2 * Copyright (c) 2012 Taobao.
3 * Written by Tao Ma <boyu.mt@taobao.com>
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of version 2.1 of the GNU Lesser General Public License
7 * as published by the Free Software Foundation.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
15 #include <linux/fiemap.h>
17 #include "ext4_jbd2.h"
22 #define EXT4_XATTR_SYSTEM_DATA "data"
23 #define EXT4_MIN_INLINE_DATA_SIZE ((sizeof(__le32) * EXT4_N_BLOCKS))
24 #define EXT4_INLINE_DOTDOT_OFFSET 2
25 #define EXT4_INLINE_DOTDOT_SIZE 4
27 static int ext4_get_inline_size(struct inode
*inode
)
29 if (EXT4_I(inode
)->i_inline_off
)
30 return EXT4_I(inode
)->i_inline_size
;
35 static int get_max_inline_xattr_value_size(struct inode
*inode
,
36 struct ext4_iloc
*iloc
)
38 struct ext4_xattr_ibody_header
*header
;
39 struct ext4_xattr_entry
*entry
;
40 struct ext4_inode
*raw_inode
;
43 min_offs
= EXT4_SB(inode
->i_sb
)->s_inode_size
-
44 EXT4_GOOD_OLD_INODE_SIZE
-
45 EXT4_I(inode
)->i_extra_isize
-
46 sizeof(struct ext4_xattr_ibody_header
);
49 * We need to subtract another sizeof(__u32) since an in-inode xattr
50 * needs an empty 4 bytes to indicate the gap between the xattr entry
51 * and the name/value pair.
53 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
))
54 return EXT4_XATTR_SIZE(min_offs
-
55 EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA
)) -
56 EXT4_XATTR_ROUND
- sizeof(__u32
));
58 raw_inode
= ext4_raw_inode(iloc
);
59 header
= IHDR(inode
, raw_inode
);
60 entry
= IFIRST(header
);
62 /* Compute min_offs. */
63 for (; !IS_LAST_ENTRY(entry
); entry
= EXT4_XATTR_NEXT(entry
)) {
64 if (!entry
->e_value_block
&& entry
->e_value_size
) {
65 size_t offs
= le16_to_cpu(entry
->e_value_offs
);
71 ((void *)entry
- (void *)IFIRST(header
)) - sizeof(__u32
);
73 if (EXT4_I(inode
)->i_inline_off
) {
74 entry
= (struct ext4_xattr_entry
*)
75 ((void *)raw_inode
+ EXT4_I(inode
)->i_inline_off
);
77 free
+= EXT4_XATTR_SIZE(le32_to_cpu(entry
->e_value_size
));
81 free
-= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA
));
83 if (free
> EXT4_XATTR_ROUND
)
84 free
= EXT4_XATTR_SIZE(free
- EXT4_XATTR_ROUND
);
93 * Get the maximum size we now can store in an inode.
94 * If we can't find the space for a xattr entry, don't use the space
95 * of the extents since we have no space to indicate the inline data.
97 int ext4_get_max_inline_size(struct inode
*inode
)
99 int error
, max_inline_size
;
100 struct ext4_iloc iloc
;
102 if (EXT4_I(inode
)->i_extra_isize
== 0)
105 error
= ext4_get_inode_loc(inode
, &iloc
);
107 ext4_error_inode(inode
, __func__
, __LINE__
, 0,
108 "can't get inode location %lu",
113 down_read(&EXT4_I(inode
)->xattr_sem
);
114 max_inline_size
= get_max_inline_xattr_value_size(inode
, &iloc
);
115 up_read(&EXT4_I(inode
)->xattr_sem
);
119 if (!max_inline_size
)
122 return max_inline_size
+ EXT4_MIN_INLINE_DATA_SIZE
;
126 * this function does not take xattr_sem, which is OK because it is
127 * currently only used in a code path coming form ext4_iget, before
128 * the new inode has been unlocked
130 int ext4_find_inline_data_nolock(struct inode
*inode
)
132 struct ext4_xattr_ibody_find is
= {
133 .s
= { .not_found
= -ENODATA
, },
135 struct ext4_xattr_info i
= {
136 .name_index
= EXT4_XATTR_INDEX_SYSTEM
,
137 .name
= EXT4_XATTR_SYSTEM_DATA
,
141 if (EXT4_I(inode
)->i_extra_isize
== 0)
144 error
= ext4_get_inode_loc(inode
, &is
.iloc
);
148 error
= ext4_xattr_ibody_find(inode
, &i
, &is
);
152 if (!is
.s
.not_found
) {
153 EXT4_I(inode
)->i_inline_off
= (u16
)((void *)is
.s
.here
-
154 (void *)ext4_raw_inode(&is
.iloc
));
155 EXT4_I(inode
)->i_inline_size
= EXT4_MIN_INLINE_DATA_SIZE
+
156 le32_to_cpu(is
.s
.here
->e_value_size
);
157 ext4_set_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
164 static int ext4_read_inline_data(struct inode
*inode
, void *buffer
,
166 struct ext4_iloc
*iloc
)
168 struct ext4_xattr_entry
*entry
;
169 struct ext4_xattr_ibody_header
*header
;
171 struct ext4_inode
*raw_inode
;
176 BUG_ON(len
> EXT4_I(inode
)->i_inline_size
);
178 cp_len
= len
< EXT4_MIN_INLINE_DATA_SIZE
?
179 len
: EXT4_MIN_INLINE_DATA_SIZE
;
181 raw_inode
= ext4_raw_inode(iloc
);
182 memcpy(buffer
, (void *)(raw_inode
->i_block
), cp_len
);
190 header
= IHDR(inode
, raw_inode
);
191 entry
= (struct ext4_xattr_entry
*)((void *)raw_inode
+
192 EXT4_I(inode
)->i_inline_off
);
193 len
= min_t(unsigned int, len
,
194 (unsigned int)le32_to_cpu(entry
->e_value_size
));
197 (void *)IFIRST(header
) + le16_to_cpu(entry
->e_value_offs
), len
);
205 * write the buffer to the inline inode.
206 * If 'create' is set, we don't need to do the extra copy in the xattr
207 * value since it is already handled by ext4_xattr_ibody_inline_set.
208 * That saves us one memcpy.
210 static void ext4_write_inline_data(struct inode
*inode
, struct ext4_iloc
*iloc
,
211 void *buffer
, loff_t pos
, unsigned int len
)
213 struct ext4_xattr_entry
*entry
;
214 struct ext4_xattr_ibody_header
*header
;
215 struct ext4_inode
*raw_inode
;
218 BUG_ON(!EXT4_I(inode
)->i_inline_off
);
219 BUG_ON(pos
+ len
> EXT4_I(inode
)->i_inline_size
);
221 raw_inode
= ext4_raw_inode(iloc
);
224 if (pos
< EXT4_MIN_INLINE_DATA_SIZE
) {
225 cp_len
= pos
+ len
> EXT4_MIN_INLINE_DATA_SIZE
?
226 EXT4_MIN_INLINE_DATA_SIZE
- pos
: len
;
227 memcpy((void *)raw_inode
->i_block
+ pos
, buffer
, cp_len
);
237 pos
-= EXT4_MIN_INLINE_DATA_SIZE
;
238 header
= IHDR(inode
, raw_inode
);
239 entry
= (struct ext4_xattr_entry
*)((void *)raw_inode
+
240 EXT4_I(inode
)->i_inline_off
);
242 memcpy((void *)IFIRST(header
) + le16_to_cpu(entry
->e_value_offs
) + pos
,
246 static int ext4_create_inline_data(handle_t
*handle
,
247 struct inode
*inode
, unsigned len
)
251 struct ext4_xattr_ibody_find is
= {
252 .s
= { .not_found
= -ENODATA
, },
254 struct ext4_xattr_info i
= {
255 .name_index
= EXT4_XATTR_INDEX_SYSTEM
,
256 .name
= EXT4_XATTR_SYSTEM_DATA
,
259 error
= ext4_get_inode_loc(inode
, &is
.iloc
);
263 BUFFER_TRACE(is
.iloc
.bh
, "get_write_access");
264 error
= ext4_journal_get_write_access(handle
, is
.iloc
.bh
);
268 if (len
> EXT4_MIN_INLINE_DATA_SIZE
) {
269 value
= EXT4_ZERO_XATTR_VALUE
;
270 len
-= EXT4_MIN_INLINE_DATA_SIZE
;
276 /* Insert the the xttr entry. */
280 error
= ext4_xattr_ibody_find(inode
, &i
, &is
);
284 BUG_ON(!is
.s
.not_found
);
286 error
= ext4_xattr_ibody_inline_set(handle
, inode
, &i
, &is
);
288 if (error
== -ENOSPC
)
289 ext4_clear_inode_state(inode
,
290 EXT4_STATE_MAY_INLINE_DATA
);
294 memset((void *)ext4_raw_inode(&is
.iloc
)->i_block
,
295 0, EXT4_MIN_INLINE_DATA_SIZE
);
297 EXT4_I(inode
)->i_inline_off
= (u16
)((void *)is
.s
.here
-
298 (void *)ext4_raw_inode(&is
.iloc
));
299 EXT4_I(inode
)->i_inline_size
= len
+ EXT4_MIN_INLINE_DATA_SIZE
;
300 ext4_clear_inode_flag(inode
, EXT4_INODE_EXTENTS
);
301 ext4_set_inode_flag(inode
, EXT4_INODE_INLINE_DATA
);
303 * Propagate changes to inode->i_flags as well - e.g. S_DAX may
306 ext4_set_inode_flags(inode
);
308 error
= ext4_mark_iloc_dirty(handle
, inode
, &is
.iloc
);
315 static int ext4_update_inline_data(handle_t
*handle
, struct inode
*inode
,
320 struct ext4_xattr_ibody_find is
= {
321 .s
= { .not_found
= -ENODATA
, },
323 struct ext4_xattr_info i
= {
324 .name_index
= EXT4_XATTR_INDEX_SYSTEM
,
325 .name
= EXT4_XATTR_SYSTEM_DATA
,
328 /* If the old space is ok, write the data directly. */
329 if (len
<= EXT4_I(inode
)->i_inline_size
)
332 error
= ext4_get_inode_loc(inode
, &is
.iloc
);
336 error
= ext4_xattr_ibody_find(inode
, &i
, &is
);
340 BUG_ON(is
.s
.not_found
);
342 len
-= EXT4_MIN_INLINE_DATA_SIZE
;
343 value
= kzalloc(len
, GFP_NOFS
);
349 error
= ext4_xattr_ibody_get(inode
, i
.name_index
, i
.name
,
351 if (error
== -ENODATA
)
354 BUFFER_TRACE(is
.iloc
.bh
, "get_write_access");
355 error
= ext4_journal_get_write_access(handle
, is
.iloc
.bh
);
359 /* Update the xttr entry. */
363 error
= ext4_xattr_ibody_inline_set(handle
, inode
, &i
, &is
);
367 EXT4_I(inode
)->i_inline_off
= (u16
)((void *)is
.s
.here
-
368 (void *)ext4_raw_inode(&is
.iloc
));
369 EXT4_I(inode
)->i_inline_size
= EXT4_MIN_INLINE_DATA_SIZE
+
370 le32_to_cpu(is
.s
.here
->e_value_size
);
371 ext4_set_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
373 error
= ext4_mark_iloc_dirty(handle
, inode
, &is
.iloc
);
381 static int ext4_prepare_inline_data(handle_t
*handle
, struct inode
*inode
,
384 int ret
, size
, no_expand
;
385 struct ext4_inode_info
*ei
= EXT4_I(inode
);
387 if (!ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
))
390 size
= ext4_get_max_inline_size(inode
);
394 ext4_write_lock_xattr(inode
, &no_expand
);
396 if (ei
->i_inline_off
)
397 ret
= ext4_update_inline_data(handle
, inode
, len
);
399 ret
= ext4_create_inline_data(handle
, inode
, len
);
401 ext4_write_unlock_xattr(inode
, &no_expand
);
405 static int ext4_destroy_inline_data_nolock(handle_t
*handle
,
408 struct ext4_inode_info
*ei
= EXT4_I(inode
);
409 struct ext4_xattr_ibody_find is
= {
410 .s
= { .not_found
= 0, },
412 struct ext4_xattr_info i
= {
413 .name_index
= EXT4_XATTR_INDEX_SYSTEM
,
414 .name
= EXT4_XATTR_SYSTEM_DATA
,
420 if (!ei
->i_inline_off
)
423 error
= ext4_get_inode_loc(inode
, &is
.iloc
);
427 error
= ext4_xattr_ibody_find(inode
, &i
, &is
);
431 BUFFER_TRACE(is
.iloc
.bh
, "get_write_access");
432 error
= ext4_journal_get_write_access(handle
, is
.iloc
.bh
);
436 error
= ext4_xattr_ibody_inline_set(handle
, inode
, &i
, &is
);
440 memset((void *)ext4_raw_inode(&is
.iloc
)->i_block
,
441 0, EXT4_MIN_INLINE_DATA_SIZE
);
443 if (ext4_has_feature_extents(inode
->i_sb
)) {
444 if (S_ISDIR(inode
->i_mode
) ||
445 S_ISREG(inode
->i_mode
) || S_ISLNK(inode
->i_mode
)) {
446 ext4_set_inode_flag(inode
, EXT4_INODE_EXTENTS
);
447 ext4_ext_tree_init(handle
, inode
);
450 ext4_clear_inode_flag(inode
, EXT4_INODE_INLINE_DATA
);
452 * Propagate changes to inode->i_flags as well - e.g. S_DAX may
455 ext4_set_inode_flags(inode
);
458 error
= ext4_mark_iloc_dirty(handle
, inode
, &is
.iloc
);
460 EXT4_I(inode
)->i_inline_off
= 0;
461 EXT4_I(inode
)->i_inline_size
= 0;
462 ext4_clear_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
465 if (error
== -ENODATA
)
470 static int ext4_read_inline_page(struct inode
*inode
, struct page
*page
)
475 struct ext4_iloc iloc
;
477 BUG_ON(!PageLocked(page
));
478 BUG_ON(!ext4_has_inline_data(inode
));
481 if (!EXT4_I(inode
)->i_inline_off
) {
482 ext4_warning(inode
->i_sb
, "inode %lu doesn't have inline data.",
487 ret
= ext4_get_inode_loc(inode
, &iloc
);
491 len
= min_t(size_t, ext4_get_inline_size(inode
), i_size_read(inode
));
492 kaddr
= kmap_atomic(page
);
493 ret
= ext4_read_inline_data(inode
, kaddr
, len
, &iloc
);
494 flush_dcache_page(page
);
495 kunmap_atomic(kaddr
);
496 zero_user_segment(page
, len
, PAGE_SIZE
);
497 SetPageUptodate(page
);
504 int ext4_readpage_inline(struct inode
*inode
, struct page
*page
)
508 down_read(&EXT4_I(inode
)->xattr_sem
);
509 if (!ext4_has_inline_data(inode
)) {
510 up_read(&EXT4_I(inode
)->xattr_sem
);
515 * Current inline data can only exist in the 1st page,
516 * So for all the other pages, just set them uptodate.
519 ret
= ext4_read_inline_page(inode
, page
);
520 else if (!PageUptodate(page
)) {
521 zero_user_segment(page
, 0, PAGE_SIZE
);
522 SetPageUptodate(page
);
525 up_read(&EXT4_I(inode
)->xattr_sem
);
528 return ret
>= 0 ? 0 : ret
;
531 static int ext4_convert_inline_data_to_extent(struct address_space
*mapping
,
535 int ret
, needed_blocks
, no_expand
;
536 handle_t
*handle
= NULL
;
537 int retries
= 0, sem_held
= 0;
538 struct page
*page
= NULL
;
540 struct ext4_iloc iloc
;
542 if (!ext4_has_inline_data(inode
)) {
544 * clear the flag so that no new write
545 * will trap here again.
547 ext4_clear_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
551 needed_blocks
= ext4_writepage_trans_blocks(inode
);
553 ret
= ext4_get_inode_loc(inode
, &iloc
);
558 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
559 if (IS_ERR(handle
)) {
560 ret
= PTR_ERR(handle
);
565 /* We cannot recurse into the filesystem as the transaction is already
567 flags
|= AOP_FLAG_NOFS
;
569 page
= grab_cache_page_write_begin(mapping
, 0, flags
);
575 ext4_write_lock_xattr(inode
, &no_expand
);
577 /* If some one has already done this for us, just exit. */
578 if (!ext4_has_inline_data(inode
)) {
584 to
= ext4_get_inline_size(inode
);
585 if (!PageUptodate(page
)) {
586 ret
= ext4_read_inline_page(inode
, page
);
591 ret
= ext4_destroy_inline_data_nolock(handle
, inode
);
595 if (ext4_should_dioread_nolock(inode
)) {
596 ret
= __block_write_begin(page
, from
, to
,
597 ext4_get_block_unwritten
);
599 ret
= __block_write_begin(page
, from
, to
, ext4_get_block
);
601 if (!ret
&& ext4_should_journal_data(inode
)) {
602 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
604 do_journal_get_write_access
);
611 ext4_orphan_add(handle
, inode
);
612 ext4_write_unlock_xattr(inode
, &no_expand
);
614 ext4_journal_stop(handle
);
616 ext4_truncate_failed_write(inode
);
618 * If truncate failed early the inode might
619 * still be on the orphan list; we need to
620 * make sure the inode is removed from the
621 * orphan list in that case.
624 ext4_orphan_del(NULL
, inode
);
627 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
631 block_commit_write(page
, from
, to
);
638 ext4_write_unlock_xattr(inode
, &no_expand
);
640 ext4_journal_stop(handle
);
646 * Try to write data in the inode.
647 * If the inode has inline data, check whether the new write can be
648 * in the inode also. If not, create the page the handle, move the data
649 * to the page make it update and let the later codes create extent for it.
651 int ext4_try_to_write_inline_data(struct address_space
*mapping
,
653 loff_t pos
, unsigned len
,
660 struct ext4_iloc iloc
;
662 if (pos
+ len
> ext4_get_max_inline_size(inode
))
665 ret
= ext4_get_inode_loc(inode
, &iloc
);
670 * The possible write could happen in the inode,
671 * so try to reserve the space in inode first.
673 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
674 if (IS_ERR(handle
)) {
675 ret
= PTR_ERR(handle
);
680 ret
= ext4_prepare_inline_data(handle
, inode
, pos
+ len
);
681 if (ret
&& ret
!= -ENOSPC
)
684 /* We don't have space in inline inode, so convert it to extent. */
685 if (ret
== -ENOSPC
) {
686 ext4_journal_stop(handle
);
691 flags
|= AOP_FLAG_NOFS
;
693 page
= grab_cache_page_write_begin(mapping
, 0, flags
);
700 down_read(&EXT4_I(inode
)->xattr_sem
);
701 if (!ext4_has_inline_data(inode
)) {
708 if (!PageUptodate(page
)) {
709 ret
= ext4_read_inline_page(inode
, page
);
717 up_read(&EXT4_I(inode
)->xattr_sem
);
720 ext4_journal_stop(handle
);
724 return ext4_convert_inline_data_to_extent(mapping
,
728 int ext4_write_inline_data_end(struct inode
*inode
, loff_t pos
, unsigned len
,
729 unsigned copied
, struct page
*page
)
733 struct ext4_iloc iloc
;
735 if (unlikely(copied
< len
)) {
736 if (!PageUptodate(page
)) {
742 ret
= ext4_get_inode_loc(inode
, &iloc
);
744 ext4_std_error(inode
->i_sb
, ret
);
749 ext4_write_lock_xattr(inode
, &no_expand
);
750 BUG_ON(!ext4_has_inline_data(inode
));
752 kaddr
= kmap_atomic(page
);
753 ext4_write_inline_data(inode
, &iloc
, kaddr
, pos
, len
);
754 kunmap_atomic(kaddr
);
755 SetPageUptodate(page
);
756 /* clear page dirty so that writepages wouldn't work for us. */
757 ClearPageDirty(page
);
759 ext4_write_unlock_xattr(inode
, &no_expand
);
766 ext4_journalled_write_inline_data(struct inode
*inode
,
772 struct ext4_iloc iloc
;
774 ret
= ext4_get_inode_loc(inode
, &iloc
);
776 ext4_std_error(inode
->i_sb
, ret
);
780 ext4_write_lock_xattr(inode
, &no_expand
);
781 kaddr
= kmap_atomic(page
);
782 ext4_write_inline_data(inode
, &iloc
, kaddr
, 0, len
);
783 kunmap_atomic(kaddr
);
784 ext4_write_unlock_xattr(inode
, &no_expand
);
790 * Try to make the page cache and handle ready for the inline data case.
791 * We can call this function in 2 cases:
792 * 1. The inode is created and the first write exceeds inline size. We can
793 * clear the inode state safely.
794 * 2. The inode has inline data, then we need to read the data, make it
795 * update and dirty so that ext4_da_writepages can handle it. We don't
796 * need to start the journal since the file's metatdata isn't changed now.
798 static int ext4_da_convert_inline_data_to_extent(struct address_space
*mapping
,
803 int ret
= 0, inline_size
;
806 page
= grab_cache_page_write_begin(mapping
, 0, flags
);
810 down_read(&EXT4_I(inode
)->xattr_sem
);
811 if (!ext4_has_inline_data(inode
)) {
812 ext4_clear_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
816 inline_size
= ext4_get_inline_size(inode
);
818 if (!PageUptodate(page
)) {
819 ret
= ext4_read_inline_page(inode
, page
);
824 ret
= __block_write_begin(page
, 0, inline_size
,
825 ext4_da_get_block_prep
);
827 up_read(&EXT4_I(inode
)->xattr_sem
);
830 ext4_truncate_failed_write(inode
);
835 SetPageUptodate(page
);
836 ext4_clear_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
837 *fsdata
= (void *)CONVERT_INLINE_DATA
;
840 up_read(&EXT4_I(inode
)->xattr_sem
);
849 * Prepare the write for the inline data.
850 * If the the data can be written into the inode, we just read
851 * the page and make it uptodate, and start the journal.
852 * Otherwise read the page, makes it dirty so that it can be
853 * handle in writepages(the i_disksize update is left to the
854 * normal ext4_da_write_end).
856 int ext4_da_write_inline_data_begin(struct address_space
*mapping
,
858 loff_t pos
, unsigned len
,
863 int ret
, inline_size
;
866 struct ext4_iloc iloc
;
869 ret
= ext4_get_inode_loc(inode
, &iloc
);
874 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
875 if (IS_ERR(handle
)) {
876 ret
= PTR_ERR(handle
);
880 inline_size
= ext4_get_max_inline_size(inode
);
883 if (inline_size
>= pos
+ len
) {
884 ret
= ext4_prepare_inline_data(handle
, inode
, pos
+ len
);
885 if (ret
&& ret
!= -ENOSPC
)
890 * We cannot recurse into the filesystem as the transaction
891 * is already started.
893 flags
|= AOP_FLAG_NOFS
;
895 if (ret
== -ENOSPC
) {
896 ret
= ext4_da_convert_inline_data_to_extent(mapping
,
900 ext4_journal_stop(handle
);
901 if (ret
== -ENOSPC
&&
902 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
908 page
= grab_cache_page_write_begin(mapping
, 0, flags
);
914 down_read(&EXT4_I(inode
)->xattr_sem
);
915 if (!ext4_has_inline_data(inode
)) {
917 goto out_release_page
;
920 if (!PageUptodate(page
)) {
921 ret
= ext4_read_inline_page(inode
, page
);
923 goto out_release_page
;
926 up_read(&EXT4_I(inode
)->xattr_sem
);
931 up_read(&EXT4_I(inode
)->xattr_sem
);
935 ext4_journal_stop(handle
);
941 int ext4_da_write_inline_data_end(struct inode
*inode
, loff_t pos
,
942 unsigned len
, unsigned copied
,
945 int i_size_changed
= 0;
948 ret
= ext4_write_inline_data_end(inode
, pos
, len
, copied
, page
);
957 * No need to use i_size_read() here, the i_size
958 * cannot change under us because we hold i_mutex.
960 * But it's important to update i_size while still holding page lock:
961 * page writeout could otherwise come in and zero beyond i_size.
963 if (pos
+copied
> inode
->i_size
) {
964 i_size_write(inode
, pos
+copied
);
971 * Don't mark the inode dirty under page lock. First, it unnecessarily
972 * makes the holding time of page lock longer. Second, it forces lock
973 * ordering of page lock and transaction start for journaling
977 mark_inode_dirty(inode
);
982 #ifdef INLINE_DIR_DEBUG
983 void ext4_show_inline_dir(struct inode
*dir
, struct buffer_head
*bh
,
984 void *inline_start
, int inline_size
)
987 unsigned short de_len
;
988 struct ext4_dir_entry_2
*de
= inline_start
;
989 void *dlimit
= inline_start
+ inline_size
;
991 trace_printk("inode %lu\n", dir
->i_ino
);
993 while ((void *)de
< dlimit
) {
994 de_len
= ext4_rec_len_from_disk(de
->rec_len
, inline_size
);
995 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
996 offset
, de_len
, de
->name_len
, de
->name
,
997 de
->name_len
, le32_to_cpu(de
->inode
));
998 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
999 inline_start
, inline_size
, offset
))
1003 de
= (struct ext4_dir_entry_2
*) ((char *) de
+ de_len
);
1007 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1011 * Add a new entry into a inline dir.
1012 * It will return -ENOSPC if no space is available, and -EIO
1013 * and -EEXIST if directory entry already exists.
1015 static int ext4_add_dirent_to_inline(handle_t
*handle
,
1016 struct ext4_filename
*fname
,
1018 struct inode
*inode
,
1019 struct ext4_iloc
*iloc
,
1020 void *inline_start
, int inline_size
)
1023 struct ext4_dir_entry_2
*de
;
1025 err
= ext4_find_dest_de(dir
, inode
, iloc
->bh
, inline_start
,
1026 inline_size
, fname
, &de
);
1030 BUFFER_TRACE(iloc
->bh
, "get_write_access");
1031 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
1034 ext4_insert_dentry(dir
, inode
, de
, inline_size
, fname
);
1036 ext4_show_inline_dir(dir
, iloc
->bh
, inline_start
, inline_size
);
1039 * XXX shouldn't update any times until successful
1040 * completion of syscall, but too many callers depend
1043 * XXX similarly, too many callers depend on
1044 * ext4_new_inode() setting the times, but error
1045 * recovery deletes the inode, so the worst that can
1046 * happen is that the times are slightly out of date
1047 * and/or different from the directory change time.
1049 dir
->i_mtime
= dir
->i_ctime
= current_time(dir
);
1050 ext4_update_dx_flag(dir
);
1052 ext4_mark_inode_dirty(handle
, dir
);
1056 static void *ext4_get_inline_xattr_pos(struct inode
*inode
,
1057 struct ext4_iloc
*iloc
)
1059 struct ext4_xattr_entry
*entry
;
1060 struct ext4_xattr_ibody_header
*header
;
1062 BUG_ON(!EXT4_I(inode
)->i_inline_off
);
1064 header
= IHDR(inode
, ext4_raw_inode(iloc
));
1065 entry
= (struct ext4_xattr_entry
*)((void *)ext4_raw_inode(iloc
) +
1066 EXT4_I(inode
)->i_inline_off
);
1068 return (void *)IFIRST(header
) + le16_to_cpu(entry
->e_value_offs
);
1071 /* Set the final de to cover the whole block. */
1072 static void ext4_update_final_de(void *de_buf
, int old_size
, int new_size
)
1074 struct ext4_dir_entry_2
*de
, *prev_de
;
1078 de
= (struct ext4_dir_entry_2
*)de_buf
;
1080 limit
= de_buf
+ old_size
;
1083 de_len
= ext4_rec_len_from_disk(de
->rec_len
, old_size
);
1085 de
= (struct ext4_dir_entry_2
*)de_buf
;
1086 } while (de_buf
< limit
);
1088 prev_de
->rec_len
= ext4_rec_len_to_disk(de_len
+ new_size
-
1089 old_size
, new_size
);
1091 /* this is just created, so create an empty entry. */
1093 de
->rec_len
= ext4_rec_len_to_disk(new_size
, new_size
);
1097 static int ext4_update_inline_dir(handle_t
*handle
, struct inode
*dir
,
1098 struct ext4_iloc
*iloc
)
1101 int old_size
= EXT4_I(dir
)->i_inline_size
- EXT4_MIN_INLINE_DATA_SIZE
;
1102 int new_size
= get_max_inline_xattr_value_size(dir
, iloc
);
1104 if (new_size
- old_size
<= EXT4_DIR_REC_LEN(1))
1107 ret
= ext4_update_inline_data(handle
, dir
,
1108 new_size
+ EXT4_MIN_INLINE_DATA_SIZE
);
1112 ext4_update_final_de(ext4_get_inline_xattr_pos(dir
, iloc
), old_size
,
1113 EXT4_I(dir
)->i_inline_size
-
1114 EXT4_MIN_INLINE_DATA_SIZE
);
1115 dir
->i_size
= EXT4_I(dir
)->i_disksize
= EXT4_I(dir
)->i_inline_size
;
1119 static void ext4_restore_inline_data(handle_t
*handle
, struct inode
*inode
,
1120 struct ext4_iloc
*iloc
,
1121 void *buf
, int inline_size
)
1123 ext4_create_inline_data(handle
, inode
, inline_size
);
1124 ext4_write_inline_data(inode
, iloc
, buf
, 0, inline_size
);
1125 ext4_set_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
1128 static int ext4_finish_convert_inline_dir(handle_t
*handle
,
1129 struct inode
*inode
,
1130 struct buffer_head
*dir_block
,
1134 int err
, csum_size
= 0, header_size
= 0;
1135 struct ext4_dir_entry_2
*de
;
1136 struct ext4_dir_entry_tail
*t
;
1137 void *target
= dir_block
->b_data
;
1140 * First create "." and ".." and then copy the dir information
1141 * back to the block.
1143 de
= (struct ext4_dir_entry_2
*)target
;
1144 de
= ext4_init_dot_dotdot(inode
, de
,
1145 inode
->i_sb
->s_blocksize
, csum_size
,
1146 le32_to_cpu(((struct ext4_dir_entry_2
*)buf
)->inode
), 1);
1147 header_size
= (void *)de
- target
;
1149 memcpy((void *)de
, buf
+ EXT4_INLINE_DOTDOT_SIZE
,
1150 inline_size
- EXT4_INLINE_DOTDOT_SIZE
);
1152 if (ext4_has_metadata_csum(inode
->i_sb
))
1153 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1155 inode
->i_size
= inode
->i_sb
->s_blocksize
;
1156 i_size_write(inode
, inode
->i_sb
->s_blocksize
);
1157 EXT4_I(inode
)->i_disksize
= inode
->i_sb
->s_blocksize
;
1158 ext4_update_final_de(dir_block
->b_data
,
1159 inline_size
- EXT4_INLINE_DOTDOT_SIZE
+ header_size
,
1160 inode
->i_sb
->s_blocksize
- csum_size
);
1163 t
= EXT4_DIRENT_TAIL(dir_block
->b_data
,
1164 inode
->i_sb
->s_blocksize
);
1165 initialize_dirent_tail(t
, inode
->i_sb
->s_blocksize
);
1167 set_buffer_uptodate(dir_block
);
1168 err
= ext4_handle_dirty_dirent_node(handle
, inode
, dir_block
);
1171 set_buffer_verified(dir_block
);
1172 return ext4_mark_inode_dirty(handle
, inode
);
1175 static int ext4_convert_inline_data_nolock(handle_t
*handle
,
1176 struct inode
*inode
,
1177 struct ext4_iloc
*iloc
)
1181 struct buffer_head
*data_bh
= NULL
;
1182 struct ext4_map_blocks map
;
1185 inline_size
= ext4_get_inline_size(inode
);
1186 buf
= kmalloc(inline_size
, GFP_NOFS
);
1192 error
= ext4_read_inline_data(inode
, buf
, inline_size
, iloc
);
1197 * Make sure the inline directory entries pass checks before we try to
1198 * convert them, so that we avoid touching stuff that needs fsck.
1200 if (S_ISDIR(inode
->i_mode
)) {
1201 error
= ext4_check_all_de(inode
, iloc
->bh
,
1202 buf
+ EXT4_INLINE_DOTDOT_SIZE
,
1203 inline_size
- EXT4_INLINE_DOTDOT_SIZE
);
1208 error
= ext4_destroy_inline_data_nolock(handle
, inode
);
1215 error
= ext4_map_blocks(handle
, inode
, &map
, EXT4_GET_BLOCKS_CREATE
);
1218 if (!(map
.m_flags
& EXT4_MAP_MAPPED
)) {
1223 data_bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
1229 lock_buffer(data_bh
);
1230 error
= ext4_journal_get_create_access(handle
, data_bh
);
1232 unlock_buffer(data_bh
);
1236 memset(data_bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
1238 if (!S_ISDIR(inode
->i_mode
)) {
1239 memcpy(data_bh
->b_data
, buf
, inline_size
);
1240 set_buffer_uptodate(data_bh
);
1241 error
= ext4_handle_dirty_metadata(handle
,
1244 error
= ext4_finish_convert_inline_dir(handle
, inode
, data_bh
,
1248 unlock_buffer(data_bh
);
1251 ext4_restore_inline_data(handle
, inode
, iloc
, buf
, inline_size
);
1260 * Try to add the new entry to the inline data.
1261 * If succeeds, return 0. If not, extended the inline dir and copied data to
1262 * the new created block.
1264 int ext4_try_add_inline_entry(handle_t
*handle
, struct ext4_filename
*fname
,
1265 struct inode
*dir
, struct inode
*inode
)
1267 int ret
, inline_size
, no_expand
;
1269 struct ext4_iloc iloc
;
1271 ret
= ext4_get_inode_loc(dir
, &iloc
);
1275 ext4_write_lock_xattr(dir
, &no_expand
);
1276 if (!ext4_has_inline_data(dir
))
1279 inline_start
= (void *)ext4_raw_inode(&iloc
)->i_block
+
1280 EXT4_INLINE_DOTDOT_SIZE
;
1281 inline_size
= EXT4_MIN_INLINE_DATA_SIZE
- EXT4_INLINE_DOTDOT_SIZE
;
1283 ret
= ext4_add_dirent_to_inline(handle
, fname
, dir
, inode
, &iloc
,
1284 inline_start
, inline_size
);
1288 /* check whether it can be inserted to inline xattr space. */
1289 inline_size
= EXT4_I(dir
)->i_inline_size
-
1290 EXT4_MIN_INLINE_DATA_SIZE
;
1292 /* Try to use the xattr space.*/
1293 ret
= ext4_update_inline_dir(handle
, dir
, &iloc
);
1294 if (ret
&& ret
!= -ENOSPC
)
1297 inline_size
= EXT4_I(dir
)->i_inline_size
-
1298 EXT4_MIN_INLINE_DATA_SIZE
;
1302 inline_start
= ext4_get_inline_xattr_pos(dir
, &iloc
);
1304 ret
= ext4_add_dirent_to_inline(handle
, fname
, dir
,
1305 inode
, &iloc
, inline_start
,
1313 * The inline space is filled up, so create a new block for it.
1314 * As the extent tree will be created, we have to save the inline
1317 ret
= ext4_convert_inline_data_nolock(handle
, dir
, &iloc
);
1320 ext4_mark_inode_dirty(handle
, dir
);
1321 ext4_write_unlock_xattr(dir
, &no_expand
);
1327 * This function fills a red-black tree with information from an
1328 * inlined dir. It returns the number directory entries loaded
1329 * into the tree. If there is an error it is returned in err.
1331 int htree_inlinedir_to_tree(struct file
*dir_file
,
1332 struct inode
*dir
, ext4_lblk_t block
,
1333 struct dx_hash_info
*hinfo
,
1334 __u32 start_hash
, __u32 start_minor_hash
,
1335 int *has_inline_data
)
1337 int err
= 0, count
= 0;
1338 unsigned int parent_ino
;
1340 struct ext4_dir_entry_2
*de
;
1341 struct inode
*inode
= file_inode(dir_file
);
1342 int ret
, inline_size
= 0;
1343 struct ext4_iloc iloc
;
1344 void *dir_buf
= NULL
;
1345 struct ext4_dir_entry_2 fake
;
1346 struct fscrypt_str tmp_str
;
1348 ret
= ext4_get_inode_loc(inode
, &iloc
);
1352 down_read(&EXT4_I(inode
)->xattr_sem
);
1353 if (!ext4_has_inline_data(inode
)) {
1354 up_read(&EXT4_I(inode
)->xattr_sem
);
1355 *has_inline_data
= 0;
1359 inline_size
= ext4_get_inline_size(inode
);
1360 dir_buf
= kmalloc(inline_size
, GFP_NOFS
);
1363 up_read(&EXT4_I(inode
)->xattr_sem
);
1367 ret
= ext4_read_inline_data(inode
, dir_buf
, inline_size
, &iloc
);
1368 up_read(&EXT4_I(inode
)->xattr_sem
);
1373 parent_ino
= le32_to_cpu(((struct ext4_dir_entry_2
*)dir_buf
)->inode
);
1374 while (pos
< inline_size
) {
1376 * As inlined dir doesn't store any information about '.' and
1377 * only the inode number of '..' is stored, we have to handle
1381 fake
.inode
= cpu_to_le32(inode
->i_ino
);
1383 strcpy(fake
.name
, ".");
1384 fake
.rec_len
= ext4_rec_len_to_disk(
1385 EXT4_DIR_REC_LEN(fake
.name_len
),
1387 ext4_set_de_type(inode
->i_sb
, &fake
, S_IFDIR
);
1389 pos
= EXT4_INLINE_DOTDOT_OFFSET
;
1390 } else if (pos
== EXT4_INLINE_DOTDOT_OFFSET
) {
1391 fake
.inode
= cpu_to_le32(parent_ino
);
1393 strcpy(fake
.name
, "..");
1394 fake
.rec_len
= ext4_rec_len_to_disk(
1395 EXT4_DIR_REC_LEN(fake
.name_len
),
1397 ext4_set_de_type(inode
->i_sb
, &fake
, S_IFDIR
);
1399 pos
= EXT4_INLINE_DOTDOT_SIZE
;
1401 de
= (struct ext4_dir_entry_2
*)(dir_buf
+ pos
);
1402 pos
+= ext4_rec_len_from_disk(de
->rec_len
, inline_size
);
1403 if (ext4_check_dir_entry(inode
, dir_file
, de
,
1405 inline_size
, pos
)) {
1411 ext4fs_dirhash(de
->name
, de
->name_len
, hinfo
);
1412 if ((hinfo
->hash
< start_hash
) ||
1413 ((hinfo
->hash
== start_hash
) &&
1414 (hinfo
->minor_hash
< start_minor_hash
)))
1418 tmp_str
.name
= de
->name
;
1419 tmp_str
.len
= de
->name_len
;
1420 err
= ext4_htree_store_dirent(dir_file
, hinfo
->hash
,
1421 hinfo
->minor_hash
, de
, &tmp_str
);
1436 * So this function is called when the volume is mkfsed with
1437 * dir_index disabled. In order to keep f_pos persistent
1438 * after we convert from an inlined dir to a blocked based,
1439 * we just pretend that we are a normal dir and return the
1440 * offset as if '.' and '..' really take place.
1443 int ext4_read_inline_dir(struct file
*file
,
1444 struct dir_context
*ctx
,
1445 int *has_inline_data
)
1447 unsigned int offset
, parent_ino
;
1449 struct ext4_dir_entry_2
*de
;
1450 struct super_block
*sb
;
1451 struct inode
*inode
= file_inode(file
);
1452 int ret
, inline_size
= 0;
1453 struct ext4_iloc iloc
;
1454 void *dir_buf
= NULL
;
1455 int dotdot_offset
, dotdot_size
, extra_offset
, extra_size
;
1457 ret
= ext4_get_inode_loc(inode
, &iloc
);
1461 down_read(&EXT4_I(inode
)->xattr_sem
);
1462 if (!ext4_has_inline_data(inode
)) {
1463 up_read(&EXT4_I(inode
)->xattr_sem
);
1464 *has_inline_data
= 0;
1468 inline_size
= ext4_get_inline_size(inode
);
1469 dir_buf
= kmalloc(inline_size
, GFP_NOFS
);
1472 up_read(&EXT4_I(inode
)->xattr_sem
);
1476 ret
= ext4_read_inline_data(inode
, dir_buf
, inline_size
, &iloc
);
1477 up_read(&EXT4_I(inode
)->xattr_sem
);
1483 parent_ino
= le32_to_cpu(((struct ext4_dir_entry_2
*)dir_buf
)->inode
);
1487 * dotdot_offset and dotdot_size is the real offset and
1488 * size for ".." and "." if the dir is block based while
1489 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1490 * So we will use extra_offset and extra_size to indicate them
1491 * during the inline dir iteration.
1493 dotdot_offset
= EXT4_DIR_REC_LEN(1);
1494 dotdot_size
= dotdot_offset
+ EXT4_DIR_REC_LEN(2);
1495 extra_offset
= dotdot_size
- EXT4_INLINE_DOTDOT_SIZE
;
1496 extra_size
= extra_offset
+ inline_size
;
1499 * If the version has changed since the last call to
1500 * readdir(2), then we might be pointing to an invalid
1501 * dirent right now. Scan from the start of the inline
1504 if (file
->f_version
!= inode
->i_version
) {
1505 for (i
= 0; i
< extra_size
&& i
< offset
;) {
1507 * "." is with offset 0 and
1508 * ".." is dotdot_offset.
1513 } else if (i
== dotdot_offset
) {
1517 /* for other entry, the real offset in
1518 * the buf has to be tuned accordingly.
1520 de
= (struct ext4_dir_entry_2
*)
1521 (dir_buf
+ i
- extra_offset
);
1522 /* It's too expensive to do a full
1523 * dirent test each time round this
1524 * loop, but we do have to test at
1525 * least that it is non-zero. A
1526 * failure will be detected in the
1527 * dirent test below. */
1528 if (ext4_rec_len_from_disk(de
->rec_len
, extra_size
)
1529 < EXT4_DIR_REC_LEN(1))
1531 i
+= ext4_rec_len_from_disk(de
->rec_len
,
1536 file
->f_version
= inode
->i_version
;
1539 while (ctx
->pos
< extra_size
) {
1540 if (ctx
->pos
== 0) {
1541 if (!dir_emit(ctx
, ".", 1, inode
->i_ino
, DT_DIR
))
1543 ctx
->pos
= dotdot_offset
;
1547 if (ctx
->pos
== dotdot_offset
) {
1548 if (!dir_emit(ctx
, "..", 2, parent_ino
, DT_DIR
))
1550 ctx
->pos
= dotdot_size
;
1554 de
= (struct ext4_dir_entry_2
*)
1555 (dir_buf
+ ctx
->pos
- extra_offset
);
1556 if (ext4_check_dir_entry(inode
, file
, de
, iloc
.bh
, dir_buf
,
1557 extra_size
, ctx
->pos
))
1559 if (le32_to_cpu(de
->inode
)) {
1560 if (!dir_emit(ctx
, de
->name
, de
->name_len
,
1561 le32_to_cpu(de
->inode
),
1562 get_dtype(sb
, de
->file_type
)))
1565 ctx
->pos
+= ext4_rec_len_from_disk(de
->rec_len
, extra_size
);
1573 struct buffer_head
*ext4_get_first_inline_block(struct inode
*inode
,
1574 struct ext4_dir_entry_2
**parent_de
,
1577 struct ext4_iloc iloc
;
1579 *retval
= ext4_get_inode_loc(inode
, &iloc
);
1583 *parent_de
= (struct ext4_dir_entry_2
*)ext4_raw_inode(&iloc
)->i_block
;
1589 * Try to create the inline data for the new dir.
1590 * If it succeeds, return 0, otherwise return the error.
1591 * In case of ENOSPC, the caller should create the normal disk layout dir.
1593 int ext4_try_create_inline_dir(handle_t
*handle
, struct inode
*parent
,
1594 struct inode
*inode
)
1596 int ret
, inline_size
= EXT4_MIN_INLINE_DATA_SIZE
;
1597 struct ext4_iloc iloc
;
1598 struct ext4_dir_entry_2
*de
;
1600 ret
= ext4_get_inode_loc(inode
, &iloc
);
1604 ret
= ext4_prepare_inline_data(handle
, inode
, inline_size
);
1609 * For inline dir, we only save the inode information for the ".."
1610 * and create a fake dentry to cover the left space.
1612 de
= (struct ext4_dir_entry_2
*)ext4_raw_inode(&iloc
)->i_block
;
1613 de
->inode
= cpu_to_le32(parent
->i_ino
);
1614 de
= (struct ext4_dir_entry_2
*)((void *)de
+ EXT4_INLINE_DOTDOT_SIZE
);
1616 de
->rec_len
= ext4_rec_len_to_disk(
1617 inline_size
- EXT4_INLINE_DOTDOT_SIZE
,
1619 set_nlink(inode
, 2);
1620 inode
->i_size
= EXT4_I(inode
)->i_disksize
= inline_size
;
1626 struct buffer_head
*ext4_find_inline_entry(struct inode
*dir
,
1627 struct ext4_filename
*fname
,
1628 const struct qstr
*d_name
,
1629 struct ext4_dir_entry_2
**res_dir
,
1630 int *has_inline_data
)
1633 struct ext4_iloc iloc
;
1637 if (ext4_get_inode_loc(dir
, &iloc
))
1640 down_read(&EXT4_I(dir
)->xattr_sem
);
1641 if (!ext4_has_inline_data(dir
)) {
1642 *has_inline_data
= 0;
1646 inline_start
= (void *)ext4_raw_inode(&iloc
)->i_block
+
1647 EXT4_INLINE_DOTDOT_SIZE
;
1648 inline_size
= EXT4_MIN_INLINE_DATA_SIZE
- EXT4_INLINE_DOTDOT_SIZE
;
1649 ret
= ext4_search_dir(iloc
.bh
, inline_start
, inline_size
,
1650 dir
, fname
, d_name
, 0, res_dir
);
1656 if (ext4_get_inline_size(dir
) == EXT4_MIN_INLINE_DATA_SIZE
)
1659 inline_start
= ext4_get_inline_xattr_pos(dir
, &iloc
);
1660 inline_size
= ext4_get_inline_size(dir
) - EXT4_MIN_INLINE_DATA_SIZE
;
1662 ret
= ext4_search_dir(iloc
.bh
, inline_start
, inline_size
,
1663 dir
, fname
, d_name
, 0, res_dir
);
1671 up_read(&EXT4_I(dir
)->xattr_sem
);
1675 int ext4_delete_inline_entry(handle_t
*handle
,
1677 struct ext4_dir_entry_2
*de_del
,
1678 struct buffer_head
*bh
,
1679 int *has_inline_data
)
1681 int err
, inline_size
, no_expand
;
1682 struct ext4_iloc iloc
;
1685 err
= ext4_get_inode_loc(dir
, &iloc
);
1689 ext4_write_lock_xattr(dir
, &no_expand
);
1690 if (!ext4_has_inline_data(dir
)) {
1691 *has_inline_data
= 0;
1695 if ((void *)de_del
- ((void *)ext4_raw_inode(&iloc
)->i_block
) <
1696 EXT4_MIN_INLINE_DATA_SIZE
) {
1697 inline_start
= (void *)ext4_raw_inode(&iloc
)->i_block
+
1698 EXT4_INLINE_DOTDOT_SIZE
;
1699 inline_size
= EXT4_MIN_INLINE_DATA_SIZE
-
1700 EXT4_INLINE_DOTDOT_SIZE
;
1702 inline_start
= ext4_get_inline_xattr_pos(dir
, &iloc
);
1703 inline_size
= ext4_get_inline_size(dir
) -
1704 EXT4_MIN_INLINE_DATA_SIZE
;
1707 BUFFER_TRACE(bh
, "get_write_access");
1708 err
= ext4_journal_get_write_access(handle
, bh
);
1712 err
= ext4_generic_delete_entry(handle
, dir
, de_del
, bh
,
1713 inline_start
, inline_size
, 0);
1717 err
= ext4_mark_inode_dirty(handle
, dir
);
1721 ext4_show_inline_dir(dir
, iloc
.bh
, inline_start
, inline_size
);
1723 ext4_write_unlock_xattr(dir
, &no_expand
);
1726 ext4_std_error(dir
->i_sb
, err
);
1731 * Get the inline dentry at offset.
1733 static inline struct ext4_dir_entry_2
*
1734 ext4_get_inline_entry(struct inode
*inode
,
1735 struct ext4_iloc
*iloc
,
1736 unsigned int offset
,
1737 void **inline_start
,
1742 BUG_ON(offset
> ext4_get_inline_size(inode
));
1744 if (offset
< EXT4_MIN_INLINE_DATA_SIZE
) {
1745 inline_pos
= (void *)ext4_raw_inode(iloc
)->i_block
;
1746 *inline_size
= EXT4_MIN_INLINE_DATA_SIZE
;
1748 inline_pos
= ext4_get_inline_xattr_pos(inode
, iloc
);
1749 offset
-= EXT4_MIN_INLINE_DATA_SIZE
;
1750 *inline_size
= ext4_get_inline_size(inode
) -
1751 EXT4_MIN_INLINE_DATA_SIZE
;
1755 *inline_start
= inline_pos
;
1756 return (struct ext4_dir_entry_2
*)(inline_pos
+ offset
);
1759 bool empty_inline_dir(struct inode
*dir
, int *has_inline_data
)
1761 int err
, inline_size
;
1762 struct ext4_iloc iloc
;
1764 unsigned int offset
;
1765 struct ext4_dir_entry_2
*de
;
1768 err
= ext4_get_inode_loc(dir
, &iloc
);
1770 EXT4_ERROR_INODE(dir
, "error %d getting inode %lu block",
1775 down_read(&EXT4_I(dir
)->xattr_sem
);
1776 if (!ext4_has_inline_data(dir
)) {
1777 *has_inline_data
= 0;
1781 de
= (struct ext4_dir_entry_2
*)ext4_raw_inode(&iloc
)->i_block
;
1782 if (!le32_to_cpu(de
->inode
)) {
1783 ext4_warning(dir
->i_sb
,
1784 "bad inline directory (dir #%lu) - no `..'",
1790 offset
= EXT4_INLINE_DOTDOT_SIZE
;
1791 while (offset
< dir
->i_size
) {
1792 de
= ext4_get_inline_entry(dir
, &iloc
, offset
,
1793 &inline_pos
, &inline_size
);
1794 if (ext4_check_dir_entry(dir
, NULL
, de
,
1795 iloc
.bh
, inline_pos
,
1796 inline_size
, offset
)) {
1797 ext4_warning(dir
->i_sb
,
1798 "bad inline directory (dir #%lu) - "
1799 "inode %u, rec_len %u, name_len %d"
1801 dir
->i_ino
, le32_to_cpu(de
->inode
),
1802 le16_to_cpu(de
->rec_len
), de
->name_len
,
1807 if (le32_to_cpu(de
->inode
)) {
1811 offset
+= ext4_rec_len_from_disk(de
->rec_len
, inline_size
);
1815 up_read(&EXT4_I(dir
)->xattr_sem
);
1820 int ext4_destroy_inline_data(handle_t
*handle
, struct inode
*inode
)
1824 ext4_write_lock_xattr(inode
, &no_expand
);
1825 ret
= ext4_destroy_inline_data_nolock(handle
, inode
);
1826 ext4_write_unlock_xattr(inode
, &no_expand
);
1831 int ext4_inline_data_fiemap(struct inode
*inode
,
1832 struct fiemap_extent_info
*fieinfo
,
1833 int *has_inline
, __u64 start
, __u64 len
)
1837 __u32 flags
= FIEMAP_EXTENT_DATA_INLINE
| FIEMAP_EXTENT_NOT_ALIGNED
|
1840 struct ext4_iloc iloc
;
1842 down_read(&EXT4_I(inode
)->xattr_sem
);
1843 if (!ext4_has_inline_data(inode
)) {
1847 inline_len
= min_t(size_t, ext4_get_inline_size(inode
),
1848 i_size_read(inode
));
1849 if (start
>= inline_len
)
1851 if (start
+ len
< inline_len
)
1852 inline_len
= start
+ len
;
1853 inline_len
-= start
;
1855 error
= ext4_get_inode_loc(inode
, &iloc
);
1859 physical
= (__u64
)iloc
.bh
->b_blocknr
<< inode
->i_sb
->s_blocksize_bits
;
1860 physical
+= (char *)ext4_raw_inode(&iloc
) - iloc
.bh
->b_data
;
1861 physical
+= offsetof(struct ext4_inode
, i_block
);
1864 error
= fiemap_fill_next_extent(fieinfo
, start
, physical
,
1868 up_read(&EXT4_I(inode
)->xattr_sem
);
1869 return (error
< 0 ? error
: 0);
1873 * Called during xattr set, and if we can sparse space 'needed',
1874 * just create the extent tree evict the data to the outer block.
1876 * We use jbd2 instead of page cache to move data to the 1st block
1877 * so that the whole transaction can be committed as a whole and
1878 * the data isn't lost because of the delayed page cache write.
1880 int ext4_try_to_evict_inline_data(handle_t
*handle
,
1881 struct inode
*inode
,
1885 struct ext4_xattr_entry
*entry
;
1886 struct ext4_inode
*raw_inode
;
1887 struct ext4_iloc iloc
;
1889 error
= ext4_get_inode_loc(inode
, &iloc
);
1893 raw_inode
= ext4_raw_inode(&iloc
);
1894 entry
= (struct ext4_xattr_entry
*)((void *)raw_inode
+
1895 EXT4_I(inode
)->i_inline_off
);
1896 if (EXT4_XATTR_LEN(entry
->e_name_len
) +
1897 EXT4_XATTR_SIZE(le32_to_cpu(entry
->e_value_size
)) < needed
) {
1902 error
= ext4_convert_inline_data_nolock(handle
, inode
, &iloc
);
1908 void ext4_inline_data_truncate(struct inode
*inode
, int *has_inline
)
1911 int inline_size
, value_len
, needed_blocks
, no_expand
;
1914 struct ext4_xattr_ibody_find is
= {
1915 .s
= { .not_found
= -ENODATA
, },
1917 struct ext4_xattr_info i
= {
1918 .name_index
= EXT4_XATTR_INDEX_SYSTEM
,
1919 .name
= EXT4_XATTR_SYSTEM_DATA
,
1923 needed_blocks
= ext4_writepage_trans_blocks(inode
);
1924 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, needed_blocks
);
1928 ext4_write_lock_xattr(inode
, &no_expand
);
1929 if (!ext4_has_inline_data(inode
)) {
1931 ext4_journal_stop(handle
);
1935 if (ext4_orphan_add(handle
, inode
))
1938 if (ext4_get_inode_loc(inode
, &is
.iloc
))
1941 down_write(&EXT4_I(inode
)->i_data_sem
);
1942 i_size
= inode
->i_size
;
1943 inline_size
= ext4_get_inline_size(inode
);
1944 EXT4_I(inode
)->i_disksize
= i_size
;
1946 if (i_size
< inline_size
) {
1947 /* Clear the content in the xattr space. */
1948 if (inline_size
> EXT4_MIN_INLINE_DATA_SIZE
) {
1949 if (ext4_xattr_ibody_find(inode
, &i
, &is
))
1952 BUG_ON(is
.s
.not_found
);
1954 value_len
= le32_to_cpu(is
.s
.here
->e_value_size
);
1955 value
= kmalloc(value_len
, GFP_NOFS
);
1959 if (ext4_xattr_ibody_get(inode
, i
.name_index
, i
.name
,
1964 i
.value_len
= i_size
> EXT4_MIN_INLINE_DATA_SIZE
?
1965 i_size
- EXT4_MIN_INLINE_DATA_SIZE
: 0;
1966 if (ext4_xattr_ibody_inline_set(handle
, inode
, &i
, &is
))
1970 /* Clear the content within i_blocks. */
1971 if (i_size
< EXT4_MIN_INLINE_DATA_SIZE
) {
1972 void *p
= (void *) ext4_raw_inode(&is
.iloc
)->i_block
;
1973 memset(p
+ i_size
, 0,
1974 EXT4_MIN_INLINE_DATA_SIZE
- i_size
);
1977 EXT4_I(inode
)->i_inline_size
= i_size
<
1978 EXT4_MIN_INLINE_DATA_SIZE
?
1979 EXT4_MIN_INLINE_DATA_SIZE
: i_size
;
1983 up_write(&EXT4_I(inode
)->i_data_sem
);
1986 ext4_write_unlock_xattr(inode
, &no_expand
);
1989 ext4_orphan_del(handle
, inode
);
1991 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
1992 ext4_mark_inode_dirty(handle
, inode
);
1994 ext4_handle_sync(handle
);
1996 ext4_journal_stop(handle
);
2000 int ext4_convert_inline_data(struct inode
*inode
)
2002 int error
, needed_blocks
, no_expand
;
2004 struct ext4_iloc iloc
;
2006 if (!ext4_has_inline_data(inode
)) {
2007 ext4_clear_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
2011 needed_blocks
= ext4_writepage_trans_blocks(inode
);
2014 error
= ext4_get_inode_loc(inode
, &iloc
);
2018 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
2019 if (IS_ERR(handle
)) {
2020 error
= PTR_ERR(handle
);
2024 ext4_write_lock_xattr(inode
, &no_expand
);
2025 if (ext4_has_inline_data(inode
))
2026 error
= ext4_convert_inline_data_nolock(handle
, inode
, &iloc
);
2027 ext4_write_unlock_xattr(inode
, &no_expand
);
2028 ext4_journal_stop(handle
);