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 error
= ext4_mark_iloc_dirty(handle
, inode
, &is
.iloc
);
310 static int ext4_update_inline_data(handle_t
*handle
, struct inode
*inode
,
315 struct ext4_xattr_ibody_find is
= {
316 .s
= { .not_found
= -ENODATA
, },
318 struct ext4_xattr_info i
= {
319 .name_index
= EXT4_XATTR_INDEX_SYSTEM
,
320 .name
= EXT4_XATTR_SYSTEM_DATA
,
323 /* If the old space is ok, write the data directly. */
324 if (len
<= EXT4_I(inode
)->i_inline_size
)
327 error
= ext4_get_inode_loc(inode
, &is
.iloc
);
331 error
= ext4_xattr_ibody_find(inode
, &i
, &is
);
335 BUG_ON(is
.s
.not_found
);
337 len
-= EXT4_MIN_INLINE_DATA_SIZE
;
338 value
= kzalloc(len
, GFP_NOFS
);
342 error
= ext4_xattr_ibody_get(inode
, i
.name_index
, i
.name
,
344 if (error
== -ENODATA
)
347 BUFFER_TRACE(is
.iloc
.bh
, "get_write_access");
348 error
= ext4_journal_get_write_access(handle
, is
.iloc
.bh
);
352 /* Update the xttr entry. */
356 error
= ext4_xattr_ibody_inline_set(handle
, inode
, &i
, &is
);
360 EXT4_I(inode
)->i_inline_off
= (u16
)((void *)is
.s
.here
-
361 (void *)ext4_raw_inode(&is
.iloc
));
362 EXT4_I(inode
)->i_inline_size
= EXT4_MIN_INLINE_DATA_SIZE
+
363 le32_to_cpu(is
.s
.here
->e_value_size
);
364 ext4_set_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
366 error
= ext4_mark_iloc_dirty(handle
, inode
, &is
.iloc
);
374 static int ext4_prepare_inline_data(handle_t
*handle
, struct inode
*inode
,
378 struct ext4_inode_info
*ei
= EXT4_I(inode
);
380 if (!ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
))
383 size
= ext4_get_max_inline_size(inode
);
387 down_write(&EXT4_I(inode
)->xattr_sem
);
389 if (ei
->i_inline_off
)
390 ret
= ext4_update_inline_data(handle
, inode
, len
);
392 ret
= ext4_create_inline_data(handle
, inode
, len
);
394 up_write(&EXT4_I(inode
)->xattr_sem
);
399 static int ext4_destroy_inline_data_nolock(handle_t
*handle
,
402 struct ext4_inode_info
*ei
= EXT4_I(inode
);
403 struct ext4_xattr_ibody_find is
= {
404 .s
= { .not_found
= 0, },
406 struct ext4_xattr_info i
= {
407 .name_index
= EXT4_XATTR_INDEX_SYSTEM
,
408 .name
= EXT4_XATTR_SYSTEM_DATA
,
414 if (!ei
->i_inline_off
)
417 error
= ext4_get_inode_loc(inode
, &is
.iloc
);
421 error
= ext4_xattr_ibody_find(inode
, &i
, &is
);
425 BUFFER_TRACE(is
.iloc
.bh
, "get_write_access");
426 error
= ext4_journal_get_write_access(handle
, is
.iloc
.bh
);
430 error
= ext4_xattr_ibody_inline_set(handle
, inode
, &i
, &is
);
434 memset((void *)ext4_raw_inode(&is
.iloc
)->i_block
,
435 0, EXT4_MIN_INLINE_DATA_SIZE
);
437 if (ext4_has_feature_extents(inode
->i_sb
)) {
438 if (S_ISDIR(inode
->i_mode
) ||
439 S_ISREG(inode
->i_mode
) || S_ISLNK(inode
->i_mode
)) {
440 ext4_set_inode_flag(inode
, EXT4_INODE_EXTENTS
);
441 ext4_ext_tree_init(handle
, inode
);
444 ext4_clear_inode_flag(inode
, EXT4_INODE_INLINE_DATA
);
447 error
= ext4_mark_iloc_dirty(handle
, inode
, &is
.iloc
);
449 EXT4_I(inode
)->i_inline_off
= 0;
450 EXT4_I(inode
)->i_inline_size
= 0;
451 ext4_clear_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
454 if (error
== -ENODATA
)
459 static int ext4_read_inline_page(struct inode
*inode
, struct page
*page
)
464 struct ext4_iloc iloc
;
466 BUG_ON(!PageLocked(page
));
467 BUG_ON(!ext4_has_inline_data(inode
));
470 if (!EXT4_I(inode
)->i_inline_off
) {
471 ext4_warning(inode
->i_sb
, "inode %lu doesn't have inline data.",
476 ret
= ext4_get_inode_loc(inode
, &iloc
);
480 len
= min_t(size_t, ext4_get_inline_size(inode
), i_size_read(inode
));
481 kaddr
= kmap_atomic(page
);
482 ret
= ext4_read_inline_data(inode
, kaddr
, len
, &iloc
);
483 flush_dcache_page(page
);
484 kunmap_atomic(kaddr
);
485 zero_user_segment(page
, len
, PAGE_CACHE_SIZE
);
486 SetPageUptodate(page
);
493 int ext4_readpage_inline(struct inode
*inode
, struct page
*page
)
497 down_read(&EXT4_I(inode
)->xattr_sem
);
498 if (!ext4_has_inline_data(inode
)) {
499 up_read(&EXT4_I(inode
)->xattr_sem
);
504 * Current inline data can only exist in the 1st page,
505 * So for all the other pages, just set them uptodate.
508 ret
= ext4_read_inline_page(inode
, page
);
509 else if (!PageUptodate(page
)) {
510 zero_user_segment(page
, 0, PAGE_CACHE_SIZE
);
511 SetPageUptodate(page
);
514 up_read(&EXT4_I(inode
)->xattr_sem
);
517 return ret
>= 0 ? 0 : ret
;
520 static int ext4_convert_inline_data_to_extent(struct address_space
*mapping
,
524 int ret
, needed_blocks
;
525 handle_t
*handle
= NULL
;
526 int retries
= 0, sem_held
= 0;
527 struct page
*page
= NULL
;
529 struct ext4_iloc iloc
;
531 if (!ext4_has_inline_data(inode
)) {
533 * clear the flag so that no new write
534 * will trap here again.
536 ext4_clear_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
540 needed_blocks
= ext4_writepage_trans_blocks(inode
);
542 ret
= ext4_get_inode_loc(inode
, &iloc
);
547 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
548 if (IS_ERR(handle
)) {
549 ret
= PTR_ERR(handle
);
554 /* We cannot recurse into the filesystem as the transaction is already
556 flags
|= AOP_FLAG_NOFS
;
558 page
= grab_cache_page_write_begin(mapping
, 0, flags
);
564 down_write(&EXT4_I(inode
)->xattr_sem
);
566 /* If some one has already done this for us, just exit. */
567 if (!ext4_has_inline_data(inode
)) {
573 to
= ext4_get_inline_size(inode
);
574 if (!PageUptodate(page
)) {
575 ret
= ext4_read_inline_page(inode
, page
);
580 ret
= ext4_destroy_inline_data_nolock(handle
, inode
);
584 if (ext4_should_dioread_nolock(inode
))
585 ret
= __block_write_begin(page
, from
, to
, ext4_get_block_write
);
587 ret
= __block_write_begin(page
, from
, to
, ext4_get_block
);
589 if (!ret
&& ext4_should_journal_data(inode
)) {
590 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
592 do_journal_get_write_access
);
597 page_cache_release(page
);
599 ext4_orphan_add(handle
, inode
);
600 up_write(&EXT4_I(inode
)->xattr_sem
);
602 ext4_journal_stop(handle
);
604 ext4_truncate_failed_write(inode
);
606 * If truncate failed early the inode might
607 * still be on the orphan list; we need to
608 * make sure the inode is removed from the
609 * orphan list in that case.
612 ext4_orphan_del(NULL
, inode
);
615 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
619 block_commit_write(page
, from
, to
);
623 page_cache_release(page
);
626 up_write(&EXT4_I(inode
)->xattr_sem
);
628 ext4_journal_stop(handle
);
634 * Try to write data in the inode.
635 * If the inode has inline data, check whether the new write can be
636 * in the inode also. If not, create the page the handle, move the data
637 * to the page make it update and let the later codes create extent for it.
639 int ext4_try_to_write_inline_data(struct address_space
*mapping
,
641 loff_t pos
, unsigned len
,
648 struct ext4_iloc iloc
;
650 if (pos
+ len
> ext4_get_max_inline_size(inode
))
653 ret
= ext4_get_inode_loc(inode
, &iloc
);
658 * The possible write could happen in the inode,
659 * so try to reserve the space in inode first.
661 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
662 if (IS_ERR(handle
)) {
663 ret
= PTR_ERR(handle
);
668 ret
= ext4_prepare_inline_data(handle
, inode
, pos
+ len
);
669 if (ret
&& ret
!= -ENOSPC
)
672 /* We don't have space in inline inode, so convert it to extent. */
673 if (ret
== -ENOSPC
) {
674 ext4_journal_stop(handle
);
679 flags
|= AOP_FLAG_NOFS
;
681 page
= grab_cache_page_write_begin(mapping
, 0, flags
);
688 down_read(&EXT4_I(inode
)->xattr_sem
);
689 if (!ext4_has_inline_data(inode
)) {
692 page_cache_release(page
);
696 if (!PageUptodate(page
)) {
697 ret
= ext4_read_inline_page(inode
, page
);
705 up_read(&EXT4_I(inode
)->xattr_sem
);
708 ext4_journal_stop(handle
);
712 return ext4_convert_inline_data_to_extent(mapping
,
716 int ext4_write_inline_data_end(struct inode
*inode
, loff_t pos
, unsigned len
,
717 unsigned copied
, struct page
*page
)
721 struct ext4_iloc iloc
;
723 if (unlikely(copied
< len
)) {
724 if (!PageUptodate(page
)) {
730 ret
= ext4_get_inode_loc(inode
, &iloc
);
732 ext4_std_error(inode
->i_sb
, ret
);
737 down_write(&EXT4_I(inode
)->xattr_sem
);
738 BUG_ON(!ext4_has_inline_data(inode
));
740 kaddr
= kmap_atomic(page
);
741 ext4_write_inline_data(inode
, &iloc
, kaddr
, pos
, len
);
742 kunmap_atomic(kaddr
);
743 SetPageUptodate(page
);
744 /* clear page dirty so that writepages wouldn't work for us. */
745 ClearPageDirty(page
);
747 up_write(&EXT4_I(inode
)->xattr_sem
);
754 ext4_journalled_write_inline_data(struct inode
*inode
,
760 struct ext4_iloc iloc
;
762 ret
= ext4_get_inode_loc(inode
, &iloc
);
764 ext4_std_error(inode
->i_sb
, ret
);
768 down_write(&EXT4_I(inode
)->xattr_sem
);
769 kaddr
= kmap_atomic(page
);
770 ext4_write_inline_data(inode
, &iloc
, kaddr
, 0, len
);
771 kunmap_atomic(kaddr
);
772 up_write(&EXT4_I(inode
)->xattr_sem
);
778 * Try to make the page cache and handle ready for the inline data case.
779 * We can call this function in 2 cases:
780 * 1. The inode is created and the first write exceeds inline size. We can
781 * clear the inode state safely.
782 * 2. The inode has inline data, then we need to read the data, make it
783 * update and dirty so that ext4_da_writepages can handle it. We don't
784 * need to start the journal since the file's metatdata isn't changed now.
786 static int ext4_da_convert_inline_data_to_extent(struct address_space
*mapping
,
791 int ret
= 0, inline_size
;
794 page
= grab_cache_page_write_begin(mapping
, 0, flags
);
798 down_read(&EXT4_I(inode
)->xattr_sem
);
799 if (!ext4_has_inline_data(inode
)) {
800 ext4_clear_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
804 inline_size
= ext4_get_inline_size(inode
);
806 if (!PageUptodate(page
)) {
807 ret
= ext4_read_inline_page(inode
, page
);
812 ret
= __block_write_begin(page
, 0, inline_size
,
813 ext4_da_get_block_prep
);
815 up_read(&EXT4_I(inode
)->xattr_sem
);
817 page_cache_release(page
);
818 ext4_truncate_failed_write(inode
);
823 SetPageUptodate(page
);
824 ext4_clear_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
825 *fsdata
= (void *)CONVERT_INLINE_DATA
;
828 up_read(&EXT4_I(inode
)->xattr_sem
);
831 page_cache_release(page
);
837 * Prepare the write for the inline data.
838 * If the the data can be written into the inode, we just read
839 * the page and make it uptodate, and start the journal.
840 * Otherwise read the page, makes it dirty so that it can be
841 * handle in writepages(the i_disksize update is left to the
842 * normal ext4_da_write_end).
844 int ext4_da_write_inline_data_begin(struct address_space
*mapping
,
846 loff_t pos
, unsigned len
,
851 int ret
, inline_size
;
854 struct ext4_iloc iloc
;
857 ret
= ext4_get_inode_loc(inode
, &iloc
);
862 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
863 if (IS_ERR(handle
)) {
864 ret
= PTR_ERR(handle
);
868 inline_size
= ext4_get_max_inline_size(inode
);
871 if (inline_size
>= pos
+ len
) {
872 ret
= ext4_prepare_inline_data(handle
, inode
, pos
+ len
);
873 if (ret
&& ret
!= -ENOSPC
)
878 * We cannot recurse into the filesystem as the transaction
879 * is already started.
881 flags
|= AOP_FLAG_NOFS
;
883 if (ret
== -ENOSPC
) {
884 ret
= ext4_da_convert_inline_data_to_extent(mapping
,
888 ext4_journal_stop(handle
);
889 if (ret
== -ENOSPC
&&
890 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
896 page
= grab_cache_page_write_begin(mapping
, 0, flags
);
902 down_read(&EXT4_I(inode
)->xattr_sem
);
903 if (!ext4_has_inline_data(inode
)) {
905 goto out_release_page
;
908 if (!PageUptodate(page
)) {
909 ret
= ext4_read_inline_page(inode
, page
);
911 goto out_release_page
;
914 up_read(&EXT4_I(inode
)->xattr_sem
);
919 up_read(&EXT4_I(inode
)->xattr_sem
);
921 page_cache_release(page
);
923 ext4_journal_stop(handle
);
929 int ext4_da_write_inline_data_end(struct inode
*inode
, loff_t pos
,
930 unsigned len
, unsigned copied
,
933 int i_size_changed
= 0;
935 copied
= ext4_write_inline_data_end(inode
, pos
, len
, copied
, page
);
938 * No need to use i_size_read() here, the i_size
939 * cannot change under us because we hold i_mutex.
941 * But it's important to update i_size while still holding page lock:
942 * page writeout could otherwise come in and zero beyond i_size.
944 if (pos
+copied
> inode
->i_size
) {
945 i_size_write(inode
, pos
+copied
);
949 page_cache_release(page
);
952 * Don't mark the inode dirty under page lock. First, it unnecessarily
953 * makes the holding time of page lock longer. Second, it forces lock
954 * ordering of page lock and transaction start for journaling
958 mark_inode_dirty(inode
);
963 #ifdef INLINE_DIR_DEBUG
964 void ext4_show_inline_dir(struct inode
*dir
, struct buffer_head
*bh
,
965 void *inline_start
, int inline_size
)
968 unsigned short de_len
;
969 struct ext4_dir_entry_2
*de
= inline_start
;
970 void *dlimit
= inline_start
+ inline_size
;
972 trace_printk("inode %lu\n", dir
->i_ino
);
974 while ((void *)de
< dlimit
) {
975 de_len
= ext4_rec_len_from_disk(de
->rec_len
, inline_size
);
976 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
977 offset
, de_len
, de
->name_len
, de
->name
,
978 de
->name_len
, le32_to_cpu(de
->inode
));
979 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
980 inline_start
, inline_size
, offset
))
984 de
= (struct ext4_dir_entry_2
*) ((char *) de
+ de_len
);
988 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
992 * Add a new entry into a inline dir.
993 * It will return -ENOSPC if no space is available, and -EIO
994 * and -EEXIST if directory entry already exists.
996 static int ext4_add_dirent_to_inline(handle_t
*handle
,
997 struct ext4_filename
*fname
,
1000 struct ext4_iloc
*iloc
,
1001 void *inline_start
, int inline_size
)
1004 struct ext4_dir_entry_2
*de
;
1006 err
= ext4_find_dest_de(dir
, inode
, iloc
->bh
, inline_start
,
1007 inline_size
, fname
, &de
);
1011 BUFFER_TRACE(iloc
->bh
, "get_write_access");
1012 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
1015 ext4_insert_dentry(dir
, inode
, de
, inline_size
, fname
);
1017 ext4_show_inline_dir(dir
, iloc
->bh
, inline_start
, inline_size
);
1020 * XXX shouldn't update any times until successful
1021 * completion of syscall, but too many callers depend
1024 * XXX similarly, too many callers depend on
1025 * ext4_new_inode() setting the times, but error
1026 * recovery deletes the inode, so the worst that can
1027 * happen is that the times are slightly out of date
1028 * and/or different from the directory change time.
1030 dir
->i_mtime
= dir
->i_ctime
= ext4_current_time(dir
);
1031 ext4_update_dx_flag(dir
);
1033 ext4_mark_inode_dirty(handle
, dir
);
1037 static void *ext4_get_inline_xattr_pos(struct inode
*inode
,
1038 struct ext4_iloc
*iloc
)
1040 struct ext4_xattr_entry
*entry
;
1041 struct ext4_xattr_ibody_header
*header
;
1043 BUG_ON(!EXT4_I(inode
)->i_inline_off
);
1045 header
= IHDR(inode
, ext4_raw_inode(iloc
));
1046 entry
= (struct ext4_xattr_entry
*)((void *)ext4_raw_inode(iloc
) +
1047 EXT4_I(inode
)->i_inline_off
);
1049 return (void *)IFIRST(header
) + le16_to_cpu(entry
->e_value_offs
);
1052 /* Set the final de to cover the whole block. */
1053 static void ext4_update_final_de(void *de_buf
, int old_size
, int new_size
)
1055 struct ext4_dir_entry_2
*de
, *prev_de
;
1059 de
= (struct ext4_dir_entry_2
*)de_buf
;
1061 limit
= de_buf
+ old_size
;
1064 de_len
= ext4_rec_len_from_disk(de
->rec_len
, old_size
);
1066 de
= (struct ext4_dir_entry_2
*)de_buf
;
1067 } while (de_buf
< limit
);
1069 prev_de
->rec_len
= ext4_rec_len_to_disk(de_len
+ new_size
-
1070 old_size
, new_size
);
1072 /* this is just created, so create an empty entry. */
1074 de
->rec_len
= ext4_rec_len_to_disk(new_size
, new_size
);
1078 static int ext4_update_inline_dir(handle_t
*handle
, struct inode
*dir
,
1079 struct ext4_iloc
*iloc
)
1082 int old_size
= EXT4_I(dir
)->i_inline_size
- EXT4_MIN_INLINE_DATA_SIZE
;
1083 int new_size
= get_max_inline_xattr_value_size(dir
, iloc
);
1085 if (new_size
- old_size
<= EXT4_DIR_REC_LEN(1))
1088 ret
= ext4_update_inline_data(handle
, dir
,
1089 new_size
+ EXT4_MIN_INLINE_DATA_SIZE
);
1093 ext4_update_final_de(ext4_get_inline_xattr_pos(dir
, iloc
), old_size
,
1094 EXT4_I(dir
)->i_inline_size
-
1095 EXT4_MIN_INLINE_DATA_SIZE
);
1096 dir
->i_size
= EXT4_I(dir
)->i_disksize
= EXT4_I(dir
)->i_inline_size
;
1100 static void ext4_restore_inline_data(handle_t
*handle
, struct inode
*inode
,
1101 struct ext4_iloc
*iloc
,
1102 void *buf
, int inline_size
)
1104 ext4_create_inline_data(handle
, inode
, inline_size
);
1105 ext4_write_inline_data(inode
, iloc
, buf
, 0, inline_size
);
1106 ext4_set_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
1109 static int ext4_finish_convert_inline_dir(handle_t
*handle
,
1110 struct inode
*inode
,
1111 struct buffer_head
*dir_block
,
1115 int err
, csum_size
= 0, header_size
= 0;
1116 struct ext4_dir_entry_2
*de
;
1117 struct ext4_dir_entry_tail
*t
;
1118 void *target
= dir_block
->b_data
;
1121 * First create "." and ".." and then copy the dir information
1122 * back to the block.
1124 de
= (struct ext4_dir_entry_2
*)target
;
1125 de
= ext4_init_dot_dotdot(inode
, de
,
1126 inode
->i_sb
->s_blocksize
, csum_size
,
1127 le32_to_cpu(((struct ext4_dir_entry_2
*)buf
)->inode
), 1);
1128 header_size
= (void *)de
- target
;
1130 memcpy((void *)de
, buf
+ EXT4_INLINE_DOTDOT_SIZE
,
1131 inline_size
- EXT4_INLINE_DOTDOT_SIZE
);
1133 if (ext4_has_metadata_csum(inode
->i_sb
))
1134 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1136 inode
->i_size
= inode
->i_sb
->s_blocksize
;
1137 i_size_write(inode
, inode
->i_sb
->s_blocksize
);
1138 EXT4_I(inode
)->i_disksize
= inode
->i_sb
->s_blocksize
;
1139 ext4_update_final_de(dir_block
->b_data
,
1140 inline_size
- EXT4_INLINE_DOTDOT_SIZE
+ header_size
,
1141 inode
->i_sb
->s_blocksize
- csum_size
);
1144 t
= EXT4_DIRENT_TAIL(dir_block
->b_data
,
1145 inode
->i_sb
->s_blocksize
);
1146 initialize_dirent_tail(t
, inode
->i_sb
->s_blocksize
);
1148 set_buffer_uptodate(dir_block
);
1149 err
= ext4_handle_dirty_dirent_node(handle
, inode
, dir_block
);
1152 set_buffer_verified(dir_block
);
1157 static int ext4_convert_inline_data_nolock(handle_t
*handle
,
1158 struct inode
*inode
,
1159 struct ext4_iloc
*iloc
)
1163 struct buffer_head
*data_bh
= NULL
;
1164 struct ext4_map_blocks map
;
1167 inline_size
= ext4_get_inline_size(inode
);
1168 buf
= kmalloc(inline_size
, GFP_NOFS
);
1174 error
= ext4_read_inline_data(inode
, buf
, inline_size
, iloc
);
1179 * Make sure the inline directory entries pass checks before we try to
1180 * convert them, so that we avoid touching stuff that needs fsck.
1182 if (S_ISDIR(inode
->i_mode
)) {
1183 error
= ext4_check_all_de(inode
, iloc
->bh
,
1184 buf
+ EXT4_INLINE_DOTDOT_SIZE
,
1185 inline_size
- EXT4_INLINE_DOTDOT_SIZE
);
1190 error
= ext4_destroy_inline_data_nolock(handle
, inode
);
1197 error
= ext4_map_blocks(handle
, inode
, &map
, EXT4_GET_BLOCKS_CREATE
);
1200 if (!(map
.m_flags
& EXT4_MAP_MAPPED
)) {
1205 data_bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
1211 lock_buffer(data_bh
);
1212 error
= ext4_journal_get_create_access(handle
, data_bh
);
1214 unlock_buffer(data_bh
);
1218 memset(data_bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
1220 if (!S_ISDIR(inode
->i_mode
)) {
1221 memcpy(data_bh
->b_data
, buf
, inline_size
);
1222 set_buffer_uptodate(data_bh
);
1223 error
= ext4_handle_dirty_metadata(handle
,
1226 error
= ext4_finish_convert_inline_dir(handle
, inode
, data_bh
,
1230 unlock_buffer(data_bh
);
1233 ext4_restore_inline_data(handle
, inode
, iloc
, buf
, inline_size
);
1242 * Try to add the new entry to the inline data.
1243 * If succeeds, return 0. If not, extended the inline dir and copied data to
1244 * the new created block.
1246 int ext4_try_add_inline_entry(handle_t
*handle
, struct ext4_filename
*fname
,
1247 struct inode
*dir
, struct inode
*inode
)
1249 int ret
, inline_size
;
1251 struct ext4_iloc iloc
;
1253 ret
= ext4_get_inode_loc(dir
, &iloc
);
1257 down_write(&EXT4_I(dir
)->xattr_sem
);
1258 if (!ext4_has_inline_data(dir
))
1261 inline_start
= (void *)ext4_raw_inode(&iloc
)->i_block
+
1262 EXT4_INLINE_DOTDOT_SIZE
;
1263 inline_size
= EXT4_MIN_INLINE_DATA_SIZE
- EXT4_INLINE_DOTDOT_SIZE
;
1265 ret
= ext4_add_dirent_to_inline(handle
, fname
, dir
, inode
, &iloc
,
1266 inline_start
, inline_size
);
1270 /* check whether it can be inserted to inline xattr space. */
1271 inline_size
= EXT4_I(dir
)->i_inline_size
-
1272 EXT4_MIN_INLINE_DATA_SIZE
;
1274 /* Try to use the xattr space.*/
1275 ret
= ext4_update_inline_dir(handle
, dir
, &iloc
);
1276 if (ret
&& ret
!= -ENOSPC
)
1279 inline_size
= EXT4_I(dir
)->i_inline_size
-
1280 EXT4_MIN_INLINE_DATA_SIZE
;
1284 inline_start
= ext4_get_inline_xattr_pos(dir
, &iloc
);
1286 ret
= ext4_add_dirent_to_inline(handle
, fname
, dir
,
1287 inode
, &iloc
, inline_start
,
1295 * The inline space is filled up, so create a new block for it.
1296 * As the extent tree will be created, we have to save the inline
1299 ret
= ext4_convert_inline_data_nolock(handle
, dir
, &iloc
);
1302 ext4_mark_inode_dirty(handle
, dir
);
1303 up_write(&EXT4_I(dir
)->xattr_sem
);
1309 * This function fills a red-black tree with information from an
1310 * inlined dir. It returns the number directory entries loaded
1311 * into the tree. If there is an error it is returned in err.
1313 int htree_inlinedir_to_tree(struct file
*dir_file
,
1314 struct inode
*dir
, ext4_lblk_t block
,
1315 struct dx_hash_info
*hinfo
,
1316 __u32 start_hash
, __u32 start_minor_hash
,
1317 int *has_inline_data
)
1319 int err
= 0, count
= 0;
1320 unsigned int parent_ino
;
1322 struct ext4_dir_entry_2
*de
;
1323 struct inode
*inode
= file_inode(dir_file
);
1324 int ret
, inline_size
= 0;
1325 struct ext4_iloc iloc
;
1326 void *dir_buf
= NULL
;
1327 struct ext4_dir_entry_2 fake
;
1328 struct ext4_str tmp_str
;
1330 ret
= ext4_get_inode_loc(inode
, &iloc
);
1334 down_read(&EXT4_I(inode
)->xattr_sem
);
1335 if (!ext4_has_inline_data(inode
)) {
1336 up_read(&EXT4_I(inode
)->xattr_sem
);
1337 *has_inline_data
= 0;
1341 inline_size
= ext4_get_inline_size(inode
);
1342 dir_buf
= kmalloc(inline_size
, GFP_NOFS
);
1345 up_read(&EXT4_I(inode
)->xattr_sem
);
1349 ret
= ext4_read_inline_data(inode
, dir_buf
, inline_size
, &iloc
);
1350 up_read(&EXT4_I(inode
)->xattr_sem
);
1355 parent_ino
= le32_to_cpu(((struct ext4_dir_entry_2
*)dir_buf
)->inode
);
1356 while (pos
< inline_size
) {
1358 * As inlined dir doesn't store any information about '.' and
1359 * only the inode number of '..' is stored, we have to handle
1363 fake
.inode
= cpu_to_le32(inode
->i_ino
);
1365 strcpy(fake
.name
, ".");
1366 fake
.rec_len
= ext4_rec_len_to_disk(
1367 EXT4_DIR_REC_LEN(fake
.name_len
),
1369 ext4_set_de_type(inode
->i_sb
, &fake
, S_IFDIR
);
1371 pos
= EXT4_INLINE_DOTDOT_OFFSET
;
1372 } else if (pos
== EXT4_INLINE_DOTDOT_OFFSET
) {
1373 fake
.inode
= cpu_to_le32(parent_ino
);
1375 strcpy(fake
.name
, "..");
1376 fake
.rec_len
= ext4_rec_len_to_disk(
1377 EXT4_DIR_REC_LEN(fake
.name_len
),
1379 ext4_set_de_type(inode
->i_sb
, &fake
, S_IFDIR
);
1381 pos
= EXT4_INLINE_DOTDOT_SIZE
;
1383 de
= (struct ext4_dir_entry_2
*)(dir_buf
+ pos
);
1384 pos
+= ext4_rec_len_from_disk(de
->rec_len
, inline_size
);
1385 if (ext4_check_dir_entry(inode
, dir_file
, de
,
1387 inline_size
, pos
)) {
1393 ext4fs_dirhash(de
->name
, de
->name_len
, hinfo
);
1394 if ((hinfo
->hash
< start_hash
) ||
1395 ((hinfo
->hash
== start_hash
) &&
1396 (hinfo
->minor_hash
< start_minor_hash
)))
1400 tmp_str
.name
= de
->name
;
1401 tmp_str
.len
= de
->name_len
;
1402 err
= ext4_htree_store_dirent(dir_file
, hinfo
->hash
,
1403 hinfo
->minor_hash
, de
, &tmp_str
);
1418 * So this function is called when the volume is mkfsed with
1419 * dir_index disabled. In order to keep f_pos persistent
1420 * after we convert from an inlined dir to a blocked based,
1421 * we just pretend that we are a normal dir and return the
1422 * offset as if '.' and '..' really take place.
1425 int ext4_read_inline_dir(struct file
*file
,
1426 struct dir_context
*ctx
,
1427 int *has_inline_data
)
1429 unsigned int offset
, parent_ino
;
1431 struct ext4_dir_entry_2
*de
;
1432 struct super_block
*sb
;
1433 struct inode
*inode
= file_inode(file
);
1434 int ret
, inline_size
= 0;
1435 struct ext4_iloc iloc
;
1436 void *dir_buf
= NULL
;
1437 int dotdot_offset
, dotdot_size
, extra_offset
, extra_size
;
1439 ret
= ext4_get_inode_loc(inode
, &iloc
);
1443 down_read(&EXT4_I(inode
)->xattr_sem
);
1444 if (!ext4_has_inline_data(inode
)) {
1445 up_read(&EXT4_I(inode
)->xattr_sem
);
1446 *has_inline_data
= 0;
1450 inline_size
= ext4_get_inline_size(inode
);
1451 dir_buf
= kmalloc(inline_size
, GFP_NOFS
);
1454 up_read(&EXT4_I(inode
)->xattr_sem
);
1458 ret
= ext4_read_inline_data(inode
, dir_buf
, inline_size
, &iloc
);
1459 up_read(&EXT4_I(inode
)->xattr_sem
);
1465 parent_ino
= le32_to_cpu(((struct ext4_dir_entry_2
*)dir_buf
)->inode
);
1469 * dotdot_offset and dotdot_size is the real offset and
1470 * size for ".." and "." if the dir is block based while
1471 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1472 * So we will use extra_offset and extra_size to indicate them
1473 * during the inline dir iteration.
1475 dotdot_offset
= EXT4_DIR_REC_LEN(1);
1476 dotdot_size
= dotdot_offset
+ EXT4_DIR_REC_LEN(2);
1477 extra_offset
= dotdot_size
- EXT4_INLINE_DOTDOT_SIZE
;
1478 extra_size
= extra_offset
+ inline_size
;
1481 * If the version has changed since the last call to
1482 * readdir(2), then we might be pointing to an invalid
1483 * dirent right now. Scan from the start of the inline
1486 if (file
->f_version
!= inode
->i_version
) {
1487 for (i
= 0; i
< extra_size
&& i
< offset
;) {
1489 * "." is with offset 0 and
1490 * ".." is dotdot_offset.
1495 } else if (i
== dotdot_offset
) {
1499 /* for other entry, the real offset in
1500 * the buf has to be tuned accordingly.
1502 de
= (struct ext4_dir_entry_2
*)
1503 (dir_buf
+ i
- extra_offset
);
1504 /* It's too expensive to do a full
1505 * dirent test each time round this
1506 * loop, but we do have to test at
1507 * least that it is non-zero. A
1508 * failure will be detected in the
1509 * dirent test below. */
1510 if (ext4_rec_len_from_disk(de
->rec_len
, extra_size
)
1511 < EXT4_DIR_REC_LEN(1))
1513 i
+= ext4_rec_len_from_disk(de
->rec_len
,
1518 file
->f_version
= inode
->i_version
;
1521 while (ctx
->pos
< extra_size
) {
1522 if (ctx
->pos
== 0) {
1523 if (!dir_emit(ctx
, ".", 1, inode
->i_ino
, DT_DIR
))
1525 ctx
->pos
= dotdot_offset
;
1529 if (ctx
->pos
== dotdot_offset
) {
1530 if (!dir_emit(ctx
, "..", 2, parent_ino
, DT_DIR
))
1532 ctx
->pos
= dotdot_size
;
1536 de
= (struct ext4_dir_entry_2
*)
1537 (dir_buf
+ ctx
->pos
- extra_offset
);
1538 if (ext4_check_dir_entry(inode
, file
, de
, iloc
.bh
, dir_buf
,
1539 extra_size
, ctx
->pos
))
1541 if (le32_to_cpu(de
->inode
)) {
1542 if (!dir_emit(ctx
, de
->name
, de
->name_len
,
1543 le32_to_cpu(de
->inode
),
1544 get_dtype(sb
, de
->file_type
)))
1547 ctx
->pos
+= ext4_rec_len_from_disk(de
->rec_len
, extra_size
);
1555 struct buffer_head
*ext4_get_first_inline_block(struct inode
*inode
,
1556 struct ext4_dir_entry_2
**parent_de
,
1559 struct ext4_iloc iloc
;
1561 *retval
= ext4_get_inode_loc(inode
, &iloc
);
1565 *parent_de
= (struct ext4_dir_entry_2
*)ext4_raw_inode(&iloc
)->i_block
;
1571 * Try to create the inline data for the new dir.
1572 * If it succeeds, return 0, otherwise return the error.
1573 * In case of ENOSPC, the caller should create the normal disk layout dir.
1575 int ext4_try_create_inline_dir(handle_t
*handle
, struct inode
*parent
,
1576 struct inode
*inode
)
1578 int ret
, inline_size
= EXT4_MIN_INLINE_DATA_SIZE
;
1579 struct ext4_iloc iloc
;
1580 struct ext4_dir_entry_2
*de
;
1582 ret
= ext4_get_inode_loc(inode
, &iloc
);
1586 ret
= ext4_prepare_inline_data(handle
, inode
, inline_size
);
1591 * For inline dir, we only save the inode information for the ".."
1592 * and create a fake dentry to cover the left space.
1594 de
= (struct ext4_dir_entry_2
*)ext4_raw_inode(&iloc
)->i_block
;
1595 de
->inode
= cpu_to_le32(parent
->i_ino
);
1596 de
= (struct ext4_dir_entry_2
*)((void *)de
+ EXT4_INLINE_DOTDOT_SIZE
);
1598 de
->rec_len
= ext4_rec_len_to_disk(
1599 inline_size
- EXT4_INLINE_DOTDOT_SIZE
,
1601 set_nlink(inode
, 2);
1602 inode
->i_size
= EXT4_I(inode
)->i_disksize
= inline_size
;
1608 struct buffer_head
*ext4_find_inline_entry(struct inode
*dir
,
1609 struct ext4_filename
*fname
,
1610 const struct qstr
*d_name
,
1611 struct ext4_dir_entry_2
**res_dir
,
1612 int *has_inline_data
)
1615 struct ext4_iloc iloc
;
1619 if (ext4_get_inode_loc(dir
, &iloc
))
1622 down_read(&EXT4_I(dir
)->xattr_sem
);
1623 if (!ext4_has_inline_data(dir
)) {
1624 *has_inline_data
= 0;
1628 inline_start
= (void *)ext4_raw_inode(&iloc
)->i_block
+
1629 EXT4_INLINE_DOTDOT_SIZE
;
1630 inline_size
= EXT4_MIN_INLINE_DATA_SIZE
- EXT4_INLINE_DOTDOT_SIZE
;
1631 ret
= ext4_search_dir(iloc
.bh
, inline_start
, inline_size
,
1632 dir
, fname
, d_name
, 0, res_dir
);
1638 if (ext4_get_inline_size(dir
) == EXT4_MIN_INLINE_DATA_SIZE
)
1641 inline_start
= ext4_get_inline_xattr_pos(dir
, &iloc
);
1642 inline_size
= ext4_get_inline_size(dir
) - EXT4_MIN_INLINE_DATA_SIZE
;
1644 ret
= ext4_search_dir(iloc
.bh
, inline_start
, inline_size
,
1645 dir
, fname
, d_name
, 0, res_dir
);
1653 up_read(&EXT4_I(dir
)->xattr_sem
);
1657 int ext4_delete_inline_entry(handle_t
*handle
,
1659 struct ext4_dir_entry_2
*de_del
,
1660 struct buffer_head
*bh
,
1661 int *has_inline_data
)
1663 int err
, inline_size
;
1664 struct ext4_iloc iloc
;
1667 err
= ext4_get_inode_loc(dir
, &iloc
);
1671 down_write(&EXT4_I(dir
)->xattr_sem
);
1672 if (!ext4_has_inline_data(dir
)) {
1673 *has_inline_data
= 0;
1677 if ((void *)de_del
- ((void *)ext4_raw_inode(&iloc
)->i_block
) <
1678 EXT4_MIN_INLINE_DATA_SIZE
) {
1679 inline_start
= (void *)ext4_raw_inode(&iloc
)->i_block
+
1680 EXT4_INLINE_DOTDOT_SIZE
;
1681 inline_size
= EXT4_MIN_INLINE_DATA_SIZE
-
1682 EXT4_INLINE_DOTDOT_SIZE
;
1684 inline_start
= ext4_get_inline_xattr_pos(dir
, &iloc
);
1685 inline_size
= ext4_get_inline_size(dir
) -
1686 EXT4_MIN_INLINE_DATA_SIZE
;
1689 BUFFER_TRACE(bh
, "get_write_access");
1690 err
= ext4_journal_get_write_access(handle
, bh
);
1694 err
= ext4_generic_delete_entry(handle
, dir
, de_del
, bh
,
1695 inline_start
, inline_size
, 0);
1699 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1700 err
= ext4_mark_inode_dirty(handle
, dir
);
1704 ext4_show_inline_dir(dir
, iloc
.bh
, inline_start
, inline_size
);
1706 up_write(&EXT4_I(dir
)->xattr_sem
);
1709 ext4_std_error(dir
->i_sb
, err
);
1714 * Get the inline dentry at offset.
1716 static inline struct ext4_dir_entry_2
*
1717 ext4_get_inline_entry(struct inode
*inode
,
1718 struct ext4_iloc
*iloc
,
1719 unsigned int offset
,
1720 void **inline_start
,
1725 BUG_ON(offset
> ext4_get_inline_size(inode
));
1727 if (offset
< EXT4_MIN_INLINE_DATA_SIZE
) {
1728 inline_pos
= (void *)ext4_raw_inode(iloc
)->i_block
;
1729 *inline_size
= EXT4_MIN_INLINE_DATA_SIZE
;
1731 inline_pos
= ext4_get_inline_xattr_pos(inode
, iloc
);
1732 offset
-= EXT4_MIN_INLINE_DATA_SIZE
;
1733 *inline_size
= ext4_get_inline_size(inode
) -
1734 EXT4_MIN_INLINE_DATA_SIZE
;
1738 *inline_start
= inline_pos
;
1739 return (struct ext4_dir_entry_2
*)(inline_pos
+ offset
);
1742 int empty_inline_dir(struct inode
*dir
, int *has_inline_data
)
1744 int err
, inline_size
;
1745 struct ext4_iloc iloc
;
1747 unsigned int offset
;
1748 struct ext4_dir_entry_2
*de
;
1751 err
= ext4_get_inode_loc(dir
, &iloc
);
1753 EXT4_ERROR_INODE(dir
, "error %d getting inode %lu block",
1758 down_read(&EXT4_I(dir
)->xattr_sem
);
1759 if (!ext4_has_inline_data(dir
)) {
1760 *has_inline_data
= 0;
1764 de
= (struct ext4_dir_entry_2
*)ext4_raw_inode(&iloc
)->i_block
;
1765 if (!le32_to_cpu(de
->inode
)) {
1766 ext4_warning(dir
->i_sb
,
1767 "bad inline directory (dir #%lu) - no `..'",
1773 offset
= EXT4_INLINE_DOTDOT_SIZE
;
1774 while (offset
< dir
->i_size
) {
1775 de
= ext4_get_inline_entry(dir
, &iloc
, offset
,
1776 &inline_pos
, &inline_size
);
1777 if (ext4_check_dir_entry(dir
, NULL
, de
,
1778 iloc
.bh
, inline_pos
,
1779 inline_size
, offset
)) {
1780 ext4_warning(dir
->i_sb
,
1781 "bad inline directory (dir #%lu) - "
1782 "inode %u, rec_len %u, name_len %d"
1784 dir
->i_ino
, le32_to_cpu(de
->inode
),
1785 le16_to_cpu(de
->rec_len
), de
->name_len
,
1790 if (le32_to_cpu(de
->inode
)) {
1794 offset
+= ext4_rec_len_from_disk(de
->rec_len
, inline_size
);
1798 up_read(&EXT4_I(dir
)->xattr_sem
);
1803 int ext4_destroy_inline_data(handle_t
*handle
, struct inode
*inode
)
1807 down_write(&EXT4_I(inode
)->xattr_sem
);
1808 ret
= ext4_destroy_inline_data_nolock(handle
, inode
);
1809 up_write(&EXT4_I(inode
)->xattr_sem
);
1814 int ext4_inline_data_fiemap(struct inode
*inode
,
1815 struct fiemap_extent_info
*fieinfo
,
1816 int *has_inline
, __u64 start
, __u64 len
)
1820 __u32 flags
= FIEMAP_EXTENT_DATA_INLINE
| FIEMAP_EXTENT_NOT_ALIGNED
|
1823 struct ext4_iloc iloc
;
1825 down_read(&EXT4_I(inode
)->xattr_sem
);
1826 if (!ext4_has_inline_data(inode
)) {
1830 inline_len
= min_t(size_t, ext4_get_inline_size(inode
),
1831 i_size_read(inode
));
1832 if (start
>= inline_len
)
1834 if (start
+ len
< inline_len
)
1835 inline_len
= start
+ len
;
1836 inline_len
-= start
;
1838 error
= ext4_get_inode_loc(inode
, &iloc
);
1842 physical
= (__u64
)iloc
.bh
->b_blocknr
<< inode
->i_sb
->s_blocksize_bits
;
1843 physical
+= (char *)ext4_raw_inode(&iloc
) - iloc
.bh
->b_data
;
1844 physical
+= offsetof(struct ext4_inode
, i_block
);
1847 error
= fiemap_fill_next_extent(fieinfo
, start
, physical
,
1851 up_read(&EXT4_I(inode
)->xattr_sem
);
1852 return (error
< 0 ? error
: 0);
1856 * Called during xattr set, and if we can sparse space 'needed',
1857 * just create the extent tree evict the data to the outer block.
1859 * We use jbd2 instead of page cache to move data to the 1st block
1860 * so that the whole transaction can be committed as a whole and
1861 * the data isn't lost because of the delayed page cache write.
1863 int ext4_try_to_evict_inline_data(handle_t
*handle
,
1864 struct inode
*inode
,
1868 struct ext4_xattr_entry
*entry
;
1869 struct ext4_inode
*raw_inode
;
1870 struct ext4_iloc iloc
;
1872 error
= ext4_get_inode_loc(inode
, &iloc
);
1876 raw_inode
= ext4_raw_inode(&iloc
);
1877 entry
= (struct ext4_xattr_entry
*)((void *)raw_inode
+
1878 EXT4_I(inode
)->i_inline_off
);
1879 if (EXT4_XATTR_LEN(entry
->e_name_len
) +
1880 EXT4_XATTR_SIZE(le32_to_cpu(entry
->e_value_size
)) < needed
) {
1885 error
= ext4_convert_inline_data_nolock(handle
, inode
, &iloc
);
1891 void ext4_inline_data_truncate(struct inode
*inode
, int *has_inline
)
1894 int inline_size
, value_len
, needed_blocks
;
1897 struct ext4_xattr_ibody_find is
= {
1898 .s
= { .not_found
= -ENODATA
, },
1900 struct ext4_xattr_info i
= {
1901 .name_index
= EXT4_XATTR_INDEX_SYSTEM
,
1902 .name
= EXT4_XATTR_SYSTEM_DATA
,
1906 needed_blocks
= ext4_writepage_trans_blocks(inode
);
1907 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, needed_blocks
);
1911 down_write(&EXT4_I(inode
)->xattr_sem
);
1912 if (!ext4_has_inline_data(inode
)) {
1914 ext4_journal_stop(handle
);
1918 if (ext4_orphan_add(handle
, inode
))
1921 if (ext4_get_inode_loc(inode
, &is
.iloc
))
1924 down_write(&EXT4_I(inode
)->i_data_sem
);
1925 i_size
= inode
->i_size
;
1926 inline_size
= ext4_get_inline_size(inode
);
1927 EXT4_I(inode
)->i_disksize
= i_size
;
1929 if (i_size
< inline_size
) {
1930 /* Clear the content in the xattr space. */
1931 if (inline_size
> EXT4_MIN_INLINE_DATA_SIZE
) {
1932 if (ext4_xattr_ibody_find(inode
, &i
, &is
))
1935 BUG_ON(is
.s
.not_found
);
1937 value_len
= le32_to_cpu(is
.s
.here
->e_value_size
);
1938 value
= kmalloc(value_len
, GFP_NOFS
);
1942 if (ext4_xattr_ibody_get(inode
, i
.name_index
, i
.name
,
1947 i
.value_len
= i_size
> EXT4_MIN_INLINE_DATA_SIZE
?
1948 i_size
- EXT4_MIN_INLINE_DATA_SIZE
: 0;
1949 if (ext4_xattr_ibody_inline_set(handle
, inode
, &i
, &is
))
1953 /* Clear the content within i_blocks. */
1954 if (i_size
< EXT4_MIN_INLINE_DATA_SIZE
) {
1955 void *p
= (void *) ext4_raw_inode(&is
.iloc
)->i_block
;
1956 memset(p
+ i_size
, 0,
1957 EXT4_MIN_INLINE_DATA_SIZE
- i_size
);
1960 EXT4_I(inode
)->i_inline_size
= i_size
<
1961 EXT4_MIN_INLINE_DATA_SIZE
?
1962 EXT4_MIN_INLINE_DATA_SIZE
: i_size
;
1966 up_write(&EXT4_I(inode
)->i_data_sem
);
1969 up_write(&EXT4_I(inode
)->xattr_sem
);
1972 ext4_orphan_del(handle
, inode
);
1974 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
1975 ext4_mark_inode_dirty(handle
, inode
);
1977 ext4_handle_sync(handle
);
1979 ext4_journal_stop(handle
);
1983 int ext4_convert_inline_data(struct inode
*inode
)
1985 int error
, needed_blocks
;
1987 struct ext4_iloc iloc
;
1989 if (!ext4_has_inline_data(inode
)) {
1990 ext4_clear_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
1994 needed_blocks
= ext4_writepage_trans_blocks(inode
);
1997 error
= ext4_get_inode_loc(inode
, &iloc
);
2001 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
2002 if (IS_ERR(handle
)) {
2003 error
= PTR_ERR(handle
);
2007 down_write(&EXT4_I(inode
)->xattr_sem
);
2008 if (!ext4_has_inline_data(inode
)) {
2009 up_write(&EXT4_I(inode
)->xattr_sem
);
2013 error
= ext4_convert_inline_data_nolock(handle
, inode
, &iloc
);
2014 up_write(&EXT4_I(inode
)->xattr_sem
);
2016 ext4_journal_stop(handle
);