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;
39 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
42 min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
43 EXT4_GOOD_OLD_INODE_SIZE -
44 EXT4_I(inode)->i_extra_isize -
45 sizeof(struct ext4_xattr_ibody_header);
48 * We need to subtract another sizeof(__u32) since an in-inode xattr
49 * needs an empty 4 bytes to indicate the gap between the xattr entry
50 * and the name/value pair.
52 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
53 return EXT4_XATTR_SIZE(min_offs -
54 EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
55 EXT4_XATTR_ROUND - sizeof(__u32));
57 raw_inode = ext4_raw_inode(iloc);
58 header = IHDR(inode, raw_inode);
59 entry = IFIRST(header);
61 /* Compute min_offs. */
62 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
63 if (!entry->e_value_inum && entry->e_value_size) {
64 size_t offs = le16_to_cpu(entry->e_value_offs);
70 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
72 if (EXT4_I(inode)->i_inline_off) {
73 entry = (struct ext4_xattr_entry *)
74 ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
76 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
80 free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
82 if (free > EXT4_XATTR_ROUND)
83 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
92 * Get the maximum size we now can store in an inode.
93 * If we can't find the space for a xattr entry, don't use the space
94 * of the extents since we have no space to indicate the inline data.
96 int ext4_get_max_inline_size(struct inode *inode)
98 int error, max_inline_size;
99 struct ext4_iloc iloc;
101 if (EXT4_I(inode)->i_extra_isize == 0)
104 error = ext4_get_inode_loc(inode, &iloc);
106 ext4_error_inode_err(inode, __func__, __LINE__, 0, -error,
107 "can't get inode location %lu",
112 down_read(&EXT4_I(inode)->xattr_sem);
113 max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
114 up_read(&EXT4_I(inode)->xattr_sem);
118 if (!max_inline_size)
121 return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
125 * this function does not take xattr_sem, which is OK because it is
126 * currently only used in a code path coming form ext4_iget, before
127 * the new inode has been unlocked
129 int ext4_find_inline_data_nolock(struct inode *inode)
131 struct ext4_xattr_ibody_find is = {
132 .s = { .not_found = -ENODATA, },
134 struct ext4_xattr_info i = {
135 .name_index = EXT4_XATTR_INDEX_SYSTEM,
136 .name = EXT4_XATTR_SYSTEM_DATA,
140 if (EXT4_I(inode)->i_extra_isize == 0)
143 error = ext4_get_inode_loc(inode, &is.iloc);
147 error = ext4_xattr_ibody_find(inode, &i, &is);
151 if (!is.s.not_found) {
152 if (is.s.here->e_value_inum) {
153 EXT4_ERROR_INODE(inode, "inline data xattr refers "
154 "to an external xattr inode");
155 error = -EFSCORRUPTED;
158 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
159 (void *)ext4_raw_inode(&is.iloc));
160 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
161 le32_to_cpu(is.s.here->e_value_size);
168 static int ext4_read_inline_data(struct inode *inode, void *buffer,
170 struct ext4_iloc *iloc)
172 struct ext4_xattr_entry *entry;
173 struct ext4_xattr_ibody_header *header;
175 struct ext4_inode *raw_inode;
180 BUG_ON(len > EXT4_I(inode)->i_inline_size);
182 cp_len = min_t(unsigned int, len, EXT4_MIN_INLINE_DATA_SIZE);
184 raw_inode = ext4_raw_inode(iloc);
185 memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
193 header = IHDR(inode, raw_inode);
194 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
195 EXT4_I(inode)->i_inline_off);
196 len = min_t(unsigned int, len,
197 (unsigned int)le32_to_cpu(entry->e_value_size));
200 (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
208 * write the buffer to the inline inode.
209 * If 'create' is set, we don't need to do the extra copy in the xattr
210 * value since it is already handled by ext4_xattr_ibody_set.
211 * That saves us one memcpy.
213 static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
214 void *buffer, loff_t pos, unsigned int len)
216 struct ext4_xattr_entry *entry;
217 struct ext4_xattr_ibody_header *header;
218 struct ext4_inode *raw_inode;
221 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
224 BUG_ON(!EXT4_I(inode)->i_inline_off);
225 BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
227 raw_inode = ext4_raw_inode(iloc);
230 if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
231 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
232 EXT4_MIN_INLINE_DATA_SIZE - pos : len;
233 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
243 pos -= EXT4_MIN_INLINE_DATA_SIZE;
244 header = IHDR(inode, raw_inode);
245 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
246 EXT4_I(inode)->i_inline_off);
248 memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
252 static int ext4_create_inline_data(handle_t *handle,
253 struct inode *inode, unsigned len)
257 struct ext4_xattr_ibody_find is = {
258 .s = { .not_found = -ENODATA, },
260 struct ext4_xattr_info i = {
261 .name_index = EXT4_XATTR_INDEX_SYSTEM,
262 .name = EXT4_XATTR_SYSTEM_DATA,
265 error = ext4_get_inode_loc(inode, &is.iloc);
269 BUFFER_TRACE(is.iloc.bh, "get_write_access");
270 error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh,
275 if (len > EXT4_MIN_INLINE_DATA_SIZE) {
276 value = EXT4_ZERO_XATTR_VALUE;
277 len -= EXT4_MIN_INLINE_DATA_SIZE;
283 /* Insert the xttr entry. */
287 error = ext4_xattr_ibody_find(inode, &i, &is);
291 BUG_ON(!is.s.not_found);
293 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
295 if (error == -ENOSPC)
296 ext4_clear_inode_state(inode,
297 EXT4_STATE_MAY_INLINE_DATA);
301 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
302 0, EXT4_MIN_INLINE_DATA_SIZE);
304 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
305 (void *)ext4_raw_inode(&is.iloc));
306 EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
307 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
308 ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
310 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
317 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
322 struct ext4_xattr_ibody_find is = {
323 .s = { .not_found = -ENODATA, },
325 struct ext4_xattr_info i = {
326 .name_index = EXT4_XATTR_INDEX_SYSTEM,
327 .name = EXT4_XATTR_SYSTEM_DATA,
330 /* If the old space is ok, write the data directly. */
331 if (len <= EXT4_I(inode)->i_inline_size)
334 error = ext4_get_inode_loc(inode, &is.iloc);
338 error = ext4_xattr_ibody_find(inode, &i, &is);
342 BUG_ON(is.s.not_found);
344 len -= EXT4_MIN_INLINE_DATA_SIZE;
345 value = kzalloc(len, GFP_NOFS);
351 error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
353 if (error == -ENODATA)
356 BUFFER_TRACE(is.iloc.bh, "get_write_access");
357 error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh,
362 /* Update the xattr entry. */
366 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
370 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
371 (void *)ext4_raw_inode(&is.iloc));
372 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
373 le32_to_cpu(is.s.here->e_value_size);
374 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
376 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
384 static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
387 int ret, size, no_expand;
388 struct ext4_inode_info *ei = EXT4_I(inode);
390 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
393 size = ext4_get_max_inline_size(inode);
397 ext4_write_lock_xattr(inode, &no_expand);
399 if (ei->i_inline_off)
400 ret = ext4_update_inline_data(handle, inode, len);
402 ret = ext4_create_inline_data(handle, inode, len);
404 ext4_write_unlock_xattr(inode, &no_expand);
408 static int ext4_destroy_inline_data_nolock(handle_t *handle,
411 struct ext4_inode_info *ei = EXT4_I(inode);
412 struct ext4_xattr_ibody_find is = {
413 .s = { .not_found = 0, },
415 struct ext4_xattr_info i = {
416 .name_index = EXT4_XATTR_INDEX_SYSTEM,
417 .name = EXT4_XATTR_SYSTEM_DATA,
423 if (!ei->i_inline_off)
426 error = ext4_get_inode_loc(inode, &is.iloc);
430 error = ext4_xattr_ibody_find(inode, &i, &is);
434 BUFFER_TRACE(is.iloc.bh, "get_write_access");
435 error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh,
440 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
444 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
445 0, EXT4_MIN_INLINE_DATA_SIZE);
446 memset(ei->i_data, 0, EXT4_MIN_INLINE_DATA_SIZE);
448 if (ext4_has_feature_extents(inode->i_sb)) {
449 if (S_ISDIR(inode->i_mode) ||
450 S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
451 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
452 ext4_ext_tree_init(handle, inode);
455 ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
458 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
460 EXT4_I(inode)->i_inline_off = 0;
461 EXT4_I(inode)->i_inline_size = 0;
462 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
465 if (error == -ENODATA)
470 static int ext4_read_inline_folio(struct inode *inode, struct folio *folio)
475 struct ext4_iloc iloc;
477 BUG_ON(!folio_test_locked(folio));
478 BUG_ON(!ext4_has_inline_data(inode));
479 BUG_ON(folio->index);
481 if (!EXT4_I(inode)->i_inline_off) {
482 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
487 ret = ext4_get_inode_loc(inode, &iloc);
491 len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
492 BUG_ON(len > PAGE_SIZE);
493 kaddr = kmap_local_folio(folio, 0);
494 ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
495 flush_dcache_folio(folio);
497 folio_zero_segment(folio, len, folio_size(folio));
498 folio_mark_uptodate(folio);
505 int ext4_readpage_inline(struct inode *inode, struct folio *folio)
509 down_read(&EXT4_I(inode)->xattr_sem);
510 if (!ext4_has_inline_data(inode)) {
511 up_read(&EXT4_I(inode)->xattr_sem);
516 * Current inline data can only exist in the 1st page,
517 * So for all the other pages, just set them uptodate.
520 ret = ext4_read_inline_folio(inode, folio);
521 else if (!folio_test_uptodate(folio)) {
522 folio_zero_segment(folio, 0, folio_size(folio));
523 folio_mark_uptodate(folio);
526 up_read(&EXT4_I(inode)->xattr_sem);
529 return ret >= 0 ? 0 : ret;
532 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
535 int ret, needed_blocks, no_expand;
536 handle_t *handle = NULL;
537 int retries = 0, sem_held = 0;
538 struct folio *folio = NULL;
540 struct ext4_iloc iloc;
542 if (!ext4_has_inline_data(inode)) {
544 * clear the flag so that no new write
545 * will trap here again.
547 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
551 needed_blocks = ext4_writepage_trans_blocks(inode);
553 ret = ext4_get_inode_loc(inode, &iloc);
558 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
559 if (IS_ERR(handle)) {
560 ret = PTR_ERR(handle);
565 /* We cannot recurse into the filesystem as the transaction is already
567 folio = __filemap_get_folio(mapping, 0, FGP_WRITEBEGIN | FGP_NOFS,
568 mapping_gfp_mask(mapping));
570 ret = PTR_ERR(folio);
574 ext4_write_lock_xattr(inode, &no_expand);
576 /* If some one has already done this for us, just exit. */
577 if (!ext4_has_inline_data(inode)) {
583 to = ext4_get_inline_size(inode);
584 if (!folio_test_uptodate(folio)) {
585 ret = ext4_read_inline_folio(inode, folio);
590 ret = ext4_destroy_inline_data_nolock(handle, inode);
594 if (ext4_should_dioread_nolock(inode)) {
595 ret = __block_write_begin(&folio->page, from, to,
596 ext4_get_block_unwritten);
598 ret = __block_write_begin(&folio->page, from, to, ext4_get_block);
600 if (!ret && ext4_should_journal_data(inode)) {
601 ret = ext4_walk_page_buffers(handle, inode,
602 folio_buffers(folio), from, to,
603 NULL, do_journal_get_write_access);
610 ext4_orphan_add(handle, inode);
611 ext4_write_unlock_xattr(inode, &no_expand);
613 ext4_journal_stop(handle);
615 ext4_truncate_failed_write(inode);
617 * If truncate failed early the inode might
618 * still be on the orphan list; we need to
619 * make sure the inode is removed from the
620 * orphan list in that case.
623 ext4_orphan_del(NULL, inode);
626 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
630 block_commit_write(&folio->page, from, to);
638 ext4_write_unlock_xattr(inode, &no_expand);
640 ext4_journal_stop(handle);
646 * Try to write data in the inode.
647 * If the inode has inline data, check whether the new write can be
648 * in the inode also. If not, create the page the handle, move the data
649 * to the page make it update and let the later codes create extent for it.
651 int ext4_try_to_write_inline_data(struct address_space *mapping,
653 loff_t pos, unsigned len,
659 struct ext4_iloc iloc;
661 if (pos + len > ext4_get_max_inline_size(inode))
664 ret = ext4_get_inode_loc(inode, &iloc);
669 * The possible write could happen in the inode,
670 * so try to reserve the space in inode first.
672 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
673 if (IS_ERR(handle)) {
674 ret = PTR_ERR(handle);
679 ret = ext4_prepare_inline_data(handle, inode, pos + len);
680 if (ret && ret != -ENOSPC)
683 /* We don't have space in inline inode, so convert it to extent. */
684 if (ret == -ENOSPC) {
685 ext4_journal_stop(handle);
690 ret = ext4_journal_get_write_access(handle, inode->i_sb, iloc.bh,
695 folio = __filemap_get_folio(mapping, 0, FGP_WRITEBEGIN | FGP_NOFS,
696 mapping_gfp_mask(mapping));
698 ret = PTR_ERR(folio);
702 *pagep = &folio->page;
703 down_read(&EXT4_I(inode)->xattr_sem);
704 if (!ext4_has_inline_data(inode)) {
711 if (!folio_test_uptodate(folio)) {
712 ret = ext4_read_inline_folio(inode, folio);
723 up_read(&EXT4_I(inode)->xattr_sem);
725 if (handle && (ret != 1))
726 ext4_journal_stop(handle);
730 return ext4_convert_inline_data_to_extent(mapping, inode);
733 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
734 unsigned copied, struct page *page)
736 struct folio *folio = page_folio(page);
737 handle_t *handle = ext4_journal_current_handle();
740 struct ext4_iloc iloc;
743 if (unlikely(copied < len) && !folio_test_uptodate(folio))
746 if (likely(copied)) {
747 ret = ext4_get_inode_loc(inode, &iloc);
751 ext4_std_error(inode->i_sb, ret);
754 ext4_write_lock_xattr(inode, &no_expand);
755 BUG_ON(!ext4_has_inline_data(inode));
758 * ei->i_inline_off may have changed since
759 * ext4_write_begin() called
760 * ext4_try_to_write_inline_data()
762 (void) ext4_find_inline_data_nolock(inode);
764 kaddr = kmap_local_folio(folio, 0);
765 ext4_write_inline_data(inode, &iloc, kaddr, pos, copied);
767 folio_mark_uptodate(folio);
768 /* clear dirty flag so that writepages wouldn't work for us. */
769 folio_clear_dirty(folio);
771 ext4_write_unlock_xattr(inode, &no_expand);
775 * It's important to update i_size while still holding folio
776 * lock: page writeout could otherwise come in and zero
779 ext4_update_inode_size(inode, pos + copied);
785 * Don't mark the inode dirty under folio lock. First, it unnecessarily
786 * makes the holding time of folio lock longer. Second, it forces lock
787 * ordering of folio lock and transaction start for journaling
791 mark_inode_dirty(inode);
794 * If we didn't copy as much data as expected, we need to trim back
795 * size of xattr containing inline data.
797 if (pos + len > inode->i_size && ext4_can_truncate(inode))
798 ext4_orphan_add(handle, inode);
800 ret2 = ext4_journal_stop(handle);
803 if (pos + len > inode->i_size) {
804 ext4_truncate_failed_write(inode);
806 * If truncate failed early the inode might still be
807 * on the orphan list; we need to make sure the inode
808 * is removed from the orphan list in that case.
811 ext4_orphan_del(NULL, inode);
813 return ret ? ret : copied;
817 ext4_journalled_write_inline_data(struct inode *inode,
823 struct ext4_iloc iloc;
825 ret = ext4_get_inode_loc(inode, &iloc);
827 ext4_std_error(inode->i_sb, ret);
831 ext4_write_lock_xattr(inode, &no_expand);
832 kaddr = kmap_atomic(page);
833 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
834 kunmap_atomic(kaddr);
835 ext4_write_unlock_xattr(inode, &no_expand);
841 * Try to make the page cache and handle ready for the inline data case.
842 * We can call this function in 2 cases:
843 * 1. The inode is created and the first write exceeds inline size. We can
844 * clear the inode state safely.
845 * 2. The inode has inline data, then we need to read the data, make it
846 * update and dirty so that ext4_da_writepages can handle it. We don't
847 * need to start the journal since the file's metadata isn't changed now.
849 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
853 int ret = 0, inline_size;
856 folio = __filemap_get_folio(mapping, 0, FGP_WRITEBEGIN,
857 mapping_gfp_mask(mapping));
859 return PTR_ERR(folio);
861 down_read(&EXT4_I(inode)->xattr_sem);
862 if (!ext4_has_inline_data(inode)) {
863 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
867 inline_size = ext4_get_inline_size(inode);
869 if (!folio_test_uptodate(folio)) {
870 ret = ext4_read_inline_folio(inode, folio);
875 ret = __block_write_begin(&folio->page, 0, inline_size,
876 ext4_da_get_block_prep);
878 up_read(&EXT4_I(inode)->xattr_sem);
881 ext4_truncate_failed_write(inode);
885 folio_mark_dirty(folio);
886 folio_mark_uptodate(folio);
887 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
888 *fsdata = (void *)CONVERT_INLINE_DATA;
891 up_read(&EXT4_I(inode)->xattr_sem);
900 * Prepare the write for the inline data.
901 * If the data can be written into the inode, we just read
902 * the page and make it uptodate, and start the journal.
903 * Otherwise read the page, makes it dirty so that it can be
904 * handle in writepages(the i_disksize update is left to the
905 * normal ext4_da_write_end).
907 int ext4_da_write_inline_data_begin(struct address_space *mapping,
909 loff_t pos, unsigned len,
916 struct ext4_iloc iloc;
919 ret = ext4_get_inode_loc(inode, &iloc);
924 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
925 if (IS_ERR(handle)) {
926 ret = PTR_ERR(handle);
930 ret = ext4_prepare_inline_data(handle, inode, pos + len);
931 if (ret && ret != -ENOSPC)
934 if (ret == -ENOSPC) {
935 ext4_journal_stop(handle);
936 ret = ext4_da_convert_inline_data_to_extent(mapping,
939 if (ret == -ENOSPC &&
940 ext4_should_retry_alloc(inode->i_sb, &retries))
946 * We cannot recurse into the filesystem as the transaction
947 * is already started.
949 folio = __filemap_get_folio(mapping, 0, FGP_WRITEBEGIN | FGP_NOFS,
950 mapping_gfp_mask(mapping));
952 ret = PTR_ERR(folio);
956 down_read(&EXT4_I(inode)->xattr_sem);
957 if (!ext4_has_inline_data(inode)) {
959 goto out_release_page;
962 if (!folio_test_uptodate(folio)) {
963 ret = ext4_read_inline_folio(inode, folio);
965 goto out_release_page;
967 ret = ext4_journal_get_write_access(handle, inode->i_sb, iloc.bh,
970 goto out_release_page;
972 up_read(&EXT4_I(inode)->xattr_sem);
973 *pagep = &folio->page;
977 up_read(&EXT4_I(inode)->xattr_sem);
981 ext4_journal_stop(handle);
987 #ifdef INLINE_DIR_DEBUG
988 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
989 void *inline_start, int inline_size)
992 unsigned short de_len;
993 struct ext4_dir_entry_2 *de = inline_start;
994 void *dlimit = inline_start + inline_size;
996 trace_printk("inode %lu\n", dir->i_ino);
998 while ((void *)de < dlimit) {
999 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
1000 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
1001 offset, de_len, de->name_len, de->name,
1002 de->name_len, le32_to_cpu(de->inode));
1003 if (ext4_check_dir_entry(dir, NULL, de, bh,
1004 inline_start, inline_size, offset))
1008 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1012 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1016 * Add a new entry into a inline dir.
1017 * It will return -ENOSPC if no space is available, and -EIO
1018 * and -EEXIST if directory entry already exists.
1020 static int ext4_add_dirent_to_inline(handle_t *handle,
1021 struct ext4_filename *fname,
1023 struct inode *inode,
1024 struct ext4_iloc *iloc,
1025 void *inline_start, int inline_size)
1028 struct ext4_dir_entry_2 *de;
1030 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1031 inline_size, fname, &de);
1035 BUFFER_TRACE(iloc->bh, "get_write_access");
1036 err = ext4_journal_get_write_access(handle, dir->i_sb, iloc->bh,
1040 ext4_insert_dentry(dir, inode, de, inline_size, fname);
1042 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1045 * XXX shouldn't update any times until successful
1046 * completion of syscall, but too many callers depend
1049 * XXX similarly, too many callers depend on
1050 * ext4_new_inode() setting the times, but error
1051 * recovery deletes the inode, so the worst that can
1052 * happen is that the times are slightly out of date
1053 * and/or different from the directory change time.
1055 dir->i_mtime = dir->i_ctime = current_time(dir);
1056 ext4_update_dx_flag(dir);
1057 inode_inc_iversion(dir);
1061 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1062 struct ext4_iloc *iloc)
1064 struct ext4_xattr_entry *entry;
1065 struct ext4_xattr_ibody_header *header;
1067 BUG_ON(!EXT4_I(inode)->i_inline_off);
1069 header = IHDR(inode, ext4_raw_inode(iloc));
1070 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1071 EXT4_I(inode)->i_inline_off);
1073 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1076 /* Set the final de to cover the whole block. */
1077 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1079 struct ext4_dir_entry_2 *de, *prev_de;
1085 limit = de_buf + old_size;
1088 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1091 } while (de_buf < limit);
1093 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1094 old_size, new_size);
1096 /* this is just created, so create an empty entry. */
1098 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1102 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1103 struct ext4_iloc *iloc)
1106 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1107 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1109 if (new_size - old_size <= ext4_dir_rec_len(1, NULL))
1112 ret = ext4_update_inline_data(handle, dir,
1113 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1117 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1118 EXT4_I(dir)->i_inline_size -
1119 EXT4_MIN_INLINE_DATA_SIZE);
1120 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1124 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1125 struct ext4_iloc *iloc,
1126 void *buf, int inline_size)
1130 ret = ext4_create_inline_data(handle, inode, inline_size);
1132 ext4_msg(inode->i_sb, KERN_EMERG,
1133 "error restoring inline_data for inode -- potential data loss! (inode %lu, error %d)",
1137 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1138 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1141 static int ext4_finish_convert_inline_dir(handle_t *handle,
1142 struct inode *inode,
1143 struct buffer_head *dir_block,
1147 int err, csum_size = 0, header_size = 0;
1148 struct ext4_dir_entry_2 *de;
1149 void *target = dir_block->b_data;
1152 * First create "." and ".." and then copy the dir information
1153 * back to the block.
1156 de = ext4_init_dot_dotdot(inode, de,
1157 inode->i_sb->s_blocksize, csum_size,
1158 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1159 header_size = (void *)de - target;
1161 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1162 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1164 if (ext4_has_metadata_csum(inode->i_sb))
1165 csum_size = sizeof(struct ext4_dir_entry_tail);
1167 inode->i_size = inode->i_sb->s_blocksize;
1168 i_size_write(inode, inode->i_sb->s_blocksize);
1169 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1170 ext4_update_final_de(dir_block->b_data,
1171 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1172 inode->i_sb->s_blocksize - csum_size);
1175 ext4_initialize_dirent_tail(dir_block,
1176 inode->i_sb->s_blocksize);
1177 set_buffer_uptodate(dir_block);
1178 err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
1181 set_buffer_verified(dir_block);
1182 return ext4_mark_inode_dirty(handle, inode);
1185 static int ext4_convert_inline_data_nolock(handle_t *handle,
1186 struct inode *inode,
1187 struct ext4_iloc *iloc)
1191 struct buffer_head *data_bh = NULL;
1192 struct ext4_map_blocks map;
1195 inline_size = ext4_get_inline_size(inode);
1196 buf = kmalloc(inline_size, GFP_NOFS);
1202 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1207 * Make sure the inline directory entries pass checks before we try to
1208 * convert them, so that we avoid touching stuff that needs fsck.
1210 if (S_ISDIR(inode->i_mode)) {
1211 error = ext4_check_all_de(inode, iloc->bh,
1212 buf + EXT4_INLINE_DOTDOT_SIZE,
1213 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1218 error = ext4_destroy_inline_data_nolock(handle, inode);
1225 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1228 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1233 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1239 lock_buffer(data_bh);
1240 error = ext4_journal_get_create_access(handle, inode->i_sb, data_bh,
1243 unlock_buffer(data_bh);
1247 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1249 if (!S_ISDIR(inode->i_mode)) {
1250 memcpy(data_bh->b_data, buf, inline_size);
1251 set_buffer_uptodate(data_bh);
1252 error = ext4_handle_dirty_metadata(handle,
1255 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1259 unlock_buffer(data_bh);
1262 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1271 * Try to add the new entry to the inline data.
1272 * If succeeds, return 0. If not, extended the inline dir and copied data to
1273 * the new created block.
1275 int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1276 struct inode *dir, struct inode *inode)
1278 int ret, ret2, inline_size, no_expand;
1280 struct ext4_iloc iloc;
1282 ret = ext4_get_inode_loc(dir, &iloc);
1286 ext4_write_lock_xattr(dir, &no_expand);
1287 if (!ext4_has_inline_data(dir))
1290 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1291 EXT4_INLINE_DOTDOT_SIZE;
1292 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1294 ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1295 inline_start, inline_size);
1299 /* check whether it can be inserted to inline xattr space. */
1300 inline_size = EXT4_I(dir)->i_inline_size -
1301 EXT4_MIN_INLINE_DATA_SIZE;
1303 /* Try to use the xattr space.*/
1304 ret = ext4_update_inline_dir(handle, dir, &iloc);
1305 if (ret && ret != -ENOSPC)
1308 inline_size = EXT4_I(dir)->i_inline_size -
1309 EXT4_MIN_INLINE_DATA_SIZE;
1313 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1315 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1316 inode, &iloc, inline_start,
1324 * The inline space is filled up, so create a new block for it.
1325 * As the extent tree will be created, we have to save the inline
1328 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1331 ext4_write_unlock_xattr(dir, &no_expand);
1332 ret2 = ext4_mark_inode_dirty(handle, dir);
1333 if (unlikely(ret2 && !ret))
1340 * This function fills a red-black tree with information from an
1341 * inlined dir. It returns the number directory entries loaded
1342 * into the tree. If there is an error it is returned in err.
1344 int ext4_inlinedir_to_tree(struct file *dir_file,
1345 struct inode *dir, ext4_lblk_t block,
1346 struct dx_hash_info *hinfo,
1347 __u32 start_hash, __u32 start_minor_hash,
1348 int *has_inline_data)
1350 int err = 0, count = 0;
1351 unsigned int parent_ino;
1353 struct ext4_dir_entry_2 *de;
1354 struct inode *inode = file_inode(dir_file);
1355 int ret, inline_size = 0;
1356 struct ext4_iloc iloc;
1357 void *dir_buf = NULL;
1358 struct ext4_dir_entry_2 fake;
1359 struct fscrypt_str tmp_str;
1361 ret = ext4_get_inode_loc(inode, &iloc);
1365 down_read(&EXT4_I(inode)->xattr_sem);
1366 if (!ext4_has_inline_data(inode)) {
1367 up_read(&EXT4_I(inode)->xattr_sem);
1368 *has_inline_data = 0;
1372 inline_size = ext4_get_inline_size(inode);
1373 dir_buf = kmalloc(inline_size, GFP_NOFS);
1376 up_read(&EXT4_I(inode)->xattr_sem);
1380 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1381 up_read(&EXT4_I(inode)->xattr_sem);
1386 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1387 while (pos < inline_size) {
1389 * As inlined dir doesn't store any information about '.' and
1390 * only the inode number of '..' is stored, we have to handle
1394 fake.inode = cpu_to_le32(inode->i_ino);
1396 strcpy(fake.name, ".");
1397 fake.rec_len = ext4_rec_len_to_disk(
1398 ext4_dir_rec_len(fake.name_len, NULL),
1400 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1402 pos = EXT4_INLINE_DOTDOT_OFFSET;
1403 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1404 fake.inode = cpu_to_le32(parent_ino);
1406 strcpy(fake.name, "..");
1407 fake.rec_len = ext4_rec_len_to_disk(
1408 ext4_dir_rec_len(fake.name_len, NULL),
1410 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1412 pos = EXT4_INLINE_DOTDOT_SIZE;
1414 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1415 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1416 if (ext4_check_dir_entry(inode, dir_file, de,
1418 inline_size, pos)) {
1424 if (ext4_hash_in_dirent(dir)) {
1425 hinfo->hash = EXT4_DIRENT_HASH(de);
1426 hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1428 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1430 if ((hinfo->hash < start_hash) ||
1431 ((hinfo->hash == start_hash) &&
1432 (hinfo->minor_hash < start_minor_hash)))
1436 tmp_str.name = de->name;
1437 tmp_str.len = de->name_len;
1438 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1439 hinfo->minor_hash, de, &tmp_str);
1454 * So this function is called when the volume is mkfsed with
1455 * dir_index disabled. In order to keep f_pos persistent
1456 * after we convert from an inlined dir to a blocked based,
1457 * we just pretend that we are a normal dir and return the
1458 * offset as if '.' and '..' really take place.
1461 int ext4_read_inline_dir(struct file *file,
1462 struct dir_context *ctx,
1463 int *has_inline_data)
1465 unsigned int offset, parent_ino;
1467 struct ext4_dir_entry_2 *de;
1468 struct super_block *sb;
1469 struct inode *inode = file_inode(file);
1470 int ret, inline_size = 0;
1471 struct ext4_iloc iloc;
1472 void *dir_buf = NULL;
1473 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1475 ret = ext4_get_inode_loc(inode, &iloc);
1479 down_read(&EXT4_I(inode)->xattr_sem);
1480 if (!ext4_has_inline_data(inode)) {
1481 up_read(&EXT4_I(inode)->xattr_sem);
1482 *has_inline_data = 0;
1486 inline_size = ext4_get_inline_size(inode);
1487 dir_buf = kmalloc(inline_size, GFP_NOFS);
1490 up_read(&EXT4_I(inode)->xattr_sem);
1494 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1495 up_read(&EXT4_I(inode)->xattr_sem);
1501 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1505 * dotdot_offset and dotdot_size is the real offset and
1506 * size for ".." and "." if the dir is block based while
1507 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1508 * So we will use extra_offset and extra_size to indicate them
1509 * during the inline dir iteration.
1511 dotdot_offset = ext4_dir_rec_len(1, NULL);
1512 dotdot_size = dotdot_offset + ext4_dir_rec_len(2, NULL);
1513 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1514 extra_size = extra_offset + inline_size;
1517 * If the version has changed since the last call to
1518 * readdir(2), then we might be pointing to an invalid
1519 * dirent right now. Scan from the start of the inline
1522 if (!inode_eq_iversion(inode, file->f_version)) {
1523 for (i = 0; i < extra_size && i < offset;) {
1525 * "." is with offset 0 and
1526 * ".." is dotdot_offset.
1531 } else if (i == dotdot_offset) {
1535 /* for other entry, the real offset in
1536 * the buf has to be tuned accordingly.
1538 de = (struct ext4_dir_entry_2 *)
1539 (dir_buf + i - extra_offset);
1540 /* It's too expensive to do a full
1541 * dirent test each time round this
1542 * loop, but we do have to test at
1543 * least that it is non-zero. A
1544 * failure will be detected in the
1545 * dirent test below. */
1546 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1547 < ext4_dir_rec_len(1, NULL))
1549 i += ext4_rec_len_from_disk(de->rec_len,
1554 file->f_version = inode_query_iversion(inode);
1557 while (ctx->pos < extra_size) {
1558 if (ctx->pos == 0) {
1559 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1561 ctx->pos = dotdot_offset;
1565 if (ctx->pos == dotdot_offset) {
1566 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1568 ctx->pos = dotdot_size;
1572 de = (struct ext4_dir_entry_2 *)
1573 (dir_buf + ctx->pos - extra_offset);
1574 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1575 extra_size, ctx->pos))
1577 if (le32_to_cpu(de->inode)) {
1578 if (!dir_emit(ctx, de->name, de->name_len,
1579 le32_to_cpu(de->inode),
1580 get_dtype(sb, de->file_type)))
1583 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1591 void *ext4_read_inline_link(struct inode *inode)
1593 struct ext4_iloc iloc;
1594 int ret, inline_size;
1597 ret = ext4_get_inode_loc(inode, &iloc);
1599 return ERR_PTR(ret);
1602 inline_size = ext4_get_inline_size(inode);
1603 link = kmalloc(inline_size + 1, GFP_NOFS);
1607 ret = ext4_read_inline_data(inode, link, inline_size, &iloc);
1612 nd_terminate_link(link, inode->i_size, ret);
1615 link = ERR_PTR(ret);
1620 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1621 struct ext4_dir_entry_2 **parent_de,
1624 struct ext4_iloc iloc;
1626 *retval = ext4_get_inode_loc(inode, &iloc);
1630 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1636 * Try to create the inline data for the new dir.
1637 * If it succeeds, return 0, otherwise return the error.
1638 * In case of ENOSPC, the caller should create the normal disk layout dir.
1640 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1641 struct inode *inode)
1643 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1644 struct ext4_iloc iloc;
1645 struct ext4_dir_entry_2 *de;
1647 ret = ext4_get_inode_loc(inode, &iloc);
1651 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1656 * For inline dir, we only save the inode information for the ".."
1657 * and create a fake dentry to cover the left space.
1659 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1660 de->inode = cpu_to_le32(parent->i_ino);
1661 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1663 de->rec_len = ext4_rec_len_to_disk(
1664 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1666 set_nlink(inode, 2);
1667 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1673 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1674 struct ext4_filename *fname,
1675 struct ext4_dir_entry_2 **res_dir,
1676 int *has_inline_data)
1679 struct ext4_iloc iloc;
1683 if (ext4_get_inode_loc(dir, &iloc))
1686 down_read(&EXT4_I(dir)->xattr_sem);
1687 if (!ext4_has_inline_data(dir)) {
1688 *has_inline_data = 0;
1692 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1693 EXT4_INLINE_DOTDOT_SIZE;
1694 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1695 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1696 dir, fname, 0, res_dir);
1702 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1705 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1706 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1708 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1709 dir, fname, 0, res_dir);
1717 up_read(&EXT4_I(dir)->xattr_sem);
1721 int ext4_delete_inline_entry(handle_t *handle,
1723 struct ext4_dir_entry_2 *de_del,
1724 struct buffer_head *bh,
1725 int *has_inline_data)
1727 int err, inline_size, no_expand;
1728 struct ext4_iloc iloc;
1731 err = ext4_get_inode_loc(dir, &iloc);
1735 ext4_write_lock_xattr(dir, &no_expand);
1736 if (!ext4_has_inline_data(dir)) {
1737 *has_inline_data = 0;
1741 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1742 EXT4_MIN_INLINE_DATA_SIZE) {
1743 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1744 EXT4_INLINE_DOTDOT_SIZE;
1745 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1746 EXT4_INLINE_DOTDOT_SIZE;
1748 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1749 inline_size = ext4_get_inline_size(dir) -
1750 EXT4_MIN_INLINE_DATA_SIZE;
1753 BUFFER_TRACE(bh, "get_write_access");
1754 err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
1759 err = ext4_generic_delete_entry(dir, de_del, bh,
1760 inline_start, inline_size, 0);
1764 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1766 ext4_write_unlock_xattr(dir, &no_expand);
1767 if (likely(err == 0))
1768 err = ext4_mark_inode_dirty(handle, dir);
1771 ext4_std_error(dir->i_sb, err);
1776 * Get the inline dentry at offset.
1778 static inline struct ext4_dir_entry_2 *
1779 ext4_get_inline_entry(struct inode *inode,
1780 struct ext4_iloc *iloc,
1781 unsigned int offset,
1782 void **inline_start,
1787 BUG_ON(offset > ext4_get_inline_size(inode));
1789 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1790 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1791 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1793 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1794 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1795 *inline_size = ext4_get_inline_size(inode) -
1796 EXT4_MIN_INLINE_DATA_SIZE;
1800 *inline_start = inline_pos;
1801 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1804 bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1806 int err, inline_size;
1807 struct ext4_iloc iloc;
1810 unsigned int offset;
1811 struct ext4_dir_entry_2 *de;
1814 err = ext4_get_inode_loc(dir, &iloc);
1816 EXT4_ERROR_INODE_ERR(dir, -err,
1817 "error %d getting inode %lu block",
1822 down_read(&EXT4_I(dir)->xattr_sem);
1823 if (!ext4_has_inline_data(dir)) {
1824 *has_inline_data = 0;
1829 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1830 if (!le32_to_cpu(de->inode)) {
1831 ext4_warning(dir->i_sb,
1832 "bad inline directory (dir #%lu) - no `..'",
1837 inline_len = ext4_get_inline_size(dir);
1838 offset = EXT4_INLINE_DOTDOT_SIZE;
1839 while (offset < inline_len) {
1840 de = ext4_get_inline_entry(dir, &iloc, offset,
1841 &inline_pos, &inline_size);
1842 if (ext4_check_dir_entry(dir, NULL, de,
1843 iloc.bh, inline_pos,
1844 inline_size, offset)) {
1845 ext4_warning(dir->i_sb,
1846 "bad inline directory (dir #%lu) - "
1847 "inode %u, rec_len %u, name_len %d"
1849 dir->i_ino, le32_to_cpu(de->inode),
1850 le16_to_cpu(de->rec_len), de->name_len,
1854 if (le32_to_cpu(de->inode)) {
1857 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1862 up_read(&EXT4_I(dir)->xattr_sem);
1867 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1871 ext4_write_lock_xattr(inode, &no_expand);
1872 ret = ext4_destroy_inline_data_nolock(handle, inode);
1873 ext4_write_unlock_xattr(inode, &no_expand);
1878 int ext4_inline_data_iomap(struct inode *inode, struct iomap *iomap)
1881 int error = -EAGAIN;
1882 struct ext4_iloc iloc;
1884 down_read(&EXT4_I(inode)->xattr_sem);
1885 if (!ext4_has_inline_data(inode))
1888 error = ext4_get_inode_loc(inode, &iloc);
1892 addr = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1893 addr += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1894 addr += offsetof(struct ext4_inode, i_block);
1900 iomap->length = min_t(loff_t, ext4_get_inline_size(inode),
1901 i_size_read(inode));
1902 iomap->type = IOMAP_INLINE;
1906 up_read(&EXT4_I(inode)->xattr_sem);
1910 int ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1913 int inline_size, value_len, needed_blocks, no_expand, err = 0;
1916 struct ext4_xattr_ibody_find is = {
1917 .s = { .not_found = -ENODATA, },
1919 struct ext4_xattr_info i = {
1920 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1921 .name = EXT4_XATTR_SYSTEM_DATA,
1925 needed_blocks = ext4_writepage_trans_blocks(inode);
1926 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1928 return PTR_ERR(handle);
1930 ext4_write_lock_xattr(inode, &no_expand);
1931 if (!ext4_has_inline_data(inode)) {
1932 ext4_write_unlock_xattr(inode, &no_expand);
1934 ext4_journal_stop(handle);
1938 if ((err = ext4_orphan_add(handle, inode)) != 0)
1941 if ((err = ext4_get_inode_loc(inode, &is.iloc)) != 0)
1944 down_write(&EXT4_I(inode)->i_data_sem);
1945 i_size = inode->i_size;
1946 inline_size = ext4_get_inline_size(inode);
1947 EXT4_I(inode)->i_disksize = i_size;
1949 if (i_size < inline_size) {
1951 * if there's inline data to truncate and this file was
1952 * converted to extents after that inline data was written,
1953 * the extent status cache must be cleared to avoid leaving
1954 * behind stale delayed allocated extent entries
1956 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1958 err = ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1959 if (err == -ENOMEM) {
1960 memalloc_retry_wait(GFP_ATOMIC);
1967 /* Clear the content in the xattr space. */
1968 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1969 if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0)
1972 BUG_ON(is.s.not_found);
1974 value_len = le32_to_cpu(is.s.here->e_value_size);
1975 value = kmalloc(value_len, GFP_NOFS);
1981 err = ext4_xattr_ibody_get(inode, i.name_index,
1982 i.name, value, value_len);
1987 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1988 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1989 err = ext4_xattr_ibody_set(handle, inode, &i, &is);
1994 /* Clear the content within i_blocks. */
1995 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1996 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1997 memset(p + i_size, 0,
1998 EXT4_MIN_INLINE_DATA_SIZE - i_size);
2001 EXT4_I(inode)->i_inline_size = i_size <
2002 EXT4_MIN_INLINE_DATA_SIZE ?
2003 EXT4_MIN_INLINE_DATA_SIZE : i_size;
2007 up_write(&EXT4_I(inode)->i_data_sem);
2010 ext4_write_unlock_xattr(inode, &no_expand);
2013 ext4_orphan_del(handle, inode);
2016 inode->i_mtime = inode->i_ctime = current_time(inode);
2017 err = ext4_mark_inode_dirty(handle, inode);
2019 ext4_handle_sync(handle);
2021 ext4_journal_stop(handle);
2025 int ext4_convert_inline_data(struct inode *inode)
2027 int error, needed_blocks, no_expand;
2029 struct ext4_iloc iloc;
2031 if (!ext4_has_inline_data(inode)) {
2032 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
2034 } else if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2036 * Inode has inline data but EXT4_STATE_MAY_INLINE_DATA is
2037 * cleared. This means we are in the middle of moving of
2038 * inline data to delay allocated block. Just force writeout
2039 * here to finish conversion.
2041 error = filemap_flush(inode->i_mapping);
2044 if (!ext4_has_inline_data(inode))
2048 needed_blocks = ext4_writepage_trans_blocks(inode);
2051 error = ext4_get_inode_loc(inode, &iloc);
2055 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2056 if (IS_ERR(handle)) {
2057 error = PTR_ERR(handle);
2061 ext4_write_lock_xattr(inode, &no_expand);
2062 if (ext4_has_inline_data(inode))
2063 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2064 ext4_write_unlock_xattr(inode, &no_expand);
2065 ext4_journal_stop(handle);