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(EXT4_SB(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 flush_dcache_folio(folio);
507 folio_zero_segment(folio, len, folio_size(folio));
508 folio_mark_uptodate(folio);
515 int ext4_readpage_inline(struct inode *inode, struct folio *folio)
519 down_read(&EXT4_I(inode)->xattr_sem);
520 if (!ext4_has_inline_data(inode)) {
521 up_read(&EXT4_I(inode)->xattr_sem);
526 * Current inline data can only exist in the 1st page,
527 * So for all the other pages, just set them uptodate.
530 ret = ext4_read_inline_folio(inode, folio);
531 else if (!folio_test_uptodate(folio)) {
532 folio_zero_segment(folio, 0, folio_size(folio));
533 folio_mark_uptodate(folio);
536 up_read(&EXT4_I(inode)->xattr_sem);
539 return ret >= 0 ? 0 : ret;
542 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
545 int ret, needed_blocks, no_expand;
546 handle_t *handle = NULL;
547 int retries = 0, sem_held = 0;
548 struct folio *folio = NULL;
550 struct ext4_iloc iloc;
552 if (!ext4_has_inline_data(inode)) {
554 * clear the flag so that no new write
555 * will trap here again.
557 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
561 needed_blocks = ext4_writepage_trans_blocks(inode);
563 ret = ext4_get_inode_loc(inode, &iloc);
568 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
569 if (IS_ERR(handle)) {
570 ret = PTR_ERR(handle);
575 /* We cannot recurse into the filesystem as the transaction is already
577 folio = __filemap_get_folio(mapping, 0, FGP_WRITEBEGIN | FGP_NOFS,
578 mapping_gfp_mask(mapping));
580 ret = PTR_ERR(folio);
584 ext4_write_lock_xattr(inode, &no_expand);
586 /* If some one has already done this for us, just exit. */
587 if (!ext4_has_inline_data(inode)) {
593 to = ext4_get_inline_size(inode);
594 if (!folio_test_uptodate(folio)) {
595 ret = ext4_read_inline_folio(inode, folio);
600 ret = ext4_destroy_inline_data_nolock(handle, inode);
604 if (ext4_should_dioread_nolock(inode)) {
605 ret = __block_write_begin(&folio->page, from, to,
606 ext4_get_block_unwritten);
608 ret = __block_write_begin(&folio->page, from, to, ext4_get_block);
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,
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);
712 *pagep = &folio->page;
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 page *page)
746 struct folio *folio = page_folio(page);
747 handle_t *handle = ext4_journal_current_handle();
750 struct ext4_iloc iloc;
753 if (unlikely(copied < len) && !folio_test_uptodate(folio))
756 if (likely(copied)) {
757 ret = ext4_get_inode_loc(inode, &iloc);
761 ext4_std_error(inode->i_sb, ret);
764 ext4_write_lock_xattr(inode, &no_expand);
765 BUG_ON(!ext4_has_inline_data(inode));
768 * ei->i_inline_off may have changed since
769 * ext4_write_begin() called
770 * ext4_try_to_write_inline_data()
772 (void) ext4_find_inline_data_nolock(inode);
774 kaddr = kmap_local_folio(folio, 0);
775 ext4_write_inline_data(inode, &iloc, kaddr, pos, copied);
777 folio_mark_uptodate(folio);
778 /* clear dirty flag so that writepages wouldn't work for us. */
779 folio_clear_dirty(folio);
781 ext4_write_unlock_xattr(inode, &no_expand);
785 * It's important to update i_size while still holding folio
786 * lock: page writeout could otherwise come in and zero
789 ext4_update_inode_size(inode, pos + copied);
795 * Don't mark the inode dirty under folio lock. First, it unnecessarily
796 * makes the holding time of folio lock longer. Second, it forces lock
797 * ordering of folio lock and transaction start for journaling
801 mark_inode_dirty(inode);
804 * If we didn't copy as much data as expected, we need to trim back
805 * size of xattr containing inline data.
807 if (pos + len > inode->i_size && ext4_can_truncate(inode))
808 ext4_orphan_add(handle, inode);
810 ret2 = ext4_journal_stop(handle);
813 if (pos + len > inode->i_size) {
814 ext4_truncate_failed_write(inode);
816 * If truncate failed early the inode might still be
817 * on the orphan list; we need to make sure the inode
818 * is removed from the orphan list in that case.
821 ext4_orphan_del(NULL, inode);
823 return ret ? ret : copied;
827 ext4_journalled_write_inline_data(struct inode *inode,
833 struct ext4_iloc iloc;
835 ret = ext4_get_inode_loc(inode, &iloc);
837 ext4_std_error(inode->i_sb, ret);
841 ext4_write_lock_xattr(inode, &no_expand);
842 kaddr = kmap_atomic(page);
843 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
844 kunmap_atomic(kaddr);
845 ext4_write_unlock_xattr(inode, &no_expand);
851 * Try to make the page cache and handle ready for the inline data case.
852 * We can call this function in 2 cases:
853 * 1. The inode is created and the first write exceeds inline size. We can
854 * clear the inode state safely.
855 * 2. The inode has inline data, then we need to read the data, make it
856 * update and dirty so that ext4_da_writepages can handle it. We don't
857 * need to start the journal since the file's metadata isn't changed now.
859 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
863 int ret = 0, inline_size;
866 folio = __filemap_get_folio(mapping, 0, FGP_WRITEBEGIN,
867 mapping_gfp_mask(mapping));
869 return PTR_ERR(folio);
871 down_read(&EXT4_I(inode)->xattr_sem);
872 if (!ext4_has_inline_data(inode)) {
873 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
877 inline_size = ext4_get_inline_size(inode);
879 if (!folio_test_uptodate(folio)) {
880 ret = ext4_read_inline_folio(inode, folio);
885 ret = __block_write_begin(&folio->page, 0, inline_size,
886 ext4_da_get_block_prep);
888 up_read(&EXT4_I(inode)->xattr_sem);
891 ext4_truncate_failed_write(inode);
895 folio_mark_dirty(folio);
896 folio_mark_uptodate(folio);
897 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
898 *fsdata = (void *)CONVERT_INLINE_DATA;
901 up_read(&EXT4_I(inode)->xattr_sem);
910 * Prepare the write for the inline data.
911 * If the data can be written into the inode, we just read
912 * the page and make it uptodate, and start the journal.
913 * Otherwise read the page, makes it dirty so that it can be
914 * handle in writepages(the i_disksize update is left to the
915 * normal ext4_da_write_end).
917 int ext4_da_write_inline_data_begin(struct address_space *mapping,
919 loff_t pos, unsigned len,
926 struct ext4_iloc iloc;
929 ret = ext4_get_inode_loc(inode, &iloc);
934 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
935 if (IS_ERR(handle)) {
936 ret = PTR_ERR(handle);
940 ret = ext4_prepare_inline_data(handle, inode, pos + len);
941 if (ret && ret != -ENOSPC)
944 if (ret == -ENOSPC) {
945 ext4_journal_stop(handle);
946 ret = ext4_da_convert_inline_data_to_extent(mapping,
949 if (ret == -ENOSPC &&
950 ext4_should_retry_alloc(inode->i_sb, &retries))
956 * We cannot recurse into the filesystem as the transaction
957 * is already started.
959 folio = __filemap_get_folio(mapping, 0, FGP_WRITEBEGIN | FGP_NOFS,
960 mapping_gfp_mask(mapping));
962 ret = PTR_ERR(folio);
966 down_read(&EXT4_I(inode)->xattr_sem);
967 if (!ext4_has_inline_data(inode)) {
969 goto out_release_page;
972 if (!folio_test_uptodate(folio)) {
973 ret = ext4_read_inline_folio(inode, folio);
975 goto out_release_page;
977 ret = ext4_journal_get_write_access(handle, inode->i_sb, iloc.bh,
980 goto out_release_page;
982 up_read(&EXT4_I(inode)->xattr_sem);
983 *pagep = &folio->page;
987 up_read(&EXT4_I(inode)->xattr_sem);
991 ext4_journal_stop(handle);
997 #ifdef INLINE_DIR_DEBUG
998 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
999 void *inline_start, int inline_size)
1002 unsigned short de_len;
1003 struct ext4_dir_entry_2 *de = inline_start;
1004 void *dlimit = inline_start + inline_size;
1006 trace_printk("inode %lu\n", dir->i_ino);
1008 while ((void *)de < dlimit) {
1009 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
1010 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
1011 offset, de_len, de->name_len, de->name,
1012 de->name_len, le32_to_cpu(de->inode));
1013 if (ext4_check_dir_entry(dir, NULL, de, bh,
1014 inline_start, inline_size, offset))
1018 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1022 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1026 * Add a new entry into a inline dir.
1027 * It will return -ENOSPC if no space is available, and -EIO
1028 * and -EEXIST if directory entry already exists.
1030 static int ext4_add_dirent_to_inline(handle_t *handle,
1031 struct ext4_filename *fname,
1033 struct inode *inode,
1034 struct ext4_iloc *iloc,
1035 void *inline_start, int inline_size)
1038 struct ext4_dir_entry_2 *de;
1040 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1041 inline_size, fname, &de);
1045 BUFFER_TRACE(iloc->bh, "get_write_access");
1046 err = ext4_journal_get_write_access(handle, dir->i_sb, iloc->bh,
1050 ext4_insert_dentry(dir, inode, de, inline_size, fname);
1052 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1055 * XXX shouldn't update any times until successful
1056 * completion of syscall, but too many callers depend
1059 * XXX similarly, too many callers depend on
1060 * ext4_new_inode() setting the times, but error
1061 * recovery deletes the inode, so the worst that can
1062 * happen is that the times are slightly out of date
1063 * and/or different from the directory change time.
1065 dir->i_mtime = dir->i_ctime = current_time(dir);
1066 ext4_update_dx_flag(dir);
1067 inode_inc_iversion(dir);
1071 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1072 struct ext4_iloc *iloc)
1074 struct ext4_xattr_entry *entry;
1075 struct ext4_xattr_ibody_header *header;
1077 BUG_ON(!EXT4_I(inode)->i_inline_off);
1079 header = IHDR(inode, ext4_raw_inode(iloc));
1080 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1081 EXT4_I(inode)->i_inline_off);
1083 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1086 /* Set the final de to cover the whole block. */
1087 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1089 struct ext4_dir_entry_2 *de, *prev_de;
1095 limit = de_buf + old_size;
1098 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1101 } while (de_buf < limit);
1103 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1104 old_size, new_size);
1106 /* this is just created, so create an empty entry. */
1108 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1112 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1113 struct ext4_iloc *iloc)
1116 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1117 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1119 if (new_size - old_size <= ext4_dir_rec_len(1, NULL))
1122 ret = ext4_update_inline_data(handle, dir,
1123 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1127 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1128 EXT4_I(dir)->i_inline_size -
1129 EXT4_MIN_INLINE_DATA_SIZE);
1130 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1134 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1135 struct ext4_iloc *iloc,
1136 void *buf, int inline_size)
1140 ret = ext4_create_inline_data(handle, inode, inline_size);
1142 ext4_msg(inode->i_sb, KERN_EMERG,
1143 "error restoring inline_data for inode -- potential data loss! (inode %lu, error %d)",
1147 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1148 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1151 static int ext4_finish_convert_inline_dir(handle_t *handle,
1152 struct inode *inode,
1153 struct buffer_head *dir_block,
1157 int err, csum_size = 0, header_size = 0;
1158 struct ext4_dir_entry_2 *de;
1159 void *target = dir_block->b_data;
1162 * First create "." and ".." and then copy the dir information
1163 * back to the block.
1166 de = ext4_init_dot_dotdot(inode, de,
1167 inode->i_sb->s_blocksize, csum_size,
1168 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1169 header_size = (void *)de - target;
1171 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1172 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1174 if (ext4_has_metadata_csum(inode->i_sb))
1175 csum_size = sizeof(struct ext4_dir_entry_tail);
1177 inode->i_size = inode->i_sb->s_blocksize;
1178 i_size_write(inode, inode->i_sb->s_blocksize);
1179 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1180 ext4_update_final_de(dir_block->b_data,
1181 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1182 inode->i_sb->s_blocksize - csum_size);
1185 ext4_initialize_dirent_tail(dir_block,
1186 inode->i_sb->s_blocksize);
1187 set_buffer_uptodate(dir_block);
1188 unlock_buffer(dir_block);
1189 err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
1192 set_buffer_verified(dir_block);
1193 return ext4_mark_inode_dirty(handle, inode);
1196 static int ext4_convert_inline_data_nolock(handle_t *handle,
1197 struct inode *inode,
1198 struct ext4_iloc *iloc)
1202 struct buffer_head *data_bh = NULL;
1203 struct ext4_map_blocks map;
1206 inline_size = ext4_get_inline_size(inode);
1207 buf = kmalloc(inline_size, GFP_NOFS);
1213 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1218 * Make sure the inline directory entries pass checks before we try to
1219 * convert them, so that we avoid touching stuff that needs fsck.
1221 if (S_ISDIR(inode->i_mode)) {
1222 error = ext4_check_all_de(inode, iloc->bh,
1223 buf + EXT4_INLINE_DOTDOT_SIZE,
1224 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1229 error = ext4_destroy_inline_data_nolock(handle, inode);
1236 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1239 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1244 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1250 lock_buffer(data_bh);
1251 error = ext4_journal_get_create_access(handle, inode->i_sb, data_bh,
1254 unlock_buffer(data_bh);
1258 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1260 if (!S_ISDIR(inode->i_mode)) {
1261 memcpy(data_bh->b_data, buf, inline_size);
1262 set_buffer_uptodate(data_bh);
1263 unlock_buffer(data_bh);
1264 error = ext4_handle_dirty_metadata(handle,
1267 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1273 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1282 * Try to add the new entry to the inline data.
1283 * If succeeds, return 0. If not, extended the inline dir and copied data to
1284 * the new created block.
1286 int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1287 struct inode *dir, struct inode *inode)
1289 int ret, ret2, inline_size, no_expand;
1291 struct ext4_iloc iloc;
1293 ret = ext4_get_inode_loc(dir, &iloc);
1297 ext4_write_lock_xattr(dir, &no_expand);
1298 if (!ext4_has_inline_data(dir))
1301 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1302 EXT4_INLINE_DOTDOT_SIZE;
1303 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1305 ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1306 inline_start, inline_size);
1310 /* check whether it can be inserted to inline xattr space. */
1311 inline_size = EXT4_I(dir)->i_inline_size -
1312 EXT4_MIN_INLINE_DATA_SIZE;
1314 /* Try to use the xattr space.*/
1315 ret = ext4_update_inline_dir(handle, dir, &iloc);
1316 if (ret && ret != -ENOSPC)
1319 inline_size = EXT4_I(dir)->i_inline_size -
1320 EXT4_MIN_INLINE_DATA_SIZE;
1324 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1326 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1327 inode, &iloc, inline_start,
1335 * The inline space is filled up, so create a new block for it.
1336 * As the extent tree will be created, we have to save the inline
1339 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1342 ext4_write_unlock_xattr(dir, &no_expand);
1343 ret2 = ext4_mark_inode_dirty(handle, dir);
1344 if (unlikely(ret2 && !ret))
1351 * This function fills a red-black tree with information from an
1352 * inlined dir. It returns the number directory entries loaded
1353 * into the tree. If there is an error it is returned in err.
1355 int ext4_inlinedir_to_tree(struct file *dir_file,
1356 struct inode *dir, ext4_lblk_t block,
1357 struct dx_hash_info *hinfo,
1358 __u32 start_hash, __u32 start_minor_hash,
1359 int *has_inline_data)
1361 int err = 0, count = 0;
1362 unsigned int parent_ino;
1364 struct ext4_dir_entry_2 *de;
1365 struct inode *inode = file_inode(dir_file);
1366 int ret, inline_size = 0;
1367 struct ext4_iloc iloc;
1368 void *dir_buf = NULL;
1369 struct ext4_dir_entry_2 fake;
1370 struct fscrypt_str tmp_str;
1372 ret = ext4_get_inode_loc(inode, &iloc);
1376 down_read(&EXT4_I(inode)->xattr_sem);
1377 if (!ext4_has_inline_data(inode)) {
1378 up_read(&EXT4_I(inode)->xattr_sem);
1379 *has_inline_data = 0;
1383 inline_size = ext4_get_inline_size(inode);
1384 dir_buf = kmalloc(inline_size, GFP_NOFS);
1387 up_read(&EXT4_I(inode)->xattr_sem);
1391 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1392 up_read(&EXT4_I(inode)->xattr_sem);
1397 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1398 while (pos < inline_size) {
1400 * As inlined dir doesn't store any information about '.' and
1401 * only the inode number of '..' is stored, we have to handle
1405 fake.inode = cpu_to_le32(inode->i_ino);
1407 strcpy(fake.name, ".");
1408 fake.rec_len = ext4_rec_len_to_disk(
1409 ext4_dir_rec_len(fake.name_len, NULL),
1411 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1413 pos = EXT4_INLINE_DOTDOT_OFFSET;
1414 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1415 fake.inode = cpu_to_le32(parent_ino);
1417 strcpy(fake.name, "..");
1418 fake.rec_len = ext4_rec_len_to_disk(
1419 ext4_dir_rec_len(fake.name_len, NULL),
1421 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1423 pos = EXT4_INLINE_DOTDOT_SIZE;
1425 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1426 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1427 if (ext4_check_dir_entry(inode, dir_file, de,
1429 inline_size, pos)) {
1435 if (ext4_hash_in_dirent(dir)) {
1436 hinfo->hash = EXT4_DIRENT_HASH(de);
1437 hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1439 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1441 if ((hinfo->hash < start_hash) ||
1442 ((hinfo->hash == start_hash) &&
1443 (hinfo->minor_hash < start_minor_hash)))
1447 tmp_str.name = de->name;
1448 tmp_str.len = de->name_len;
1449 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1450 hinfo->minor_hash, de, &tmp_str);
1465 * So this function is called when the volume is mkfsed with
1466 * dir_index disabled. In order to keep f_pos persistent
1467 * after we convert from an inlined dir to a blocked based,
1468 * we just pretend that we are a normal dir and return the
1469 * offset as if '.' and '..' really take place.
1472 int ext4_read_inline_dir(struct file *file,
1473 struct dir_context *ctx,
1474 int *has_inline_data)
1476 unsigned int offset, parent_ino;
1478 struct ext4_dir_entry_2 *de;
1479 struct super_block *sb;
1480 struct inode *inode = file_inode(file);
1481 int ret, inline_size = 0;
1482 struct ext4_iloc iloc;
1483 void *dir_buf = NULL;
1484 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1486 ret = ext4_get_inode_loc(inode, &iloc);
1490 down_read(&EXT4_I(inode)->xattr_sem);
1491 if (!ext4_has_inline_data(inode)) {
1492 up_read(&EXT4_I(inode)->xattr_sem);
1493 *has_inline_data = 0;
1497 inline_size = ext4_get_inline_size(inode);
1498 dir_buf = kmalloc(inline_size, GFP_NOFS);
1501 up_read(&EXT4_I(inode)->xattr_sem);
1505 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1506 up_read(&EXT4_I(inode)->xattr_sem);
1512 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1516 * dotdot_offset and dotdot_size is the real offset and
1517 * size for ".." and "." if the dir is block based while
1518 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1519 * So we will use extra_offset and extra_size to indicate them
1520 * during the inline dir iteration.
1522 dotdot_offset = ext4_dir_rec_len(1, NULL);
1523 dotdot_size = dotdot_offset + ext4_dir_rec_len(2, NULL);
1524 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1525 extra_size = extra_offset + inline_size;
1528 * If the version has changed since the last call to
1529 * readdir(2), then we might be pointing to an invalid
1530 * dirent right now. Scan from the start of the inline
1533 if (!inode_eq_iversion(inode, file->f_version)) {
1534 for (i = 0; i < extra_size && i < offset;) {
1536 * "." is with offset 0 and
1537 * ".." is dotdot_offset.
1542 } else if (i == dotdot_offset) {
1546 /* for other entry, the real offset in
1547 * the buf has to be tuned accordingly.
1549 de = (struct ext4_dir_entry_2 *)
1550 (dir_buf + i - extra_offset);
1551 /* It's too expensive to do a full
1552 * dirent test each time round this
1553 * loop, but we do have to test at
1554 * least that it is non-zero. A
1555 * failure will be detected in the
1556 * dirent test below. */
1557 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1558 < ext4_dir_rec_len(1, NULL))
1560 i += ext4_rec_len_from_disk(de->rec_len,
1565 file->f_version = inode_query_iversion(inode);
1568 while (ctx->pos < extra_size) {
1569 if (ctx->pos == 0) {
1570 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1572 ctx->pos = dotdot_offset;
1576 if (ctx->pos == dotdot_offset) {
1577 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1579 ctx->pos = dotdot_size;
1583 de = (struct ext4_dir_entry_2 *)
1584 (dir_buf + ctx->pos - extra_offset);
1585 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1586 extra_size, ctx->pos))
1588 if (le32_to_cpu(de->inode)) {
1589 if (!dir_emit(ctx, de->name, de->name_len,
1590 le32_to_cpu(de->inode),
1591 get_dtype(sb, de->file_type)))
1594 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1602 void *ext4_read_inline_link(struct inode *inode)
1604 struct ext4_iloc iloc;
1605 int ret, inline_size;
1608 ret = ext4_get_inode_loc(inode, &iloc);
1610 return ERR_PTR(ret);
1613 inline_size = ext4_get_inline_size(inode);
1614 link = kmalloc(inline_size + 1, GFP_NOFS);
1618 ret = ext4_read_inline_data(inode, link, inline_size, &iloc);
1623 nd_terminate_link(link, inode->i_size, ret);
1626 link = ERR_PTR(ret);
1631 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1632 struct ext4_dir_entry_2 **parent_de,
1635 struct ext4_iloc iloc;
1637 *retval = ext4_get_inode_loc(inode, &iloc);
1641 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1647 * Try to create the inline data for the new dir.
1648 * If it succeeds, return 0, otherwise return the error.
1649 * In case of ENOSPC, the caller should create the normal disk layout dir.
1651 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1652 struct inode *inode)
1654 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1655 struct ext4_iloc iloc;
1656 struct ext4_dir_entry_2 *de;
1658 ret = ext4_get_inode_loc(inode, &iloc);
1662 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1667 * For inline dir, we only save the inode information for the ".."
1668 * and create a fake dentry to cover the left space.
1670 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1671 de->inode = cpu_to_le32(parent->i_ino);
1672 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1674 de->rec_len = ext4_rec_len_to_disk(
1675 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1677 set_nlink(inode, 2);
1678 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1684 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1685 struct ext4_filename *fname,
1686 struct ext4_dir_entry_2 **res_dir,
1687 int *has_inline_data)
1690 struct ext4_iloc iloc;
1694 if (ext4_get_inode_loc(dir, &iloc))
1697 down_read(&EXT4_I(dir)->xattr_sem);
1698 if (!ext4_has_inline_data(dir)) {
1699 *has_inline_data = 0;
1703 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1704 EXT4_INLINE_DOTDOT_SIZE;
1705 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1706 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1707 dir, fname, 0, res_dir);
1713 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1716 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1717 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1719 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1720 dir, fname, 0, res_dir);
1728 up_read(&EXT4_I(dir)->xattr_sem);
1732 int ext4_delete_inline_entry(handle_t *handle,
1734 struct ext4_dir_entry_2 *de_del,
1735 struct buffer_head *bh,
1736 int *has_inline_data)
1738 int err, inline_size, no_expand;
1739 struct ext4_iloc iloc;
1742 err = ext4_get_inode_loc(dir, &iloc);
1746 ext4_write_lock_xattr(dir, &no_expand);
1747 if (!ext4_has_inline_data(dir)) {
1748 *has_inline_data = 0;
1752 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1753 EXT4_MIN_INLINE_DATA_SIZE) {
1754 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1755 EXT4_INLINE_DOTDOT_SIZE;
1756 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1757 EXT4_INLINE_DOTDOT_SIZE;
1759 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1760 inline_size = ext4_get_inline_size(dir) -
1761 EXT4_MIN_INLINE_DATA_SIZE;
1764 BUFFER_TRACE(bh, "get_write_access");
1765 err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
1770 err = ext4_generic_delete_entry(dir, de_del, bh,
1771 inline_start, inline_size, 0);
1775 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1777 ext4_write_unlock_xattr(dir, &no_expand);
1778 if (likely(err == 0))
1779 err = ext4_mark_inode_dirty(handle, dir);
1782 ext4_std_error(dir->i_sb, err);
1787 * Get the inline dentry at offset.
1789 static inline struct ext4_dir_entry_2 *
1790 ext4_get_inline_entry(struct inode *inode,
1791 struct ext4_iloc *iloc,
1792 unsigned int offset,
1793 void **inline_start,
1798 BUG_ON(offset > ext4_get_inline_size(inode));
1800 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1801 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1802 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1804 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1805 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1806 *inline_size = ext4_get_inline_size(inode) -
1807 EXT4_MIN_INLINE_DATA_SIZE;
1811 *inline_start = inline_pos;
1812 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1815 bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1817 int err, inline_size;
1818 struct ext4_iloc iloc;
1821 unsigned int offset;
1822 struct ext4_dir_entry_2 *de;
1825 err = ext4_get_inode_loc(dir, &iloc);
1827 EXT4_ERROR_INODE_ERR(dir, -err,
1828 "error %d getting inode %lu block",
1833 down_read(&EXT4_I(dir)->xattr_sem);
1834 if (!ext4_has_inline_data(dir)) {
1835 *has_inline_data = 0;
1840 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1841 if (!le32_to_cpu(de->inode)) {
1842 ext4_warning(dir->i_sb,
1843 "bad inline directory (dir #%lu) - no `..'",
1848 inline_len = ext4_get_inline_size(dir);
1849 offset = EXT4_INLINE_DOTDOT_SIZE;
1850 while (offset < inline_len) {
1851 de = ext4_get_inline_entry(dir, &iloc, offset,
1852 &inline_pos, &inline_size);
1853 if (ext4_check_dir_entry(dir, NULL, de,
1854 iloc.bh, inline_pos,
1855 inline_size, offset)) {
1856 ext4_warning(dir->i_sb,
1857 "bad inline directory (dir #%lu) - "
1858 "inode %u, rec_len %u, name_len %d"
1860 dir->i_ino, le32_to_cpu(de->inode),
1861 le16_to_cpu(de->rec_len), de->name_len,
1865 if (le32_to_cpu(de->inode)) {
1868 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1873 up_read(&EXT4_I(dir)->xattr_sem);
1878 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1882 ext4_write_lock_xattr(inode, &no_expand);
1883 ret = ext4_destroy_inline_data_nolock(handle, inode);
1884 ext4_write_unlock_xattr(inode, &no_expand);
1889 int ext4_inline_data_iomap(struct inode *inode, struct iomap *iomap)
1892 int error = -EAGAIN;
1893 struct ext4_iloc iloc;
1895 down_read(&EXT4_I(inode)->xattr_sem);
1896 if (!ext4_has_inline_data(inode))
1899 error = ext4_get_inode_loc(inode, &iloc);
1903 addr = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1904 addr += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1905 addr += offsetof(struct ext4_inode, i_block);
1911 iomap->length = min_t(loff_t, ext4_get_inline_size(inode),
1912 i_size_read(inode));
1913 iomap->type = IOMAP_INLINE;
1917 up_read(&EXT4_I(inode)->xattr_sem);
1921 int ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1924 int inline_size, value_len, needed_blocks, no_expand, err = 0;
1927 struct ext4_xattr_ibody_find is = {
1928 .s = { .not_found = -ENODATA, },
1930 struct ext4_xattr_info i = {
1931 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1932 .name = EXT4_XATTR_SYSTEM_DATA,
1936 needed_blocks = ext4_writepage_trans_blocks(inode);
1937 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1939 return PTR_ERR(handle);
1941 ext4_write_lock_xattr(inode, &no_expand);
1942 if (!ext4_has_inline_data(inode)) {
1943 ext4_write_unlock_xattr(inode, &no_expand);
1945 ext4_journal_stop(handle);
1949 if ((err = ext4_orphan_add(handle, inode)) != 0)
1952 if ((err = ext4_get_inode_loc(inode, &is.iloc)) != 0)
1955 down_write(&EXT4_I(inode)->i_data_sem);
1956 i_size = inode->i_size;
1957 inline_size = ext4_get_inline_size(inode);
1958 EXT4_I(inode)->i_disksize = i_size;
1960 if (i_size < inline_size) {
1962 * if there's inline data to truncate and this file was
1963 * converted to extents after that inline data was written,
1964 * the extent status cache must be cleared to avoid leaving
1965 * behind stale delayed allocated extent entries
1967 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1969 err = ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1970 if (err == -ENOMEM) {
1971 memalloc_retry_wait(GFP_ATOMIC);
1978 /* Clear the content in the xattr space. */
1979 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1980 if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0)
1983 BUG_ON(is.s.not_found);
1985 value_len = le32_to_cpu(is.s.here->e_value_size);
1986 value = kmalloc(value_len, GFP_NOFS);
1992 err = ext4_xattr_ibody_get(inode, i.name_index,
1993 i.name, value, value_len);
1998 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1999 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
2000 err = ext4_xattr_ibody_set(handle, inode, &i, &is);
2005 /* Clear the content within i_blocks. */
2006 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
2007 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
2008 memset(p + i_size, 0,
2009 EXT4_MIN_INLINE_DATA_SIZE - i_size);
2012 EXT4_I(inode)->i_inline_size = i_size <
2013 EXT4_MIN_INLINE_DATA_SIZE ?
2014 EXT4_MIN_INLINE_DATA_SIZE : i_size;
2018 up_write(&EXT4_I(inode)->i_data_sem);
2021 ext4_write_unlock_xattr(inode, &no_expand);
2024 ext4_orphan_del(handle, inode);
2027 inode->i_mtime = inode->i_ctime = current_time(inode);
2028 err = ext4_mark_inode_dirty(handle, inode);
2030 ext4_handle_sync(handle);
2032 ext4_journal_stop(handle);
2036 int ext4_convert_inline_data(struct inode *inode)
2038 int error, needed_blocks, no_expand;
2040 struct ext4_iloc iloc;
2042 if (!ext4_has_inline_data(inode)) {
2043 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
2045 } else if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2047 * Inode has inline data but EXT4_STATE_MAY_INLINE_DATA is
2048 * cleared. This means we are in the middle of moving of
2049 * inline data to delay allocated block. Just force writeout
2050 * here to finish conversion.
2052 error = filemap_flush(inode->i_mapping);
2055 if (!ext4_has_inline_data(inode))
2059 needed_blocks = ext4_writepage_trans_blocks(inode);
2062 error = ext4_get_inode_loc(inode, &iloc);
2066 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2067 if (IS_ERR(handle)) {
2068 error = PTR_ERR(handle);
2072 ext4_write_lock_xattr(inode, &no_expand);
2073 if (ext4_has_inline_data(inode))
2074 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2075 ext4_write_unlock_xattr(inode, &no_expand);
2076 ext4_journal_stop(handle);