size_t len;
4601	int ret;
4602	char label[BTRFS_LABEL_SIZE];
4603
4604	spin_lock(&fs_info->super_lock);
4605	memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4606	spin_unlock(&fs_info->super_lock);
4607
4608	len = strnlen(label, BTRFS_LABEL_SIZE);
4609
4610	if (len == BTRFS_LABEL_SIZE) {
4611		btrfs_warn(fs_info,
4612			   "label is too long, return the first %zu bytes",
4613			   --len);
4614	}
4615
4616	ret = copy_to_user(arg, label, len);
4617
4618	return ret ? -EFAULT : 0;
4619}
4620
4621static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4622{
4623	struct inode *inode = file_inode(file);
4624	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4625	struct btrfs_root *root = BTRFS_I(inode)->root;
4626	struct btrfs_super_block *super_block = fs_info->super_copy;
4627	struct btrfs_trans_handle *trans;
4628	char label[BTRFS_LABEL_SIZE];
4629	int ret;
4630
4631	if (!capable(CAP_SYS_ADMIN))
4632		return -EPERM;
4633
4634	if (copy_from_user(label, arg, sizeof(label)))
4635		return -EFAULT;
4636
4637	if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4638		btrfs_err(fs_info,
4639			  "unable to set label with more than %d bytes",
4640			  BTRFS_LABEL_SIZE - 1);
4641		return -EINVAL;
4642	}
4643
4644	ret = mnt_want_write_file(file);
4645	if (ret)
4646		return ret;
4647
4648	trans = btrfs_start_transaction(root, 0);
4649	if (IS_ERR(trans)) {
4650		ret = PTR_ERR(trans);
4651		goto out_unlock;
4652	}
4653
4654	spin_lock(&fs_info->super_lock);
4655	strcpy(super_block->label, label);
4656	spin_unlock(&fs_info->super_lock);
4657	ret = btrfs_commit_transaction(trans);
4658
4659out_unlock:
4660	mnt_drop_write_file(file);
4661	return ret;
4662}
4663
4664#define INIT_FEATURE_FLAGS(suffix) \
4665	{ .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
4666	  .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
4667	  .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
4668
4669int btrfs_ioctl_get_supported_features(void __user *arg)
4670{
4671	static const struct btrfs_ioctl_feature_flags features[3] = {
4672		INIT_FEATURE_FLAGS(SUPP),
4673		INIT_FEATURE_FLAGS(SAFE_SET),
4674		INIT_FEATURE_FLAGS(SAFE_CLEAR)
4675	};
4676
4677	if (copy_to_user(arg, &features, sizeof(features)))
4678		return -EFAULT;
4679
4680	return 0;
4681}
4682
4683static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info,
4684					void __user *arg)
4685{
4686	struct btrfs_super_block *super_block = fs_info->super_copy;
4687	struct btrfs_ioctl_feature_flags features;
4688
4689	features.compat_flags = btrfs_super_compat_flags(super_block);
4690	features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
4691	features.incompat_flags = btrfs_super_incompat_flags(super_block);
4692
4693	if (copy_to_user(arg, &features, sizeof(features)))
4694		return -EFAULT;
4695
4696	return 0;
4697}
4698
4699static int check_feature_bits(struct btrfs_fs_info *fs_info,
4700			      enum btrfs_feature_set set,
4701			      u64 change_mask, u64 flags, u64 supported_flags,
4702			      u64 safe_set, u64 safe_clear)
4703{
4704	const char *type = btrfs_feature_set_name(set);
4705	char *names;
4706	u64 disallowed, unsupported;
4707	u64 set_mask = flags & change_mask;
4708	u64 clear_mask = ~flags & change_mask;
4709
4710	unsupported = set_mask & ~supported_flags;
4711	if (unsupported) {
4712		names = btrfs_printable_features(set, unsupported);
4713		if (names) {
4714			btrfs_warn(fs_info,
4715				   "this kernel does not support the %s feature bit%s",
4716				   names, strchr(names, ',') ? "s" : "");
4717			kfree(names);
4718		} else
4719			btrfs_warn(fs_info,
4720				   "this kernel does not support %s bits 0x%llx",
4721				   type, unsupported);
4722		return -EOPNOTSUPP;
4723	}
4724
4725	disallowed = set_mask & ~safe_set;
4726	if (disallowed) {
4727		names = btrfs_printable_features(set, disallowed);
4728		if (names) {
4729			btrfs_warn(fs_info,
4730				   "can't set the %s feature bit%s while mounted",
4731				   names, strchr(names, ',') ? "s" : "");
4732			kfree(names);
4733		} else
4734			btrfs_warn(fs_info,
4735				   "can't set %s bits 0x%llx while mounted",
4736				   type, disallowed);
4737		return -EPERM;
4738	}
4739
4740	disallowed = clear_mask & ~safe_clear;
4741	if (disallowed) {
4742		names = btrfs_printable_features(set, disallowed);
4743		if (names) {
4744			btrfs_warn(fs_info,
4745				   "can't clear the %s feature bit%s while mounted",
4746				   names, strchr(names, ',') ? "s" : "");
4747			kfree(names);
4748		} else
4749			btrfs_warn(fs_info,
4750				   "can't clear %s bits 0x%llx while mounted",
4751				   type, disallowed);
4752		return -EPERM;
4753	}
4754
4755	return 0;
4756}
4757
4758#define check_feature(fs_info, change_mask, flags, mask_base)	\
4759check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,	\
4760		   BTRFS_FEATURE_ ## mask_base ## _SUPP,	\
4761		   BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,	\
4762		   BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
4763
4764static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
4765{
4766	struct inode *inode = file_inode(file);
4767	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4768	struct btrfs_root *root = BTRFS_I(inode)->root;
4769	struct btrfs_super_block *super_block = fs_info->super_copy;
4770	struct btrfs_ioctl_feature_flags flags[2];
4771	struct btrfs_trans_handle *trans;
4772	u64 newflags;
4773	int ret;
4774
4775	if (!capable(CAP_SYS_ADMIN))
4776		return -EPERM;
4777
4778	if (copy_from_user(flags, arg, sizeof(flags)))
4779		return -EFAULT;
4780
4781	/* Nothing to do */
4782	if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
4783	    !flags[0].incompat_flags)
4784		return 0;
4785
4786	ret = check_feature(fs_info, flags[0].compat_flags,
4787			    flags[1].compat_flags, COMPAT);
4788	if (ret)
4789		return ret;
4790
4791	ret = check_feature(fs_info, flags[0].compat_ro_flags,
4792			    flags[1].compat_ro_flags, COMPAT_RO);
4793	if (ret)
4794		return ret;
4795
4796	ret = check_feature(fs_info, flags[0].incompat_flags,
4797			    flags[1].incompat_flags, INCOMPAT);
4798	if (ret)
4799		return ret;
4800
4801	ret = mnt_want_write_file(file);
4802	if (ret)
4803		return ret;
4804
4805	trans = btrfs_start_transaction(root, 0);
4806	if (IS_ERR(trans)) {
4807		ret = PTR_ERR(trans);
4808		goto out_drop_write;
4809	}
4810
4811	spin_lock(&fs_info->super_lock);
4812	newflags = btrfs_super_compat_flags(super_block);
4813	newflags |= flags[0].compat_flags & flags[1].compat_flags;
4814	newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
4815	btrfs_set_super_compat_flags(super_block, newflags);
4816
4817	newflags = btrfs_super_compat_ro_flags(super_block);
4818	newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
4819	newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
4820	btrfs_set_super_compat_ro_flags(super_block, newflags);
4821
4822	newflags = btrfs_super_incompat_flags(super_block);
4823	newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
4824	newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
4825	btrfs_set_super_incompat_flags(super_block, newflags);
4826	spin_unlock(&fs_info->super_lock);
4827
4828	ret = btrfs_commit_transaction(trans);
4829out_drop_write:
4830	mnt_drop_write_file(file);
4831
4832	return ret;
4833}
4834
4835static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
4836{
4837	struct btrfs_ioctl_send_args *arg;
4838	int ret;
4839
4840	if (compat) {
4841#if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4842		struct btrfs_ioctl_send_args_32 args32;
4843
4844		ret = copy_from_user(&args32, argp, sizeof(args32));
4845		if (ret)
4846			return -EFAULT;
4847		arg = kzalloc(sizeof(*arg), GFP_KERNEL);
4848		if (!arg)
4849			return -ENOMEM;
4850		arg->send_fd = args32.send_fd;
4851		arg->clone_sources_count = args32.clone_sources_count;
4852		arg->clone_sources = compat_ptr(args32.clone_sources);
4853		arg->parent_root = args32.parent_root;
4854		arg->flags = args32.flags;
4855		memcpy(arg->reserved, args32.reserved,
4856		       sizeof(args32.reserved));
4857#else
4858		return -ENOTTY;
4859#endif
4860	} else {
4861		arg = memdup_user(argp, sizeof(*arg));
4862		if (IS_ERR(arg))
4863			return PTR_ERR(arg);
4864	}
4865	ret = btrfs_ioctl_send(file, arg);
4866	kfree(arg);
4867	return ret;
4868}
4869
4870long btrfs_ioctl(struct file *file, unsigned int
4871		cmd, unsigned long arg)
4872{
4873	struct inode *inode = file_inode(file);
4874	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4875	struct btrfs_root *root = BTRFS_I(inode)->root;
4876	void __user *argp = (void __user *)arg;
4877
4878	switch (cmd) {
4879	case FS_IOC_GETVERSION:
4880		return btrfs_ioctl_getversion(file, argp);
4881	case FS_IOC_GETFSLABEL:
4882		return btrfs_ioctl_get_fslabel(fs_info, argp);
4883	case FS_IOC_SETFSLABEL:
4884		return btrfs_ioctl_set_fslabel(file, argp);
4885	case FITRIM:
4886		return btrfs_ioctl_fitrim(fs_info, argp);
4887	case BTRFS_IOC_SNAP_CREATE:
4888		return btrfs_ioctl_snap_create(file, argp, 0);
4889	case BTRFS_IOC_SNAP_CREATE_V2:
4890		return btrfs_ioctl_snap_create_v2(file, argp, 0);
4891	case BTRFS_IOC_SUBVOL_CREATE:
4892		return btrfs_ioctl_snap_create(file, argp, 1);
4893	case BTRFS_IOC_SUBVOL_CREATE_V2:
4894		return btrfs_ioctl_snap_create_v2(file, argp, 1);
4895	case BTRFS_IOC_SNAP_DESTROY:
4896		return btrfs_ioctl_snap_destroy(file, argp, false);
4897	case BTRFS_IOC_SNAP_DESTROY_V2:
4898		return btrfs_ioctl_snap_destroy(file, argp, true);
4899	case BTRFS_IOC_SUBVOL_GETFLAGS:
4900		return btrfs_ioctl_subvol_getflags(file, argp);
4901	case BTRFS_IOC_SUBVOL_SETFLAGS:
4902		return btrfs_ioctl_subvol_setflags(file, argp);
4903	case BTRFS_IOC_DEFAULT_SUBVOL:
4904		return btrfs_ioctl_default_subvol(file, argp);
4905	case BTRFS_IOC_DEFRAG:
4906		return btrfs_ioctl_defrag(file, NULL);
4907	case BTRFS_IOC_DEFRAG_RANGE:
4908		return btrfs_ioctl_defrag(file, argp);
4909	case BTRFS_IOC_RESIZE:
4910		return btrfs_ioctl_resize(file, argp);
4911	case BTRFS_IOC_ADD_DEV:
4912		return btrfs_ioctl_add_dev(fs_info, argp);
4913	case BTRFS_IOC_RM_DEV:
4914		return btrfs_ioctl_rm_dev(file, argp);
4915	case BTRFS_IOC_RM_DEV_V2:
4916		return btrfs_ioctl_rm_dev_v2(file, argp);
4917	case BTRFS_IOC_FS_INFO:
4918		return btrfs_ioctl_fs_info(fs_info, argp);
4919	case BTRFS_IOC_DEV_INFO:
4920		return btrfs_ioctl_dev_info(fs_info, argp);
4921	case BTRFS_IOC_BALANCE:
4922		return btrfs_ioctl_balance(file, NULL);
4923	case BTRFS_IOC_TREE_SEARCH:
4924		return btrfs_ioctl_tree_search(file, argp);
4925	case BTRFS_IOC_TREE_SEARCH_V2:
4926		return btrfs_ioctl_tree_search_v2(file, argp);
4927	case BTRFS_IOC_INO_LOOKUP:
4928		return btrfs_ioctl_ino_lookup(file, argp);
4929	case BTRFS_IOC_INO_PATHS:
4930		return btrfs_ioctl_ino_to_path(root, argp);
4931	case BTRFS_IOC_LOGICAL_INO:
4932		return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
4933	case BTRFS_IOC_LOGICAL_INO_V2:
4934		return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
4935	case BTRFS_IOC_SPACE_INFO:
4936		return btrfs_ioctl_space_info(fs_info, argp);
4937	case BTRFS_IOC_SYNC: {
4938		int ret;
4939
4940		ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false);
4941		if (ret)
4942			return ret;
4943		ret = btrfs_sync_fs(inode->i_sb, 1);
4944		/*
4945		 * The transaction thread may want to do more work,
4946		 * namely it pokes the cleaner kthread that will start
4947		 * processing uncleaned subvols.
4948		 */
4949		wake_up_process(fs_info->transaction_kthread);
4950		return ret;
4951	}
4952	case BTRFS_IOC_START_SYNC:
4953		return btrfs_ioctl_start_sync(root, argp);
4954	case BTRFS_IOC_WAIT_SYNC:
4955		return btrfs_ioctl_wait_sync(fs_info, argp);
4956	case BTRFS_IOC_SCRUB:
4957		return btrfs_ioctl_scrub(file, argp);
4958	case BTRFS_IOC_SCRUB_CANCEL:
4959		return btrfs_ioctl_scrub_cancel(fs_info);
4960	case BTRFS_IOC_SCRUB_PROGRESS:
4961		return btrfs_ioctl_scrub_progress(fs_info, argp);
4962	case BTRFS_IOC_BALANCE_V2:
4963		return btrfs_ioctl_balance(file, argp);
4964	case BTRFS_IOC_BALANCE_CTL:
4965		return btrfs_ioctl_balance_ctl(fs_info, arg);
4966	case BTRFS_IOC_BALANCE_PROGRESS:
4967		return btrfs_ioctl_balance_progress(fs_info, argp);
4968	case BTRFS_IOC_SET_RECEIVED_SUBVOL:
4969		return btrfs_ioctl_set_received_subvol(file, argp);
4970#ifdef CONFIG_64BIT
4971	case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
4972		return btrfs_ioctl_set_received_subvol_32(file, argp);
4973#endif
4974	case BTRFS_IOC_SEND:
4975		return _btrfs_ioctl_send(file, argp, false);
4976#if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4977	case BTRFS_IOC_SEND_32:
4978		return _btrfs_ioctl_send(file, argp, true);
4979#endif
4980	case BTRFS_IOC_GET_DEV_STATS:
4981		return btrfs_ioctl_get_dev_stats(fs_info, argp);
4982	case BTRFS_IOC_QUOTA_CTL:
4983		return btrfs_ioctl_quota_ctl(file, argp);
4984	case BTRFS_IOC_QGROUP_ASSIGN:
4985		return btrfs_ioctl_qgroup_assign(file, argp);
4986	case BTRFS_IOC_QGROUP_CREATE:
4987		return btrfs_ioctl_qgroup_create(file, argp);
4988	case BTRFS_IOC_QGROUP_LIMIT:
4989		return btrfs_ioctl_qgroup_limit(file, argp);
4990	case BTRFS_IOC_QUOTA_RESCAN:
4991		return btrfs_ioctl_quota_rescan(file, argp);
4992	case BTRFS_IOC_QUOTA_RESCAN_STATUS:
4993		return btrfs_ioctl_quota_rescan_status(fs_info, argp);
4994	case BTRFS_IOC_QUOTA_RESCAN_WAIT:
4995		return btrfs_ioctl_quota_rescan_wait(fs_info, argp);
4996	case BTRFS_IOC_DEV_REPLACE:
4997		return btrfs_ioctl_dev_replace(fs_info, argp);
4998	case BTRFS_IOC_GET_SUPPORTED_FEATURES:
4999		return btrfs_ioctl_get_supported_features(argp);
5000	case BTRFS_IOC_GET_FEATURES:
5001		return btrfs_ioctl_get_features(fs_info, argp);
5002	case BTRFS_IOC_SET_FEATURES:
5003		return btrfs_ioctl_set_features(file, argp);
5004	case BTRFS_IOC_GET_SUBVOL_INFO:
5005		return btrfs_ioctl_get_subvol_info(file, argp);
5006	case BTRFS_IOC_GET_SUBVOL_ROOTREF:
5007		return btrfs_ioctl_get_subvol_rootref(file, argp);
5008	case BTRFS_IOC_INO_LOOKUP_USER:
5009		return btrfs_ioctl_ino_lookup_user(file, argp);
5010	case FS_IOC_ENABLE_VERITY:
5011		return fsverity_ioctl_enable(file, (const void __user *)argp);
5012	case FS_IOC_MEASURE_VERITY:
5013		return fsverity_ioctl_measure(file, argp);
5014	}
5015
5016	return -ENOTTY;
5017}
5018
5019#ifdef CONFIG_COMPAT
5020long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5021{
5022	/*
5023	 * These all access 32-bit values anyway so no further
5024	 * handling is necessary.
5025	 */
5026	switch (cmd) {
5027	case FS_IOC32_GETVERSION:
5028		cmd = FS_IOC_GETVERSION;
5029		break;
5030	}
5031
5032	return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5033}
5034#endif
5035