1 // SPDX-License-Identifier: LGPL-2.1
3 * Copyright (c) 2012 Taobao.
4 * Written by Tao Ma <boyu.mt@taobao.com>
7 #include <linux/iomap.h>
8 #include <linux/fiemap.h>
9 #include <linux/namei.h>
10 #include <linux/iversion.h>
11 #include <linux/sched/mm.h>
13 #include "ext4_jbd2.h"
18 #define EXT4_XATTR_SYSTEM_DATA "data"
19 #define EXT4_MIN_INLINE_DATA_SIZE ((sizeof(__le32) * EXT4_N_BLOCKS))
20 #define EXT4_INLINE_DOTDOT_OFFSET 2
21 #define EXT4_INLINE_DOTDOT_SIZE 4
23 static int ext4_get_inline_size(struct inode
*inode
)
25 if (EXT4_I(inode
)->i_inline_off
)
26 return EXT4_I(inode
)->i_inline_size
;
31 static int get_max_inline_xattr_value_size(struct inode
*inode
,
32 struct ext4_iloc
*iloc
)
34 struct ext4_xattr_ibody_header
*header
;
35 struct ext4_xattr_entry
*entry
;
36 struct ext4_inode
*raw_inode
;
40 if (!EXT4_INODE_HAS_XATTR_SPACE(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
);
61 end
= (void *)raw_inode
+ EXT4_SB(inode
->i_sb
)->s_inode_size
;
63 /* Compute min_offs. */
64 while (!IS_LAST_ENTRY(entry
)) {
65 void *next
= EXT4_XATTR_NEXT(entry
);
68 EXT4_ERROR_INODE(inode
,
69 "corrupt xattr in inline inode");
72 if (!entry
->e_value_inum
&& entry
->e_value_size
) {
73 size_t offs
= le16_to_cpu(entry
->e_value_offs
);
80 ((void *)entry
- (void *)IFIRST(header
)) - sizeof(__u32
);
82 if (EXT4_I(inode
)->i_inline_off
) {
83 entry
= (struct ext4_xattr_entry
*)
84 ((void *)raw_inode
+ EXT4_I(inode
)->i_inline_off
);
86 free
+= EXT4_XATTR_SIZE(le32_to_cpu(entry
->e_value_size
));
90 free
-= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA
));
92 if (free
> EXT4_XATTR_ROUND
)
93 free
= EXT4_XATTR_SIZE(free
- EXT4_XATTR_ROUND
);
102 * Get the maximum size we now can store in an inode.
103 * If we can't find the space for a xattr entry, don't use the space
104 * of the extents since we have no space to indicate the inline data.
106 int ext4_get_max_inline_size(struct inode
*inode
)
108 int error
, max_inline_size
;
109 struct ext4_iloc iloc
;
111 if (EXT4_I(inode
)->i_extra_isize
== 0)
114 error
= ext4_get_inode_loc(inode
, &iloc
);
116 ext4_error_inode_err(inode
, __func__
, __LINE__
, 0, -error
,
117 "can't get inode location %lu",
122 down_read(&EXT4_I(inode
)->xattr_sem
);
123 max_inline_size
= get_max_inline_xattr_value_size(inode
, &iloc
);
124 up_read(&EXT4_I(inode
)->xattr_sem
);
128 if (!max_inline_size
)
131 return max_inline_size
+ EXT4_MIN_INLINE_DATA_SIZE
;
135 * this function does not take xattr_sem, which is OK because it is
136 * currently only used in a code path coming form ext4_iget, before
137 * the new inode has been unlocked
139 int ext4_find_inline_data_nolock(struct inode
*inode
)
141 struct ext4_xattr_ibody_find is
= {
142 .s
= { .not_found
= -ENODATA
, },
144 struct ext4_xattr_info i
= {
145 .name_index
= EXT4_XATTR_INDEX_SYSTEM
,
146 .name
= EXT4_XATTR_SYSTEM_DATA
,
150 if (EXT4_I(inode
)->i_extra_isize
== 0)
153 error
= ext4_get_inode_loc(inode
, &is
.iloc
);
157 error
= ext4_xattr_ibody_find(inode
, &i
, &is
);
161 if (!is
.s
.not_found
) {
162 if (is
.s
.here
->e_value_inum
) {
163 EXT4_ERROR_INODE(inode
, "inline data xattr refers "
164 "to an external xattr inode");
165 error
= -EFSCORRUPTED
;
168 EXT4_I(inode
)->i_inline_off
= (u16
)((void *)is
.s
.here
-
169 (void *)ext4_raw_inode(&is
.iloc
));
170 EXT4_I(inode
)->i_inline_size
= EXT4_MIN_INLINE_DATA_SIZE
+
171 le32_to_cpu(is
.s
.here
->e_value_size
);
178 static int ext4_read_inline_data(struct inode
*inode
, void *buffer
,
180 struct ext4_iloc
*iloc
)
182 struct ext4_xattr_entry
*entry
;
183 struct ext4_xattr_ibody_header
*header
;
185 struct ext4_inode
*raw_inode
;
190 BUG_ON(len
> EXT4_I(inode
)->i_inline_size
);
192 cp_len
= min_t(unsigned int, len
, EXT4_MIN_INLINE_DATA_SIZE
);
194 raw_inode
= ext4_raw_inode(iloc
);
195 memcpy(buffer
, (void *)(raw_inode
->i_block
), cp_len
);
203 header
= IHDR(inode
, raw_inode
);
204 entry
= (struct ext4_xattr_entry
*)((void *)raw_inode
+
205 EXT4_I(inode
)->i_inline_off
);
206 len
= min_t(unsigned int, len
,
207 (unsigned int)le32_to_cpu(entry
->e_value_size
));
210 (void *)IFIRST(header
) + le16_to_cpu(entry
->e_value_offs
), len
);
218 * write the buffer to the inline inode.
219 * If 'create' is set, we don't need to do the extra copy in the xattr
220 * value since it is already handled by ext4_xattr_ibody_set.
221 * That saves us one memcpy.
223 static void ext4_write_inline_data(struct inode
*inode
, struct ext4_iloc
*iloc
,
224 void *buffer
, loff_t pos
, unsigned int len
)
226 struct ext4_xattr_entry
*entry
;
227 struct ext4_xattr_ibody_header
*header
;
228 struct ext4_inode
*raw_inode
;
231 if (unlikely(ext4_forced_shutdown(inode
->i_sb
)))
234 BUG_ON(!EXT4_I(inode
)->i_inline_off
);
235 BUG_ON(pos
+ len
> EXT4_I(inode
)->i_inline_size
);
237 raw_inode
= ext4_raw_inode(iloc
);
240 if (pos
< EXT4_MIN_INLINE_DATA_SIZE
) {
241 cp_len
= pos
+ len
> EXT4_MIN_INLINE_DATA_SIZE
?
242 EXT4_MIN_INLINE_DATA_SIZE
- pos
: len
;
243 memcpy((void *)raw_inode
->i_block
+ pos
, buffer
, cp_len
);
253 pos
-= EXT4_MIN_INLINE_DATA_SIZE
;
254 header
= IHDR(inode
, raw_inode
);
255 entry
= (struct ext4_xattr_entry
*)((void *)raw_inode
+
256 EXT4_I(inode
)->i_inline_off
);
258 memcpy((void *)IFIRST(header
) + le16_to_cpu(entry
->e_value_offs
) + pos
,
262 static int ext4_create_inline_data(handle_t
*handle
,
263 struct inode
*inode
, unsigned len
)
267 struct ext4_xattr_ibody_find is
= {
268 .s
= { .not_found
= -ENODATA
, },
270 struct ext4_xattr_info i
= {
271 .name_index
= EXT4_XATTR_INDEX_SYSTEM
,
272 .name
= EXT4_XATTR_SYSTEM_DATA
,
275 error
= ext4_get_inode_loc(inode
, &is
.iloc
);
279 BUFFER_TRACE(is
.iloc
.bh
, "get_write_access");
280 error
= ext4_journal_get_write_access(handle
, inode
->i_sb
, is
.iloc
.bh
,
285 if (len
> EXT4_MIN_INLINE_DATA_SIZE
) {
286 value
= EXT4_ZERO_XATTR_VALUE
;
287 len
-= EXT4_MIN_INLINE_DATA_SIZE
;
293 /* Insert the xttr entry. */
297 error
= ext4_xattr_ibody_find(inode
, &i
, &is
);
301 BUG_ON(!is
.s
.not_found
);
303 error
= ext4_xattr_ibody_set(handle
, inode
, &i
, &is
);
305 if (error
== -ENOSPC
)
306 ext4_clear_inode_state(inode
,
307 EXT4_STATE_MAY_INLINE_DATA
);
311 memset((void *)ext4_raw_inode(&is
.iloc
)->i_block
,
312 0, EXT4_MIN_INLINE_DATA_SIZE
);
314 EXT4_I(inode
)->i_inline_off
= (u16
)((void *)is
.s
.here
-
315 (void *)ext4_raw_inode(&is
.iloc
));
316 EXT4_I(inode
)->i_inline_size
= len
+ EXT4_MIN_INLINE_DATA_SIZE
;
317 ext4_clear_inode_flag(inode
, EXT4_INODE_EXTENTS
);
318 ext4_set_inode_flag(inode
, EXT4_INODE_INLINE_DATA
);
320 error
= ext4_mark_iloc_dirty(handle
, inode
, &is
.iloc
);
327 static int ext4_update_inline_data(handle_t
*handle
, struct inode
*inode
,
332 struct ext4_xattr_ibody_find is
= {
333 .s
= { .not_found
= -ENODATA
, },
335 struct ext4_xattr_info i
= {
336 .name_index
= EXT4_XATTR_INDEX_SYSTEM
,
337 .name
= EXT4_XATTR_SYSTEM_DATA
,
340 /* If the old space is ok, write the data directly. */
341 if (len
<= EXT4_I(inode
)->i_inline_size
)
344 error
= ext4_get_inode_loc(inode
, &is
.iloc
);
348 error
= ext4_xattr_ibody_find(inode
, &i
, &is
);
352 BUG_ON(is
.s
.not_found
);
354 len
-= EXT4_MIN_INLINE_DATA_SIZE
;
355 value
= kzalloc(len
, GFP_NOFS
);
361 error
= ext4_xattr_ibody_get(inode
, i
.name_index
, i
.name
,
366 BUFFER_TRACE(is
.iloc
.bh
, "get_write_access");
367 error
= ext4_journal_get_write_access(handle
, inode
->i_sb
, is
.iloc
.bh
,
372 /* Update the xattr entry. */
376 error
= ext4_xattr_ibody_set(handle
, inode
, &i
, &is
);
380 EXT4_I(inode
)->i_inline_off
= (u16
)((void *)is
.s
.here
-
381 (void *)ext4_raw_inode(&is
.iloc
));
382 EXT4_I(inode
)->i_inline_size
= EXT4_MIN_INLINE_DATA_SIZE
+
383 le32_to_cpu(is
.s
.here
->e_value_size
);
384 ext4_set_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
386 error
= ext4_mark_iloc_dirty(handle
, inode
, &is
.iloc
);
394 static int ext4_prepare_inline_data(handle_t
*handle
, struct inode
*inode
,
397 int ret
, size
, no_expand
;
398 struct ext4_inode_info
*ei
= EXT4_I(inode
);
400 if (!ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
))
403 size
= ext4_get_max_inline_size(inode
);
407 ext4_write_lock_xattr(inode
, &no_expand
);
409 if (ei
->i_inline_off
)
410 ret
= ext4_update_inline_data(handle
, inode
, len
);
412 ret
= ext4_create_inline_data(handle
, inode
, len
);
414 ext4_write_unlock_xattr(inode
, &no_expand
);
418 static int ext4_destroy_inline_data_nolock(handle_t
*handle
,
421 struct ext4_inode_info
*ei
= EXT4_I(inode
);
422 struct ext4_xattr_ibody_find is
= {
423 .s
= { .not_found
= 0, },
425 struct ext4_xattr_info i
= {
426 .name_index
= EXT4_XATTR_INDEX_SYSTEM
,
427 .name
= EXT4_XATTR_SYSTEM_DATA
,
433 if (!ei
->i_inline_off
)
436 error
= ext4_get_inode_loc(inode
, &is
.iloc
);
440 error
= ext4_xattr_ibody_find(inode
, &i
, &is
);
444 BUFFER_TRACE(is
.iloc
.bh
, "get_write_access");
445 error
= ext4_journal_get_write_access(handle
, inode
->i_sb
, is
.iloc
.bh
,
450 error
= ext4_xattr_ibody_set(handle
, inode
, &i
, &is
);
454 memset((void *)ext4_raw_inode(&is
.iloc
)->i_block
,
455 0, EXT4_MIN_INLINE_DATA_SIZE
);
456 memset(ei
->i_data
, 0, EXT4_MIN_INLINE_DATA_SIZE
);
458 if (ext4_has_feature_extents(inode
->i_sb
)) {
459 if (S_ISDIR(inode
->i_mode
) ||
460 S_ISREG(inode
->i_mode
) || S_ISLNK(inode
->i_mode
)) {
461 ext4_set_inode_flag(inode
, EXT4_INODE_EXTENTS
);
462 ext4_ext_tree_init(handle
, inode
);
465 ext4_clear_inode_flag(inode
, EXT4_INODE_INLINE_DATA
);
468 error
= ext4_mark_iloc_dirty(handle
, inode
, &is
.iloc
);
470 EXT4_I(inode
)->i_inline_off
= 0;
471 EXT4_I(inode
)->i_inline_size
= 0;
472 ext4_clear_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
475 if (error
== -ENODATA
)
480 static int ext4_read_inline_folio(struct inode
*inode
, struct folio
*folio
)
485 struct ext4_iloc iloc
;
487 BUG_ON(!folio_test_locked(folio
));
488 BUG_ON(!ext4_has_inline_data(inode
));
489 BUG_ON(folio
->index
);
491 if (!EXT4_I(inode
)->i_inline_off
) {
492 ext4_warning(inode
->i_sb
, "inode %lu doesn't have inline data.",
497 ret
= ext4_get_inode_loc(inode
, &iloc
);
501 len
= min_t(size_t, ext4_get_inline_size(inode
), i_size_read(inode
));
502 BUG_ON(len
> PAGE_SIZE
);
503 kaddr
= kmap_local_folio(folio
, 0);
504 ret
= ext4_read_inline_data(inode
, kaddr
, len
, &iloc
);
505 kaddr
= folio_zero_tail(folio
, len
, kaddr
+ len
);
507 folio_mark_uptodate(folio
);
514 int ext4_readpage_inline(struct inode
*inode
, struct folio
*folio
)
518 down_read(&EXT4_I(inode
)->xattr_sem
);
519 if (!ext4_has_inline_data(inode
)) {
520 up_read(&EXT4_I(inode
)->xattr_sem
);
525 * Current inline data can only exist in the 1st page,
526 * So for all the other pages, just set them uptodate.
529 ret
= ext4_read_inline_folio(inode
, folio
);
530 else if (!folio_test_uptodate(folio
)) {
531 folio_zero_segment(folio
, 0, folio_size(folio
));
532 folio_mark_uptodate(folio
);
535 up_read(&EXT4_I(inode
)->xattr_sem
);
538 return ret
>= 0 ? 0 : ret
;
541 static int ext4_convert_inline_data_to_extent(struct address_space
*mapping
,
544 int ret
, needed_blocks
, no_expand
;
545 handle_t
*handle
= NULL
;
546 int retries
= 0, sem_held
= 0;
547 struct folio
*folio
= NULL
;
549 struct ext4_iloc iloc
;
551 if (!ext4_has_inline_data(inode
)) {
553 * clear the flag so that no new write
554 * will trap here again.
556 ext4_clear_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
560 needed_blocks
= ext4_writepage_trans_blocks(inode
);
562 ret
= ext4_get_inode_loc(inode
, &iloc
);
567 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
568 if (IS_ERR(handle
)) {
569 ret
= PTR_ERR(handle
);
574 /* We cannot recurse into the filesystem as the transaction is already
576 folio
= __filemap_get_folio(mapping
, 0, FGP_WRITEBEGIN
| FGP_NOFS
,
577 mapping_gfp_mask(mapping
));
579 ret
= PTR_ERR(folio
);
583 ext4_write_lock_xattr(inode
, &no_expand
);
585 /* If some one has already done this for us, just exit. */
586 if (!ext4_has_inline_data(inode
)) {
592 to
= ext4_get_inline_size(inode
);
593 if (!folio_test_uptodate(folio
)) {
594 ret
= ext4_read_inline_folio(inode
, folio
);
599 ret
= ext4_destroy_inline_data_nolock(handle
, inode
);
603 if (ext4_should_dioread_nolock(inode
)) {
604 ret
= ext4_block_write_begin(handle
, folio
, from
, to
,
605 ext4_get_block_unwritten
);
607 ret
= ext4_block_write_begin(handle
, folio
, from
, to
,
610 if (!ret
&& ext4_should_journal_data(inode
)) {
611 ret
= ext4_walk_page_buffers(handle
, inode
,
612 folio_buffers(folio
), from
, to
,
613 NULL
, do_journal_get_write_access
);
620 ext4_orphan_add(handle
, inode
);
621 ext4_write_unlock_xattr(inode
, &no_expand
);
623 ext4_journal_stop(handle
);
625 ext4_truncate_failed_write(inode
);
627 * If truncate failed early the inode might
628 * still be on the orphan list; we need to
629 * make sure the inode is removed from the
630 * orphan list in that case.
633 ext4_orphan_del(NULL
, inode
);
636 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
640 block_commit_write(&folio
->page
, from
, to
);
648 ext4_write_unlock_xattr(inode
, &no_expand
);
650 ext4_journal_stop(handle
);
656 * Try to write data in the inode.
657 * If the inode has inline data, check whether the new write can be
658 * in the inode also. If not, create the page the handle, move the data
659 * to the page make it update and let the later codes create extent for it.
661 int ext4_try_to_write_inline_data(struct address_space
*mapping
,
663 loff_t pos
, unsigned len
,
664 struct folio
**foliop
)
669 struct ext4_iloc iloc
;
671 if (pos
+ len
> ext4_get_max_inline_size(inode
))
674 ret
= ext4_get_inode_loc(inode
, &iloc
);
679 * The possible write could happen in the inode,
680 * so try to reserve the space in inode first.
682 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
683 if (IS_ERR(handle
)) {
684 ret
= PTR_ERR(handle
);
689 ret
= ext4_prepare_inline_data(handle
, inode
, pos
+ len
);
690 if (ret
&& ret
!= -ENOSPC
)
693 /* We don't have space in inline inode, so convert it to extent. */
694 if (ret
== -ENOSPC
) {
695 ext4_journal_stop(handle
);
700 ret
= ext4_journal_get_write_access(handle
, inode
->i_sb
, iloc
.bh
,
705 folio
= __filemap_get_folio(mapping
, 0, FGP_WRITEBEGIN
| FGP_NOFS
,
706 mapping_gfp_mask(mapping
));
708 ret
= PTR_ERR(folio
);
713 down_read(&EXT4_I(inode
)->xattr_sem
);
714 if (!ext4_has_inline_data(inode
)) {
721 if (!folio_test_uptodate(folio
)) {
722 ret
= ext4_read_inline_folio(inode
, folio
);
733 up_read(&EXT4_I(inode
)->xattr_sem
);
735 if (handle
&& (ret
!= 1))
736 ext4_journal_stop(handle
);
740 return ext4_convert_inline_data_to_extent(mapping
, inode
);
743 int ext4_write_inline_data_end(struct inode
*inode
, loff_t pos
, unsigned len
,
744 unsigned copied
, struct folio
*folio
)
746 handle_t
*handle
= ext4_journal_current_handle();
749 struct ext4_iloc iloc
;
752 if (unlikely(copied
< len
) && !folio_test_uptodate(folio
))
755 if (likely(copied
)) {
756 ret
= ext4_get_inode_loc(inode
, &iloc
);
760 ext4_std_error(inode
->i_sb
, ret
);
763 ext4_write_lock_xattr(inode
, &no_expand
);
764 BUG_ON(!ext4_has_inline_data(inode
));
767 * ei->i_inline_off may have changed since
768 * ext4_write_begin() called
769 * ext4_try_to_write_inline_data()
771 (void) ext4_find_inline_data_nolock(inode
);
773 kaddr
= kmap_local_folio(folio
, 0);
774 ext4_write_inline_data(inode
, &iloc
, kaddr
, pos
, copied
);
776 folio_mark_uptodate(folio
);
777 /* clear dirty flag so that writepages wouldn't work for us. */
778 folio_clear_dirty(folio
);
780 ext4_write_unlock_xattr(inode
, &no_expand
);
784 * It's important to update i_size while still holding folio
785 * lock: page writeout could otherwise come in and zero
788 ext4_update_inode_size(inode
, pos
+ copied
);
794 * Don't mark the inode dirty under folio lock. First, it unnecessarily
795 * makes the holding time of folio lock longer. Second, it forces lock
796 * ordering of folio lock and transaction start for journaling
800 mark_inode_dirty(inode
);
803 * If we didn't copy as much data as expected, we need to trim back
804 * size of xattr containing inline data.
806 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
807 ext4_orphan_add(handle
, inode
);
809 ret2
= ext4_journal_stop(handle
);
812 if (pos
+ len
> inode
->i_size
) {
813 ext4_truncate_failed_write(inode
);
815 * If truncate failed early the inode might still be
816 * on the orphan list; we need to make sure the inode
817 * is removed from the orphan list in that case.
820 ext4_orphan_del(NULL
, inode
);
822 return ret
? ret
: copied
;
826 * Try to make the page cache and handle ready for the inline data case.
827 * We can call this function in 2 cases:
828 * 1. The inode is created and the first write exceeds inline size. We can
829 * clear the inode state safely.
830 * 2. The inode has inline data, then we need to read the data, make it
831 * update and dirty so that ext4_da_writepages can handle it. We don't
832 * need to start the journal since the file's metadata isn't changed now.
834 static int ext4_da_convert_inline_data_to_extent(struct address_space
*mapping
,
838 int ret
= 0, inline_size
;
841 folio
= __filemap_get_folio(mapping
, 0, FGP_WRITEBEGIN
,
842 mapping_gfp_mask(mapping
));
844 return PTR_ERR(folio
);
846 down_read(&EXT4_I(inode
)->xattr_sem
);
847 if (!ext4_has_inline_data(inode
)) {
848 ext4_clear_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
852 inline_size
= ext4_get_inline_size(inode
);
854 if (!folio_test_uptodate(folio
)) {
855 ret
= ext4_read_inline_folio(inode
, folio
);
860 ret
= ext4_block_write_begin(NULL
, folio
, 0, inline_size
,
861 ext4_da_get_block_prep
);
863 up_read(&EXT4_I(inode
)->xattr_sem
);
866 ext4_truncate_failed_write(inode
);
870 folio_mark_dirty(folio
);
871 folio_mark_uptodate(folio
);
872 ext4_clear_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
873 *fsdata
= (void *)CONVERT_INLINE_DATA
;
876 up_read(&EXT4_I(inode
)->xattr_sem
);
885 * Prepare the write for the inline data.
886 * If the data can be written into the inode, we just read
887 * the page and make it uptodate, and start the journal.
888 * Otherwise read the page, makes it dirty so that it can be
889 * handle in writepages(the i_disksize update is left to the
890 * normal ext4_da_write_end).
892 int ext4_da_write_inline_data_begin(struct address_space
*mapping
,
894 loff_t pos
, unsigned len
,
895 struct folio
**foliop
,
901 struct ext4_iloc iloc
;
904 ret
= ext4_get_inode_loc(inode
, &iloc
);
909 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
910 if (IS_ERR(handle
)) {
911 ret
= PTR_ERR(handle
);
915 ret
= ext4_prepare_inline_data(handle
, inode
, pos
+ len
);
916 if (ret
&& ret
!= -ENOSPC
)
919 if (ret
== -ENOSPC
) {
920 ext4_journal_stop(handle
);
921 ret
= ext4_da_convert_inline_data_to_extent(mapping
,
924 if (ret
== -ENOSPC
&&
925 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
931 * We cannot recurse into the filesystem as the transaction
932 * is already started.
934 folio
= __filemap_get_folio(mapping
, 0, FGP_WRITEBEGIN
| FGP_NOFS
,
935 mapping_gfp_mask(mapping
));
937 ret
= PTR_ERR(folio
);
941 down_read(&EXT4_I(inode
)->xattr_sem
);
942 if (!ext4_has_inline_data(inode
)) {
944 goto out_release_page
;
947 if (!folio_test_uptodate(folio
)) {
948 ret
= ext4_read_inline_folio(inode
, folio
);
950 goto out_release_page
;
952 ret
= ext4_journal_get_write_access(handle
, inode
->i_sb
, iloc
.bh
,
955 goto out_release_page
;
957 up_read(&EXT4_I(inode
)->xattr_sem
);
962 up_read(&EXT4_I(inode
)->xattr_sem
);
966 ext4_journal_stop(handle
);
972 #ifdef INLINE_DIR_DEBUG
973 void ext4_show_inline_dir(struct inode
*dir
, struct buffer_head
*bh
,
974 void *inline_start
, int inline_size
)
977 unsigned short de_len
;
978 struct ext4_dir_entry_2
*de
= inline_start
;
979 void *dlimit
= inline_start
+ inline_size
;
981 trace_printk("inode %lu\n", dir
->i_ino
);
983 while ((void *)de
< dlimit
) {
984 de_len
= ext4_rec_len_from_disk(de
->rec_len
, inline_size
);
985 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
986 offset
, de_len
, de
->name_len
, de
->name
,
987 de
->name_len
, le32_to_cpu(de
->inode
));
988 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
989 inline_start
, inline_size
, offset
))
993 de
= (struct ext4_dir_entry_2
*) ((char *) de
+ de_len
);
997 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1001 * Add a new entry into a inline dir.
1002 * It will return -ENOSPC if no space is available, and -EIO
1003 * and -EEXIST if directory entry already exists.
1005 static int ext4_add_dirent_to_inline(handle_t
*handle
,
1006 struct ext4_filename
*fname
,
1008 struct inode
*inode
,
1009 struct ext4_iloc
*iloc
,
1010 void *inline_start
, int inline_size
)
1013 struct ext4_dir_entry_2
*de
;
1015 err
= ext4_find_dest_de(dir
, inode
, iloc
->bh
, inline_start
,
1016 inline_size
, fname
, &de
);
1020 BUFFER_TRACE(iloc
->bh
, "get_write_access");
1021 err
= ext4_journal_get_write_access(handle
, dir
->i_sb
, iloc
->bh
,
1025 ext4_insert_dentry(dir
, inode
, de
, inline_size
, fname
);
1027 ext4_show_inline_dir(dir
, iloc
->bh
, inline_start
, inline_size
);
1030 * XXX shouldn't update any times until successful
1031 * completion of syscall, but too many callers depend
1034 * XXX similarly, too many callers depend on
1035 * ext4_new_inode() setting the times, but error
1036 * recovery deletes the inode, so the worst that can
1037 * happen is that the times are slightly out of date
1038 * and/or different from the directory change time.
1040 inode_set_mtime_to_ts(dir
, inode_set_ctime_current(dir
));
1041 ext4_update_dx_flag(dir
);
1042 inode_inc_iversion(dir
);
1046 static void *ext4_get_inline_xattr_pos(struct inode
*inode
,
1047 struct ext4_iloc
*iloc
)
1049 struct ext4_xattr_entry
*entry
;
1050 struct ext4_xattr_ibody_header
*header
;
1052 BUG_ON(!EXT4_I(inode
)->i_inline_off
);
1054 header
= IHDR(inode
, ext4_raw_inode(iloc
));
1055 entry
= (struct ext4_xattr_entry
*)((void *)ext4_raw_inode(iloc
) +
1056 EXT4_I(inode
)->i_inline_off
);
1058 return (void *)IFIRST(header
) + le16_to_cpu(entry
->e_value_offs
);
1061 /* Set the final de to cover the whole block. */
1062 static void ext4_update_final_de(void *de_buf
, int old_size
, int new_size
)
1064 struct ext4_dir_entry_2
*de
, *prev_de
;
1070 limit
= de_buf
+ old_size
;
1073 de_len
= ext4_rec_len_from_disk(de
->rec_len
, old_size
);
1076 } while (de_buf
< limit
);
1078 prev_de
->rec_len
= ext4_rec_len_to_disk(de_len
+ new_size
-
1079 old_size
, new_size
);
1081 /* this is just created, so create an empty entry. */
1083 de
->rec_len
= ext4_rec_len_to_disk(new_size
, new_size
);
1087 static int ext4_update_inline_dir(handle_t
*handle
, struct inode
*dir
,
1088 struct ext4_iloc
*iloc
)
1091 int old_size
= EXT4_I(dir
)->i_inline_size
- EXT4_MIN_INLINE_DATA_SIZE
;
1092 int new_size
= get_max_inline_xattr_value_size(dir
, iloc
);
1094 if (new_size
- old_size
<= ext4_dir_rec_len(1, NULL
))
1097 ret
= ext4_update_inline_data(handle
, dir
,
1098 new_size
+ EXT4_MIN_INLINE_DATA_SIZE
);
1102 ext4_update_final_de(ext4_get_inline_xattr_pos(dir
, iloc
), old_size
,
1103 EXT4_I(dir
)->i_inline_size
-
1104 EXT4_MIN_INLINE_DATA_SIZE
);
1105 dir
->i_size
= EXT4_I(dir
)->i_disksize
= EXT4_I(dir
)->i_inline_size
;
1109 static void ext4_restore_inline_data(handle_t
*handle
, struct inode
*inode
,
1110 struct ext4_iloc
*iloc
,
1111 void *buf
, int inline_size
)
1115 ret
= ext4_create_inline_data(handle
, inode
, inline_size
);
1117 ext4_msg(inode
->i_sb
, KERN_EMERG
,
1118 "error restoring inline_data for inode -- potential data loss! (inode %lu, error %d)",
1122 ext4_write_inline_data(inode
, iloc
, buf
, 0, inline_size
);
1123 ext4_set_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
1126 static int ext4_finish_convert_inline_dir(handle_t
*handle
,
1127 struct inode
*inode
,
1128 struct buffer_head
*dir_block
,
1132 int err
, csum_size
= 0, header_size
= 0;
1133 struct ext4_dir_entry_2
*de
;
1134 void *target
= dir_block
->b_data
;
1137 * First create "." and ".." and then copy the dir information
1138 * back to the block.
1141 de
= ext4_init_dot_dotdot(inode
, de
,
1142 inode
->i_sb
->s_blocksize
, csum_size
,
1143 le32_to_cpu(((struct ext4_dir_entry_2
*)buf
)->inode
), 1);
1144 header_size
= (void *)de
- target
;
1146 memcpy((void *)de
, buf
+ EXT4_INLINE_DOTDOT_SIZE
,
1147 inline_size
- EXT4_INLINE_DOTDOT_SIZE
);
1149 if (ext4_has_metadata_csum(inode
->i_sb
))
1150 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1152 inode
->i_size
= inode
->i_sb
->s_blocksize
;
1153 i_size_write(inode
, inode
->i_sb
->s_blocksize
);
1154 EXT4_I(inode
)->i_disksize
= inode
->i_sb
->s_blocksize
;
1155 ext4_update_final_de(dir_block
->b_data
,
1156 inline_size
- EXT4_INLINE_DOTDOT_SIZE
+ header_size
,
1157 inode
->i_sb
->s_blocksize
- csum_size
);
1160 ext4_initialize_dirent_tail(dir_block
,
1161 inode
->i_sb
->s_blocksize
);
1162 set_buffer_uptodate(dir_block
);
1163 unlock_buffer(dir_block
);
1164 err
= ext4_handle_dirty_dirblock(handle
, inode
, dir_block
);
1167 set_buffer_verified(dir_block
);
1168 return ext4_mark_inode_dirty(handle
, inode
);
1171 static int ext4_convert_inline_data_nolock(handle_t
*handle
,
1172 struct inode
*inode
,
1173 struct ext4_iloc
*iloc
)
1177 struct buffer_head
*data_bh
= NULL
;
1178 struct ext4_map_blocks map
;
1181 inline_size
= ext4_get_inline_size(inode
);
1182 buf
= kmalloc(inline_size
, GFP_NOFS
);
1188 error
= ext4_read_inline_data(inode
, buf
, inline_size
, iloc
);
1193 * Make sure the inline directory entries pass checks before we try to
1194 * convert them, so that we avoid touching stuff that needs fsck.
1196 if (S_ISDIR(inode
->i_mode
)) {
1197 error
= ext4_check_all_de(inode
, iloc
->bh
,
1198 buf
+ EXT4_INLINE_DOTDOT_SIZE
,
1199 inline_size
- EXT4_INLINE_DOTDOT_SIZE
);
1204 error
= ext4_destroy_inline_data_nolock(handle
, inode
);
1211 error
= ext4_map_blocks(handle
, inode
, &map
, EXT4_GET_BLOCKS_CREATE
);
1214 if (!(map
.m_flags
& EXT4_MAP_MAPPED
)) {
1219 data_bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
1225 lock_buffer(data_bh
);
1226 error
= ext4_journal_get_create_access(handle
, inode
->i_sb
, data_bh
,
1229 unlock_buffer(data_bh
);
1233 memset(data_bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
1235 if (!S_ISDIR(inode
->i_mode
)) {
1236 memcpy(data_bh
->b_data
, buf
, inline_size
);
1237 set_buffer_uptodate(data_bh
);
1238 unlock_buffer(data_bh
);
1239 error
= ext4_handle_dirty_metadata(handle
,
1242 error
= ext4_finish_convert_inline_dir(handle
, inode
, data_bh
,
1248 ext4_restore_inline_data(handle
, inode
, iloc
, buf
, inline_size
);
1257 * Try to add the new entry to the inline data.
1258 * If succeeds, return 0. If not, extended the inline dir and copied data to
1259 * the new created block.
1261 int ext4_try_add_inline_entry(handle_t
*handle
, struct ext4_filename
*fname
,
1262 struct inode
*dir
, struct inode
*inode
)
1264 int ret
, ret2
, inline_size
, no_expand
;
1266 struct ext4_iloc iloc
;
1268 ret
= ext4_get_inode_loc(dir
, &iloc
);
1272 ext4_write_lock_xattr(dir
, &no_expand
);
1273 if (!ext4_has_inline_data(dir
))
1276 inline_start
= (void *)ext4_raw_inode(&iloc
)->i_block
+
1277 EXT4_INLINE_DOTDOT_SIZE
;
1278 inline_size
= EXT4_MIN_INLINE_DATA_SIZE
- EXT4_INLINE_DOTDOT_SIZE
;
1280 ret
= ext4_add_dirent_to_inline(handle
, fname
, dir
, inode
, &iloc
,
1281 inline_start
, inline_size
);
1285 /* check whether it can be inserted to inline xattr space. */
1286 inline_size
= EXT4_I(dir
)->i_inline_size
-
1287 EXT4_MIN_INLINE_DATA_SIZE
;
1289 /* Try to use the xattr space.*/
1290 ret
= ext4_update_inline_dir(handle
, dir
, &iloc
);
1291 if (ret
&& ret
!= -ENOSPC
)
1294 inline_size
= EXT4_I(dir
)->i_inline_size
-
1295 EXT4_MIN_INLINE_DATA_SIZE
;
1299 inline_start
= ext4_get_inline_xattr_pos(dir
, &iloc
);
1301 ret
= ext4_add_dirent_to_inline(handle
, fname
, dir
,
1302 inode
, &iloc
, inline_start
,
1310 * The inline space is filled up, so create a new block for it.
1311 * As the extent tree will be created, we have to save the inline
1314 ret
= ext4_convert_inline_data_nolock(handle
, dir
, &iloc
);
1317 ext4_write_unlock_xattr(dir
, &no_expand
);
1318 ret2
= ext4_mark_inode_dirty(handle
, dir
);
1319 if (unlikely(ret2
&& !ret
))
1326 * This function fills a red-black tree with information from an
1327 * inlined dir. It returns the number directory entries loaded
1328 * into the tree. If there is an error it is returned in err.
1330 int ext4_inlinedir_to_tree(struct file
*dir_file
,
1331 struct inode
*dir
, ext4_lblk_t block
,
1332 struct dx_hash_info
*hinfo
,
1333 __u32 start_hash
, __u32 start_minor_hash
,
1334 int *has_inline_data
)
1336 int err
= 0, count
= 0;
1337 unsigned int parent_ino
;
1339 struct ext4_dir_entry_2
*de
;
1340 struct inode
*inode
= file_inode(dir_file
);
1341 int ret
, inline_size
= 0;
1342 struct ext4_iloc iloc
;
1343 void *dir_buf
= NULL
;
1344 struct ext4_dir_entry_2 fake
;
1345 struct fscrypt_str tmp_str
;
1347 ret
= ext4_get_inode_loc(inode
, &iloc
);
1351 down_read(&EXT4_I(inode
)->xattr_sem
);
1352 if (!ext4_has_inline_data(inode
)) {
1353 up_read(&EXT4_I(inode
)->xattr_sem
);
1354 *has_inline_data
= 0;
1358 inline_size
= ext4_get_inline_size(inode
);
1359 dir_buf
= kmalloc(inline_size
, GFP_NOFS
);
1362 up_read(&EXT4_I(inode
)->xattr_sem
);
1366 ret
= ext4_read_inline_data(inode
, dir_buf
, inline_size
, &iloc
);
1367 up_read(&EXT4_I(inode
)->xattr_sem
);
1372 parent_ino
= le32_to_cpu(((struct ext4_dir_entry_2
*)dir_buf
)->inode
);
1373 while (pos
< inline_size
) {
1375 * As inlined dir doesn't store any information about '.' and
1376 * only the inode number of '..' is stored, we have to handle
1380 fake
.inode
= cpu_to_le32(inode
->i_ino
);
1382 strcpy(fake
.name
, ".");
1383 fake
.rec_len
= ext4_rec_len_to_disk(
1384 ext4_dir_rec_len(fake
.name_len
, NULL
),
1386 ext4_set_de_type(inode
->i_sb
, &fake
, S_IFDIR
);
1388 pos
= EXT4_INLINE_DOTDOT_OFFSET
;
1389 } else if (pos
== EXT4_INLINE_DOTDOT_OFFSET
) {
1390 fake
.inode
= cpu_to_le32(parent_ino
);
1392 strcpy(fake
.name
, "..");
1393 fake
.rec_len
= ext4_rec_len_to_disk(
1394 ext4_dir_rec_len(fake
.name_len
, NULL
),
1396 ext4_set_de_type(inode
->i_sb
, &fake
, S_IFDIR
);
1398 pos
= EXT4_INLINE_DOTDOT_SIZE
;
1400 de
= (struct ext4_dir_entry_2
*)(dir_buf
+ pos
);
1401 pos
+= ext4_rec_len_from_disk(de
->rec_len
, inline_size
);
1402 if (ext4_check_dir_entry(inode
, dir_file
, de
,
1404 inline_size
, pos
)) {
1410 if (ext4_hash_in_dirent(dir
)) {
1411 hinfo
->hash
= EXT4_DIRENT_HASH(de
);
1412 hinfo
->minor_hash
= EXT4_DIRENT_MINOR_HASH(de
);
1414 err
= ext4fs_dirhash(dir
, de
->name
, de
->name_len
, hinfo
);
1420 if ((hinfo
->hash
< start_hash
) ||
1421 ((hinfo
->hash
== start_hash
) &&
1422 (hinfo
->minor_hash
< start_minor_hash
)))
1426 tmp_str
.name
= de
->name
;
1427 tmp_str
.len
= de
->name_len
;
1428 err
= ext4_htree_store_dirent(dir_file
, hinfo
->hash
,
1429 hinfo
->minor_hash
, de
, &tmp_str
);
1444 * So this function is called when the volume is mkfsed with
1445 * dir_index disabled. In order to keep f_pos persistent
1446 * after we convert from an inlined dir to a blocked based,
1447 * we just pretend that we are a normal dir and return the
1448 * offset as if '.' and '..' really take place.
1451 int ext4_read_inline_dir(struct file
*file
,
1452 struct dir_context
*ctx
,
1453 int *has_inline_data
)
1455 unsigned int offset
, parent_ino
;
1457 struct ext4_dir_entry_2
*de
;
1458 struct super_block
*sb
;
1459 struct inode
*inode
= file_inode(file
);
1460 int ret
, inline_size
= 0;
1461 struct ext4_iloc iloc
;
1462 void *dir_buf
= NULL
;
1463 int dotdot_offset
, dotdot_size
, extra_offset
, extra_size
;
1464 struct dir_private_info
*info
= file
->private_data
;
1466 ret
= ext4_get_inode_loc(inode
, &iloc
);
1470 down_read(&EXT4_I(inode
)->xattr_sem
);
1471 if (!ext4_has_inline_data(inode
)) {
1472 up_read(&EXT4_I(inode
)->xattr_sem
);
1473 *has_inline_data
= 0;
1477 inline_size
= ext4_get_inline_size(inode
);
1478 dir_buf
= kmalloc(inline_size
, GFP_NOFS
);
1481 up_read(&EXT4_I(inode
)->xattr_sem
);
1485 ret
= ext4_read_inline_data(inode
, dir_buf
, inline_size
, &iloc
);
1486 up_read(&EXT4_I(inode
)->xattr_sem
);
1492 parent_ino
= le32_to_cpu(((struct ext4_dir_entry_2
*)dir_buf
)->inode
);
1496 * dotdot_offset and dotdot_size is the real offset and
1497 * size for ".." and "." if the dir is block based while
1498 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1499 * So we will use extra_offset and extra_size to indicate them
1500 * during the inline dir iteration.
1502 dotdot_offset
= ext4_dir_rec_len(1, NULL
);
1503 dotdot_size
= dotdot_offset
+ ext4_dir_rec_len(2, NULL
);
1504 extra_offset
= dotdot_size
- EXT4_INLINE_DOTDOT_SIZE
;
1505 extra_size
= extra_offset
+ inline_size
;
1508 * If the cookie has changed since the last call to
1509 * readdir(2), then we might be pointing to an invalid
1510 * dirent right now. Scan from the start of the inline
1513 if (!inode_eq_iversion(inode
, info
->cookie
)) {
1514 for (i
= 0; i
< extra_size
&& i
< offset
;) {
1516 * "." is with offset 0 and
1517 * ".." is dotdot_offset.
1522 } else if (i
== dotdot_offset
) {
1526 /* for other entry, the real offset in
1527 * the buf has to be tuned accordingly.
1529 de
= (struct ext4_dir_entry_2
*)
1530 (dir_buf
+ i
- extra_offset
);
1531 /* It's too expensive to do a full
1532 * dirent test each time round this
1533 * loop, but we do have to test at
1534 * least that it is non-zero. A
1535 * failure will be detected in the
1536 * dirent test below. */
1537 if (ext4_rec_len_from_disk(de
->rec_len
, extra_size
)
1538 < ext4_dir_rec_len(1, NULL
))
1540 i
+= ext4_rec_len_from_disk(de
->rec_len
,
1545 info
->cookie
= inode_query_iversion(inode
);
1548 while (ctx
->pos
< extra_size
) {
1549 if (ctx
->pos
== 0) {
1550 if (!dir_emit(ctx
, ".", 1, inode
->i_ino
, DT_DIR
))
1552 ctx
->pos
= dotdot_offset
;
1556 if (ctx
->pos
== dotdot_offset
) {
1557 if (!dir_emit(ctx
, "..", 2, parent_ino
, DT_DIR
))
1559 ctx
->pos
= dotdot_size
;
1563 de
= (struct ext4_dir_entry_2
*)
1564 (dir_buf
+ ctx
->pos
- extra_offset
);
1565 if (ext4_check_dir_entry(inode
, file
, de
, iloc
.bh
, dir_buf
,
1566 extra_size
, ctx
->pos
))
1568 if (le32_to_cpu(de
->inode
)) {
1569 if (!dir_emit(ctx
, de
->name
, de
->name_len
,
1570 le32_to_cpu(de
->inode
),
1571 get_dtype(sb
, de
->file_type
)))
1574 ctx
->pos
+= ext4_rec_len_from_disk(de
->rec_len
, extra_size
);
1582 void *ext4_read_inline_link(struct inode
*inode
)
1584 struct ext4_iloc iloc
;
1585 int ret
, inline_size
;
1588 ret
= ext4_get_inode_loc(inode
, &iloc
);
1590 return ERR_PTR(ret
);
1593 inline_size
= ext4_get_inline_size(inode
);
1594 link
= kmalloc(inline_size
+ 1, GFP_NOFS
);
1598 ret
= ext4_read_inline_data(inode
, link
, inline_size
, &iloc
);
1603 nd_terminate_link(link
, inode
->i_size
, ret
);
1606 link
= ERR_PTR(ret
);
1611 struct buffer_head
*ext4_get_first_inline_block(struct inode
*inode
,
1612 struct ext4_dir_entry_2
**parent_de
,
1615 struct ext4_iloc iloc
;
1617 *retval
= ext4_get_inode_loc(inode
, &iloc
);
1621 *parent_de
= (struct ext4_dir_entry_2
*)ext4_raw_inode(&iloc
)->i_block
;
1627 * Try to create the inline data for the new dir.
1628 * If it succeeds, return 0, otherwise return the error.
1629 * In case of ENOSPC, the caller should create the normal disk layout dir.
1631 int ext4_try_create_inline_dir(handle_t
*handle
, struct inode
*parent
,
1632 struct inode
*inode
)
1634 int ret
, inline_size
= EXT4_MIN_INLINE_DATA_SIZE
;
1635 struct ext4_iloc iloc
;
1636 struct ext4_dir_entry_2
*de
;
1638 ret
= ext4_get_inode_loc(inode
, &iloc
);
1642 ret
= ext4_prepare_inline_data(handle
, inode
, inline_size
);
1647 * For inline dir, we only save the inode information for the ".."
1648 * and create a fake dentry to cover the left space.
1650 de
= (struct ext4_dir_entry_2
*)ext4_raw_inode(&iloc
)->i_block
;
1651 de
->inode
= cpu_to_le32(parent
->i_ino
);
1652 de
= (struct ext4_dir_entry_2
*)((void *)de
+ EXT4_INLINE_DOTDOT_SIZE
);
1654 de
->rec_len
= ext4_rec_len_to_disk(
1655 inline_size
- EXT4_INLINE_DOTDOT_SIZE
,
1657 set_nlink(inode
, 2);
1658 inode
->i_size
= EXT4_I(inode
)->i_disksize
= inline_size
;
1664 struct buffer_head
*ext4_find_inline_entry(struct inode
*dir
,
1665 struct ext4_filename
*fname
,
1666 struct ext4_dir_entry_2
**res_dir
,
1667 int *has_inline_data
)
1669 struct ext4_xattr_ibody_find is
= {
1670 .s
= { .not_found
= -ENODATA
, },
1672 struct ext4_xattr_info i
= {
1673 .name_index
= EXT4_XATTR_INDEX_SYSTEM
,
1674 .name
= EXT4_XATTR_SYSTEM_DATA
,
1680 ret
= ext4_get_inode_loc(dir
, &is
.iloc
);
1682 return ERR_PTR(ret
);
1684 down_read(&EXT4_I(dir
)->xattr_sem
);
1686 ret
= ext4_xattr_ibody_find(dir
, &i
, &is
);
1690 if (!ext4_has_inline_data(dir
)) {
1691 *has_inline_data
= 0;
1695 inline_start
= (void *)ext4_raw_inode(&is
.iloc
)->i_block
+
1696 EXT4_INLINE_DOTDOT_SIZE
;
1697 inline_size
= EXT4_MIN_INLINE_DATA_SIZE
- EXT4_INLINE_DOTDOT_SIZE
;
1698 ret
= ext4_search_dir(is
.iloc
.bh
, inline_start
, inline_size
,
1699 dir
, fname
, 0, res_dir
);
1705 if (ext4_get_inline_size(dir
) == EXT4_MIN_INLINE_DATA_SIZE
)
1708 inline_start
= ext4_get_inline_xattr_pos(dir
, &is
.iloc
);
1709 inline_size
= ext4_get_inline_size(dir
) - EXT4_MIN_INLINE_DATA_SIZE
;
1711 ret
= ext4_search_dir(is
.iloc
.bh
, inline_start
, inline_size
,
1712 dir
, fname
, 0, res_dir
);
1719 is
.iloc
.bh
= ERR_PTR(ret
);
1723 up_read(&EXT4_I(dir
)->xattr_sem
);
1727 int ext4_delete_inline_entry(handle_t
*handle
,
1729 struct ext4_dir_entry_2
*de_del
,
1730 struct buffer_head
*bh
,
1731 int *has_inline_data
)
1733 int err
, inline_size
, no_expand
;
1734 struct ext4_iloc iloc
;
1737 err
= ext4_get_inode_loc(dir
, &iloc
);
1741 ext4_write_lock_xattr(dir
, &no_expand
);
1742 if (!ext4_has_inline_data(dir
)) {
1743 *has_inline_data
= 0;
1747 if ((void *)de_del
- ((void *)ext4_raw_inode(&iloc
)->i_block
) <
1748 EXT4_MIN_INLINE_DATA_SIZE
) {
1749 inline_start
= (void *)ext4_raw_inode(&iloc
)->i_block
+
1750 EXT4_INLINE_DOTDOT_SIZE
;
1751 inline_size
= EXT4_MIN_INLINE_DATA_SIZE
-
1752 EXT4_INLINE_DOTDOT_SIZE
;
1754 inline_start
= ext4_get_inline_xattr_pos(dir
, &iloc
);
1755 inline_size
= ext4_get_inline_size(dir
) -
1756 EXT4_MIN_INLINE_DATA_SIZE
;
1759 BUFFER_TRACE(bh
, "get_write_access");
1760 err
= ext4_journal_get_write_access(handle
, dir
->i_sb
, bh
,
1765 err
= ext4_generic_delete_entry(dir
, de_del
, bh
,
1766 inline_start
, inline_size
, 0);
1770 ext4_show_inline_dir(dir
, iloc
.bh
, inline_start
, inline_size
);
1772 ext4_write_unlock_xattr(dir
, &no_expand
);
1773 if (likely(err
== 0))
1774 err
= ext4_mark_inode_dirty(handle
, dir
);
1777 ext4_std_error(dir
->i_sb
, err
);
1782 * Get the inline dentry at offset.
1784 static inline struct ext4_dir_entry_2
*
1785 ext4_get_inline_entry(struct inode
*inode
,
1786 struct ext4_iloc
*iloc
,
1787 unsigned int offset
,
1788 void **inline_start
,
1793 BUG_ON(offset
> ext4_get_inline_size(inode
));
1795 if (offset
< EXT4_MIN_INLINE_DATA_SIZE
) {
1796 inline_pos
= (void *)ext4_raw_inode(iloc
)->i_block
;
1797 *inline_size
= EXT4_MIN_INLINE_DATA_SIZE
;
1799 inline_pos
= ext4_get_inline_xattr_pos(inode
, iloc
);
1800 offset
-= EXT4_MIN_INLINE_DATA_SIZE
;
1801 *inline_size
= ext4_get_inline_size(inode
) -
1802 EXT4_MIN_INLINE_DATA_SIZE
;
1806 *inline_start
= inline_pos
;
1807 return (struct ext4_dir_entry_2
*)(inline_pos
+ offset
);
1810 bool empty_inline_dir(struct inode
*dir
, int *has_inline_data
)
1812 int err
, inline_size
;
1813 struct ext4_iloc iloc
;
1816 unsigned int offset
;
1817 struct ext4_dir_entry_2
*de
;
1820 err
= ext4_get_inode_loc(dir
, &iloc
);
1822 EXT4_ERROR_INODE_ERR(dir
, -err
,
1823 "error %d getting inode %lu block",
1828 down_read(&EXT4_I(dir
)->xattr_sem
);
1829 if (!ext4_has_inline_data(dir
)) {
1830 *has_inline_data
= 0;
1835 de
= (struct ext4_dir_entry_2
*)ext4_raw_inode(&iloc
)->i_block
;
1836 if (!le32_to_cpu(de
->inode
)) {
1837 ext4_warning(dir
->i_sb
,
1838 "bad inline directory (dir #%lu) - no `..'",
1843 inline_len
= ext4_get_inline_size(dir
);
1844 offset
= EXT4_INLINE_DOTDOT_SIZE
;
1845 while (offset
< inline_len
) {
1846 de
= ext4_get_inline_entry(dir
, &iloc
, offset
,
1847 &inline_pos
, &inline_size
);
1848 if (ext4_check_dir_entry(dir
, NULL
, de
,
1849 iloc
.bh
, inline_pos
,
1850 inline_size
, offset
)) {
1851 ext4_warning(dir
->i_sb
,
1852 "bad inline directory (dir #%lu) - "
1853 "inode %u, rec_len %u, name_len %d"
1855 dir
->i_ino
, le32_to_cpu(de
->inode
),
1856 le16_to_cpu(de
->rec_len
), de
->name_len
,
1860 if (le32_to_cpu(de
->inode
)) {
1863 offset
+= ext4_rec_len_from_disk(de
->rec_len
, inline_size
);
1868 up_read(&EXT4_I(dir
)->xattr_sem
);
1873 int ext4_destroy_inline_data(handle_t
*handle
, struct inode
*inode
)
1877 ext4_write_lock_xattr(inode
, &no_expand
);
1878 ret
= ext4_destroy_inline_data_nolock(handle
, inode
);
1879 ext4_write_unlock_xattr(inode
, &no_expand
);
1884 int ext4_inline_data_iomap(struct inode
*inode
, struct iomap
*iomap
)
1887 int error
= -EAGAIN
;
1888 struct ext4_iloc iloc
;
1890 down_read(&EXT4_I(inode
)->xattr_sem
);
1891 if (!ext4_has_inline_data(inode
))
1894 error
= ext4_get_inode_loc(inode
, &iloc
);
1898 addr
= (__u64
)iloc
.bh
->b_blocknr
<< inode
->i_sb
->s_blocksize_bits
;
1899 addr
+= (char *)ext4_raw_inode(&iloc
) - iloc
.bh
->b_data
;
1900 addr
+= offsetof(struct ext4_inode
, i_block
);
1906 iomap
->length
= min_t(loff_t
, ext4_get_inline_size(inode
),
1907 i_size_read(inode
));
1908 iomap
->type
= IOMAP_INLINE
;
1912 up_read(&EXT4_I(inode
)->xattr_sem
);
1916 int ext4_inline_data_truncate(struct inode
*inode
, int *has_inline
)
1919 int inline_size
, value_len
, needed_blocks
, no_expand
, err
= 0;
1922 struct ext4_xattr_ibody_find is
= {
1923 .s
= { .not_found
= -ENODATA
, },
1925 struct ext4_xattr_info i
= {
1926 .name_index
= EXT4_XATTR_INDEX_SYSTEM
,
1927 .name
= EXT4_XATTR_SYSTEM_DATA
,
1931 needed_blocks
= ext4_writepage_trans_blocks(inode
);
1932 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, needed_blocks
);
1934 return PTR_ERR(handle
);
1936 ext4_write_lock_xattr(inode
, &no_expand
);
1937 if (!ext4_has_inline_data(inode
)) {
1938 ext4_write_unlock_xattr(inode
, &no_expand
);
1940 ext4_journal_stop(handle
);
1944 if ((err
= ext4_orphan_add(handle
, inode
)) != 0)
1947 if ((err
= ext4_get_inode_loc(inode
, &is
.iloc
)) != 0)
1950 down_write(&EXT4_I(inode
)->i_data_sem
);
1951 i_size
= inode
->i_size
;
1952 inline_size
= ext4_get_inline_size(inode
);
1953 EXT4_I(inode
)->i_disksize
= i_size
;
1955 if (i_size
< inline_size
) {
1957 * if there's inline data to truncate and this file was
1958 * converted to extents after that inline data was written,
1959 * the extent status cache must be cleared to avoid leaving
1960 * behind stale delayed allocated extent entries
1962 if (!ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
))
1963 ext4_es_remove_extent(inode
, 0, EXT_MAX_BLOCKS
);
1965 /* Clear the content in the xattr space. */
1966 if (inline_size
> EXT4_MIN_INLINE_DATA_SIZE
) {
1967 if ((err
= ext4_xattr_ibody_find(inode
, &i
, &is
)) != 0)
1970 BUG_ON(is
.s
.not_found
);
1972 value_len
= le32_to_cpu(is
.s
.here
->e_value_size
);
1973 value
= kmalloc(value_len
, GFP_NOFS
);
1979 err
= ext4_xattr_ibody_get(inode
, i
.name_index
,
1980 i
.name
, value
, value_len
);
1985 i
.value_len
= i_size
> EXT4_MIN_INLINE_DATA_SIZE
?
1986 i_size
- EXT4_MIN_INLINE_DATA_SIZE
: 0;
1987 err
= ext4_xattr_ibody_set(handle
, inode
, &i
, &is
);
1992 /* Clear the content within i_blocks. */
1993 if (i_size
< EXT4_MIN_INLINE_DATA_SIZE
) {
1994 void *p
= (void *) ext4_raw_inode(&is
.iloc
)->i_block
;
1995 memset(p
+ i_size
, 0,
1996 EXT4_MIN_INLINE_DATA_SIZE
- i_size
);
1999 EXT4_I(inode
)->i_inline_size
= i_size
<
2000 EXT4_MIN_INLINE_DATA_SIZE
?
2001 EXT4_MIN_INLINE_DATA_SIZE
: i_size
;
2005 up_write(&EXT4_I(inode
)->i_data_sem
);
2008 ext4_write_unlock_xattr(inode
, &no_expand
);
2011 ext4_orphan_del(handle
, inode
);
2014 inode_set_mtime_to_ts(inode
, inode_set_ctime_current(inode
));
2015 err
= ext4_mark_inode_dirty(handle
, inode
);
2017 ext4_handle_sync(handle
);
2019 ext4_journal_stop(handle
);
2023 int ext4_convert_inline_data(struct inode
*inode
)
2025 int error
, needed_blocks
, no_expand
;
2027 struct ext4_iloc iloc
;
2029 if (!ext4_has_inline_data(inode
)) {
2030 ext4_clear_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
2032 } else if (!ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2034 * Inode has inline data but EXT4_STATE_MAY_INLINE_DATA is
2035 * cleared. This means we are in the middle of moving of
2036 * inline data to delay allocated block. Just force writeout
2037 * here to finish conversion.
2039 error
= filemap_flush(inode
->i_mapping
);
2042 if (!ext4_has_inline_data(inode
))
2046 needed_blocks
= ext4_writepage_trans_blocks(inode
);
2049 error
= ext4_get_inode_loc(inode
, &iloc
);
2053 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
2054 if (IS_ERR(handle
)) {
2055 error
= PTR_ERR(handle
);
2059 ext4_write_lock_xattr(inode
, &no_expand
);
2060 if (ext4_has_inline_data(inode
))
2061 error
= ext4_convert_inline_data_nolock(handle
, inode
, &iloc
);
2062 ext4_write_unlock_xattr(inode
, &no_expand
);
2063 ext4_journal_stop(handle
);