2 * Copyright (c) 2012 Taobao.
3 * Written by Tao Ma <boyu.mt@taobao.com>
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of version 2.1 of the GNU Lesser General Public License
7 * as published by the Free Software Foundation.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
15 #include <linux/fiemap.h>
17 #include "ext4_jbd2.h"
22 #define EXT4_XATTR_SYSTEM_DATA "data"
23 #define EXT4_MIN_INLINE_DATA_SIZE ((sizeof(__le32) * EXT4_N_BLOCKS))
24 #define EXT4_INLINE_DOTDOT_OFFSET 2
25 #define EXT4_INLINE_DOTDOT_SIZE 4
27 static int ext4_get_inline_size(struct inode *inode)
29 if (EXT4_I(inode)->i_inline_off)
30 return EXT4_I(inode)->i_inline_size;
35 static int get_max_inline_xattr_value_size(struct inode *inode,
36 struct ext4_iloc *iloc)
38 struct ext4_xattr_ibody_header *header;
39 struct ext4_xattr_entry *entry;
40 struct ext4_inode *raw_inode;
43 min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
44 EXT4_GOOD_OLD_INODE_SIZE -
45 EXT4_I(inode)->i_extra_isize -
46 sizeof(struct ext4_xattr_ibody_header);
49 * We need to subtract another sizeof(__u32) since an in-inode xattr
50 * needs an empty 4 bytes to indicate the gap between the xattr entry
51 * and the name/value pair.
53 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
54 return EXT4_XATTR_SIZE(min_offs -
55 EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
56 EXT4_XATTR_ROUND - sizeof(__u32));
58 raw_inode = ext4_raw_inode(iloc);
59 header = IHDR(inode, raw_inode);
60 entry = IFIRST(header);
62 /* Compute min_offs. */
63 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
64 if (!entry->e_value_inum && entry->e_value_size) {
65 size_t offs = le16_to_cpu(entry->e_value_offs);
71 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
73 if (EXT4_I(inode)->i_inline_off) {
74 entry = (struct ext4_xattr_entry *)
75 ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
77 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
81 free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
83 if (free > EXT4_XATTR_ROUND)
84 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
93 * Get the maximum size we now can store in an inode.
94 * If we can't find the space for a xattr entry, don't use the space
95 * of the extents since we have no space to indicate the inline data.
97 int ext4_get_max_inline_size(struct inode *inode)
99 int error, max_inline_size;
100 struct ext4_iloc iloc;
102 if (EXT4_I(inode)->i_extra_isize == 0)
105 error = ext4_get_inode_loc(inode, &iloc);
107 ext4_error_inode(inode, __func__, __LINE__, 0,
108 "can't get inode location %lu",
113 down_read(&EXT4_I(inode)->xattr_sem);
114 max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
115 up_read(&EXT4_I(inode)->xattr_sem);
119 if (!max_inline_size)
122 return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
126 * this function does not take xattr_sem, which is OK because it is
127 * currently only used in a code path coming form ext4_iget, before
128 * the new inode has been unlocked
130 int ext4_find_inline_data_nolock(struct inode *inode)
132 struct ext4_xattr_ibody_find is = {
133 .s = { .not_found = -ENODATA, },
135 struct ext4_xattr_info i = {
136 .name_index = EXT4_XATTR_INDEX_SYSTEM,
137 .name = EXT4_XATTR_SYSTEM_DATA,
141 if (EXT4_I(inode)->i_extra_isize == 0)
144 error = ext4_get_inode_loc(inode, &is.iloc);
148 error = ext4_xattr_ibody_find(inode, &i, &is);
152 if (!is.s.not_found) {
153 if (is.s.here->e_value_inum) {
154 EXT4_ERROR_INODE(inode, "inline data xattr refers "
155 "to an external xattr inode");
156 error = -EFSCORRUPTED;
159 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
160 (void *)ext4_raw_inode(&is.iloc));
161 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
162 le32_to_cpu(is.s.here->e_value_size);
163 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
170 static int ext4_read_inline_data(struct inode *inode, void *buffer,
172 struct ext4_iloc *iloc)
174 struct ext4_xattr_entry *entry;
175 struct ext4_xattr_ibody_header *header;
177 struct ext4_inode *raw_inode;
182 BUG_ON(len > EXT4_I(inode)->i_inline_size);
184 cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
185 len : EXT4_MIN_INLINE_DATA_SIZE;
187 raw_inode = ext4_raw_inode(iloc);
188 memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
196 header = IHDR(inode, raw_inode);
197 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
198 EXT4_I(inode)->i_inline_off);
199 len = min_t(unsigned int, len,
200 (unsigned int)le32_to_cpu(entry->e_value_size));
203 (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
211 * write the buffer to the inline inode.
212 * If 'create' is set, we don't need to do the extra copy in the xattr
213 * value since it is already handled by ext4_xattr_ibody_inline_set.
214 * That saves us one memcpy.
216 static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
217 void *buffer, loff_t pos, unsigned int len)
219 struct ext4_xattr_entry *entry;
220 struct ext4_xattr_ibody_header *header;
221 struct ext4_inode *raw_inode;
224 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
227 BUG_ON(!EXT4_I(inode)->i_inline_off);
228 BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
230 raw_inode = ext4_raw_inode(iloc);
233 if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
234 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
235 EXT4_MIN_INLINE_DATA_SIZE - pos : len;
236 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
246 pos -= EXT4_MIN_INLINE_DATA_SIZE;
247 header = IHDR(inode, raw_inode);
248 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
249 EXT4_I(inode)->i_inline_off);
251 memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
255 static int ext4_create_inline_data(handle_t *handle,
256 struct inode *inode, unsigned len)
260 struct ext4_xattr_ibody_find is = {
261 .s = { .not_found = -ENODATA, },
263 struct ext4_xattr_info i = {
264 .name_index = EXT4_XATTR_INDEX_SYSTEM,
265 .name = EXT4_XATTR_SYSTEM_DATA,
268 error = ext4_get_inode_loc(inode, &is.iloc);
272 BUFFER_TRACE(is.iloc.bh, "get_write_access");
273 error = ext4_journal_get_write_access(handle, is.iloc.bh);
277 if (len > EXT4_MIN_INLINE_DATA_SIZE) {
278 value = EXT4_ZERO_XATTR_VALUE;
279 len -= EXT4_MIN_INLINE_DATA_SIZE;
285 /* Insert the the xttr entry. */
289 error = ext4_xattr_ibody_find(inode, &i, &is);
293 BUG_ON(!is.s.not_found);
295 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
297 if (error == -ENOSPC)
298 ext4_clear_inode_state(inode,
299 EXT4_STATE_MAY_INLINE_DATA);
303 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
304 0, EXT4_MIN_INLINE_DATA_SIZE);
306 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
307 (void *)ext4_raw_inode(&is.iloc));
308 EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
309 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
310 ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
312 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
319 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
324 struct ext4_xattr_ibody_find is = {
325 .s = { .not_found = -ENODATA, },
327 struct ext4_xattr_info i = {
328 .name_index = EXT4_XATTR_INDEX_SYSTEM,
329 .name = EXT4_XATTR_SYSTEM_DATA,
332 /* If the old space is ok, write the data directly. */
333 if (len <= EXT4_I(inode)->i_inline_size)
336 error = ext4_get_inode_loc(inode, &is.iloc);
340 error = ext4_xattr_ibody_find(inode, &i, &is);
344 BUG_ON(is.s.not_found);
346 len -= EXT4_MIN_INLINE_DATA_SIZE;
347 value = kzalloc(len, GFP_NOFS);
353 error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
355 if (error == -ENODATA)
358 BUFFER_TRACE(is.iloc.bh, "get_write_access");
359 error = ext4_journal_get_write_access(handle, is.iloc.bh);
363 /* Update the xttr entry. */
367 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
371 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
372 (void *)ext4_raw_inode(&is.iloc));
373 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
374 le32_to_cpu(is.s.here->e_value_size);
375 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
377 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
385 static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
388 int ret, size, no_expand;
389 struct ext4_inode_info *ei = EXT4_I(inode);
391 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
394 size = ext4_get_max_inline_size(inode);
398 ext4_write_lock_xattr(inode, &no_expand);
400 if (ei->i_inline_off)
401 ret = ext4_update_inline_data(handle, inode, len);
403 ret = ext4_create_inline_data(handle, inode, len);
405 ext4_write_unlock_xattr(inode, &no_expand);
409 static int ext4_destroy_inline_data_nolock(handle_t *handle,
412 struct ext4_inode_info *ei = EXT4_I(inode);
413 struct ext4_xattr_ibody_find is = {
414 .s = { .not_found = 0, },
416 struct ext4_xattr_info i = {
417 .name_index = EXT4_XATTR_INDEX_SYSTEM,
418 .name = EXT4_XATTR_SYSTEM_DATA,
424 if (!ei->i_inline_off)
427 error = ext4_get_inode_loc(inode, &is.iloc);
431 error = ext4_xattr_ibody_find(inode, &i, &is);
435 BUFFER_TRACE(is.iloc.bh, "get_write_access");
436 error = ext4_journal_get_write_access(handle, is.iloc.bh);
440 error = ext4_xattr_ibody_inline_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_page(struct inode *inode, struct page *page)
475 struct ext4_iloc iloc;
477 BUG_ON(!PageLocked(page));
478 BUG_ON(!ext4_has_inline_data(inode));
481 if (!EXT4_I(inode)->i_inline_off) {
482 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
487 ret = ext4_get_inode_loc(inode, &iloc);
491 len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
492 kaddr = kmap_atomic(page);
493 ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
494 flush_dcache_page(page);
495 kunmap_atomic(kaddr);
496 zero_user_segment(page, len, PAGE_SIZE);
497 SetPageUptodate(page);
504 int ext4_readpage_inline(struct inode *inode, struct page *page)
508 down_read(&EXT4_I(inode)->xattr_sem);
509 if (!ext4_has_inline_data(inode)) {
510 up_read(&EXT4_I(inode)->xattr_sem);
515 * Current inline data can only exist in the 1st page,
516 * So for all the other pages, just set them uptodate.
519 ret = ext4_read_inline_page(inode, page);
520 else if (!PageUptodate(page)) {
521 zero_user_segment(page, 0, PAGE_SIZE);
522 SetPageUptodate(page);
525 up_read(&EXT4_I(inode)->xattr_sem);
528 return ret >= 0 ? 0 : ret;
531 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
535 int ret, needed_blocks, no_expand;
536 handle_t *handle = NULL;
537 int retries = 0, sem_held = 0;
538 struct page *page = NULL;
540 struct ext4_iloc iloc;
542 if (!ext4_has_inline_data(inode)) {
544 * clear the flag so that no new write
545 * will trap here again.
547 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
551 needed_blocks = ext4_writepage_trans_blocks(inode);
553 ret = ext4_get_inode_loc(inode, &iloc);
558 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
559 if (IS_ERR(handle)) {
560 ret = PTR_ERR(handle);
565 /* We cannot recurse into the filesystem as the transaction is already
567 flags |= AOP_FLAG_NOFS;
569 page = grab_cache_page_write_begin(mapping, 0, flags);
575 ext4_write_lock_xattr(inode, &no_expand);
577 /* If some one has already done this for us, just exit. */
578 if (!ext4_has_inline_data(inode)) {
584 to = ext4_get_inline_size(inode);
585 if (!PageUptodate(page)) {
586 ret = ext4_read_inline_page(inode, page);
591 ret = ext4_destroy_inline_data_nolock(handle, inode);
595 if (ext4_should_dioread_nolock(inode)) {
596 ret = __block_write_begin(page, from, to,
597 ext4_get_block_unwritten);
599 ret = __block_write_begin(page, from, to, ext4_get_block);
601 if (!ret && ext4_should_journal_data(inode)) {
602 ret = ext4_walk_page_buffers(handle, page_buffers(page),
604 do_journal_get_write_access);
611 ext4_orphan_add(handle, inode);
612 ext4_write_unlock_xattr(inode, &no_expand);
614 ext4_journal_stop(handle);
616 ext4_truncate_failed_write(inode);
618 * If truncate failed early the inode might
619 * still be on the orphan list; we need to
620 * make sure the inode is removed from the
621 * orphan list in that case.
624 ext4_orphan_del(NULL, inode);
627 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
631 block_commit_write(page, from, to);
638 ext4_write_unlock_xattr(inode, &no_expand);
640 ext4_journal_stop(handle);
646 * Try to write data in the inode.
647 * If the inode has inline data, check whether the new write can be
648 * in the inode also. If not, create the page the handle, move the data
649 * to the page make it update and let the later codes create extent for it.
651 int ext4_try_to_write_inline_data(struct address_space *mapping,
653 loff_t pos, unsigned len,
660 struct ext4_iloc iloc;
662 if (pos + len > ext4_get_max_inline_size(inode))
665 ret = ext4_get_inode_loc(inode, &iloc);
670 * The possible write could happen in the inode,
671 * so try to reserve the space in inode first.
673 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
674 if (IS_ERR(handle)) {
675 ret = PTR_ERR(handle);
680 ret = ext4_prepare_inline_data(handle, inode, pos + len);
681 if (ret && ret != -ENOSPC)
684 /* We don't have space in inline inode, so convert it to extent. */
685 if (ret == -ENOSPC) {
686 ext4_journal_stop(handle);
691 ret = ext4_journal_get_write_access(handle, iloc.bh);
695 flags |= AOP_FLAG_NOFS;
697 page = grab_cache_page_write_begin(mapping, 0, flags);
704 down_read(&EXT4_I(inode)->xattr_sem);
705 if (!ext4_has_inline_data(inode)) {
712 if (!PageUptodate(page)) {
713 ret = ext4_read_inline_page(inode, page);
724 up_read(&EXT4_I(inode)->xattr_sem);
726 if (handle && (ret != 1))
727 ext4_journal_stop(handle);
731 return ext4_convert_inline_data_to_extent(mapping,
735 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
736 unsigned copied, struct page *page)
740 struct ext4_iloc iloc;
742 if (unlikely(copied < len)) {
743 if (!PageUptodate(page)) {
749 ret = ext4_get_inode_loc(inode, &iloc);
751 ext4_std_error(inode->i_sb, ret);
756 ext4_write_lock_xattr(inode, &no_expand);
757 BUG_ON(!ext4_has_inline_data(inode));
760 * ei->i_inline_off may have changed since ext4_write_begin()
761 * called ext4_try_to_write_inline_data()
763 (void) ext4_find_inline_data_nolock(inode);
765 kaddr = kmap_atomic(page);
766 ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
767 kunmap_atomic(kaddr);
768 SetPageUptodate(page);
769 /* clear page dirty so that writepages wouldn't work for us. */
770 ClearPageDirty(page);
772 ext4_write_unlock_xattr(inode, &no_expand);
774 mark_inode_dirty(inode);
780 ext4_journalled_write_inline_data(struct inode *inode,
786 struct ext4_iloc iloc;
788 ret = ext4_get_inode_loc(inode, &iloc);
790 ext4_std_error(inode->i_sb, ret);
794 ext4_write_lock_xattr(inode, &no_expand);
795 kaddr = kmap_atomic(page);
796 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
797 kunmap_atomic(kaddr);
798 ext4_write_unlock_xattr(inode, &no_expand);
804 * Try to make the page cache and handle ready for the inline data case.
805 * We can call this function in 2 cases:
806 * 1. The inode is created and the first write exceeds inline size. We can
807 * clear the inode state safely.
808 * 2. The inode has inline data, then we need to read the data, make it
809 * update and dirty so that ext4_da_writepages can handle it. We don't
810 * need to start the journal since the file's metatdata isn't changed now.
812 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
817 int ret = 0, inline_size;
820 page = grab_cache_page_write_begin(mapping, 0, flags);
824 down_read(&EXT4_I(inode)->xattr_sem);
825 if (!ext4_has_inline_data(inode)) {
826 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
830 inline_size = ext4_get_inline_size(inode);
832 if (!PageUptodate(page)) {
833 ret = ext4_read_inline_page(inode, page);
838 ret = __block_write_begin(page, 0, inline_size,
839 ext4_da_get_block_prep);
841 up_read(&EXT4_I(inode)->xattr_sem);
844 ext4_truncate_failed_write(inode);
849 SetPageUptodate(page);
850 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
851 *fsdata = (void *)CONVERT_INLINE_DATA;
854 up_read(&EXT4_I(inode)->xattr_sem);
863 * Prepare the write for the inline data.
864 * If the the data can be written into the inode, we just read
865 * the page and make it uptodate, and start the journal.
866 * Otherwise read the page, makes it dirty so that it can be
867 * handle in writepages(the i_disksize update is left to the
868 * normal ext4_da_write_end).
870 int ext4_da_write_inline_data_begin(struct address_space *mapping,
872 loff_t pos, unsigned len,
877 int ret, inline_size;
880 struct ext4_iloc iloc;
883 ret = ext4_get_inode_loc(inode, &iloc);
888 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
889 if (IS_ERR(handle)) {
890 ret = PTR_ERR(handle);
894 inline_size = ext4_get_max_inline_size(inode);
897 if (inline_size >= pos + len) {
898 ret = ext4_prepare_inline_data(handle, inode, pos + len);
899 if (ret && ret != -ENOSPC)
904 * We cannot recurse into the filesystem as the transaction
905 * is already started.
907 flags |= AOP_FLAG_NOFS;
909 if (ret == -ENOSPC) {
910 ext4_journal_stop(handle);
911 ret = ext4_da_convert_inline_data_to_extent(mapping,
915 if (ret == -ENOSPC &&
916 ext4_should_retry_alloc(inode->i_sb, &retries))
921 page = grab_cache_page_write_begin(mapping, 0, flags);
927 down_read(&EXT4_I(inode)->xattr_sem);
928 if (!ext4_has_inline_data(inode)) {
930 goto out_release_page;
933 if (!PageUptodate(page)) {
934 ret = ext4_read_inline_page(inode, page);
936 goto out_release_page;
938 ret = ext4_journal_get_write_access(handle, iloc.bh);
940 goto out_release_page;
942 up_read(&EXT4_I(inode)->xattr_sem);
947 up_read(&EXT4_I(inode)->xattr_sem);
951 ext4_journal_stop(handle);
957 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
958 unsigned len, unsigned copied,
963 ret = ext4_write_inline_data_end(inode, pos, len, copied, page);
972 * No need to use i_size_read() here, the i_size
973 * cannot change under us because we hold i_mutex.
975 * But it's important to update i_size while still holding page lock:
976 * page writeout could otherwise come in and zero beyond i_size.
978 if (pos+copied > inode->i_size)
979 i_size_write(inode, pos+copied);
984 * Don't mark the inode dirty under page lock. First, it unnecessarily
985 * makes the holding time of page lock longer. Second, it forces lock
986 * ordering of page lock and transaction start for journaling
989 mark_inode_dirty(inode);
994 #ifdef INLINE_DIR_DEBUG
995 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
996 void *inline_start, int inline_size)
999 unsigned short de_len;
1000 struct ext4_dir_entry_2 *de = inline_start;
1001 void *dlimit = inline_start + inline_size;
1003 trace_printk("inode %lu\n", dir->i_ino);
1005 while ((void *)de < dlimit) {
1006 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
1007 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
1008 offset, de_len, de->name_len, de->name,
1009 de->name_len, le32_to_cpu(de->inode));
1010 if (ext4_check_dir_entry(dir, NULL, de, bh,
1011 inline_start, inline_size, offset))
1015 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1019 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1023 * Add a new entry into a inline dir.
1024 * It will return -ENOSPC if no space is available, and -EIO
1025 * and -EEXIST if directory entry already exists.
1027 static int ext4_add_dirent_to_inline(handle_t *handle,
1028 struct ext4_filename *fname,
1030 struct inode *inode,
1031 struct ext4_iloc *iloc,
1032 void *inline_start, int inline_size)
1035 struct ext4_dir_entry_2 *de;
1037 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1038 inline_size, fname, &de);
1042 BUFFER_TRACE(iloc->bh, "get_write_access");
1043 err = ext4_journal_get_write_access(handle, iloc->bh);
1046 ext4_insert_dentry(inode, de, inline_size, fname);
1048 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1051 * XXX shouldn't update any times until successful
1052 * completion of syscall, but too many callers depend
1055 * XXX similarly, too many callers depend on
1056 * ext4_new_inode() setting the times, but error
1057 * recovery deletes the inode, so the worst that can
1058 * happen is that the times are slightly out of date
1059 * and/or different from the directory change time.
1061 dir->i_mtime = dir->i_ctime = current_time(dir);
1062 ext4_update_dx_flag(dir);
1067 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1068 struct ext4_iloc *iloc)
1070 struct ext4_xattr_entry *entry;
1071 struct ext4_xattr_ibody_header *header;
1073 BUG_ON(!EXT4_I(inode)->i_inline_off);
1075 header = IHDR(inode, ext4_raw_inode(iloc));
1076 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1077 EXT4_I(inode)->i_inline_off);
1079 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1082 /* Set the final de to cover the whole block. */
1083 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1085 struct ext4_dir_entry_2 *de, *prev_de;
1089 de = (struct ext4_dir_entry_2 *)de_buf;
1091 limit = de_buf + old_size;
1094 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1096 de = (struct ext4_dir_entry_2 *)de_buf;
1097 } while (de_buf < limit);
1099 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1100 old_size, new_size);
1102 /* this is just created, so create an empty entry. */
1104 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1108 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1109 struct ext4_iloc *iloc)
1112 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1113 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1115 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1118 ret = ext4_update_inline_data(handle, dir,
1119 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1123 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1124 EXT4_I(dir)->i_inline_size -
1125 EXT4_MIN_INLINE_DATA_SIZE);
1126 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1130 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1131 struct ext4_iloc *iloc,
1132 void *buf, int inline_size)
1136 ret = ext4_create_inline_data(handle, inode, inline_size);
1138 ext4_msg(inode->i_sb, KERN_EMERG,
1139 "error restoring inline_data for inode -- potential data loss! (inode %lu, error %d)",
1143 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1144 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1147 static int ext4_finish_convert_inline_dir(handle_t *handle,
1148 struct inode *inode,
1149 struct buffer_head *dir_block,
1153 int err, csum_size = 0, header_size = 0;
1154 struct ext4_dir_entry_2 *de;
1155 struct ext4_dir_entry_tail *t;
1156 void *target = dir_block->b_data;
1159 * First create "." and ".." and then copy the dir information
1160 * back to the block.
1162 de = (struct ext4_dir_entry_2 *)target;
1163 de = ext4_init_dot_dotdot(inode, de,
1164 inode->i_sb->s_blocksize, csum_size,
1165 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1166 header_size = (void *)de - target;
1168 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1169 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1171 if (ext4_has_metadata_csum(inode->i_sb))
1172 csum_size = sizeof(struct ext4_dir_entry_tail);
1174 inode->i_size = inode->i_sb->s_blocksize;
1175 i_size_write(inode, inode->i_sb->s_blocksize);
1176 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1177 ext4_update_final_de(dir_block->b_data,
1178 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1179 inode->i_sb->s_blocksize - csum_size);
1182 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1183 inode->i_sb->s_blocksize);
1184 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1186 set_buffer_uptodate(dir_block);
1187 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1190 set_buffer_verified(dir_block);
1191 return ext4_mark_inode_dirty(handle, inode);
1194 static int ext4_convert_inline_data_nolock(handle_t *handle,
1195 struct inode *inode,
1196 struct ext4_iloc *iloc)
1200 struct buffer_head *data_bh = NULL;
1201 struct ext4_map_blocks map;
1204 inline_size = ext4_get_inline_size(inode);
1205 buf = kmalloc(inline_size, GFP_NOFS);
1211 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1216 * Make sure the inline directory entries pass checks before we try to
1217 * convert them, so that we avoid touching stuff that needs fsck.
1219 if (S_ISDIR(inode->i_mode)) {
1220 error = ext4_check_all_de(inode, iloc->bh,
1221 buf + EXT4_INLINE_DOTDOT_SIZE,
1222 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1227 error = ext4_destroy_inline_data_nolock(handle, inode);
1234 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1237 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1242 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1248 lock_buffer(data_bh);
1249 error = ext4_journal_get_create_access(handle, data_bh);
1251 unlock_buffer(data_bh);
1255 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1257 if (!S_ISDIR(inode->i_mode)) {
1258 memcpy(data_bh->b_data, buf, inline_size);
1259 set_buffer_uptodate(data_bh);
1260 error = ext4_handle_dirty_metadata(handle,
1263 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1267 unlock_buffer(data_bh);
1270 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1279 * Try to add the new entry to the inline data.
1280 * If succeeds, return 0. If not, extended the inline dir and copied data to
1281 * the new created block.
1283 int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1284 struct inode *dir, struct inode *inode)
1286 int ret, inline_size, no_expand;
1288 struct ext4_iloc iloc;
1290 ret = ext4_get_inode_loc(dir, &iloc);
1294 ext4_write_lock_xattr(dir, &no_expand);
1295 if (!ext4_has_inline_data(dir))
1298 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1299 EXT4_INLINE_DOTDOT_SIZE;
1300 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1302 ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1303 inline_start, inline_size);
1307 /* check whether it can be inserted to inline xattr space. */
1308 inline_size = EXT4_I(dir)->i_inline_size -
1309 EXT4_MIN_INLINE_DATA_SIZE;
1311 /* Try to use the xattr space.*/
1312 ret = ext4_update_inline_dir(handle, dir, &iloc);
1313 if (ret && ret != -ENOSPC)
1316 inline_size = EXT4_I(dir)->i_inline_size -
1317 EXT4_MIN_INLINE_DATA_SIZE;
1321 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1323 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1324 inode, &iloc, inline_start,
1332 * The inline space is filled up, so create a new block for it.
1333 * As the extent tree will be created, we have to save the inline
1336 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1339 ext4_write_unlock_xattr(dir, &no_expand);
1340 ext4_mark_inode_dirty(handle, dir);
1346 * This function fills a red-black tree with information from an
1347 * inlined dir. It returns the number directory entries loaded
1348 * into the tree. If there is an error it is returned in err.
1350 int htree_inlinedir_to_tree(struct file *dir_file,
1351 struct inode *dir, ext4_lblk_t block,
1352 struct dx_hash_info *hinfo,
1353 __u32 start_hash, __u32 start_minor_hash,
1354 int *has_inline_data)
1356 int err = 0, count = 0;
1357 unsigned int parent_ino;
1359 struct ext4_dir_entry_2 *de;
1360 struct inode *inode = file_inode(dir_file);
1361 int ret, inline_size = 0;
1362 struct ext4_iloc iloc;
1363 void *dir_buf = NULL;
1364 struct ext4_dir_entry_2 fake;
1365 struct fscrypt_str tmp_str;
1367 ret = ext4_get_inode_loc(inode, &iloc);
1371 down_read(&EXT4_I(inode)->xattr_sem);
1372 if (!ext4_has_inline_data(inode)) {
1373 up_read(&EXT4_I(inode)->xattr_sem);
1374 *has_inline_data = 0;
1378 inline_size = ext4_get_inline_size(inode);
1379 dir_buf = kmalloc(inline_size, GFP_NOFS);
1382 up_read(&EXT4_I(inode)->xattr_sem);
1386 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1387 up_read(&EXT4_I(inode)->xattr_sem);
1392 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1393 while (pos < inline_size) {
1395 * As inlined dir doesn't store any information about '.' and
1396 * only the inode number of '..' is stored, we have to handle
1400 fake.inode = cpu_to_le32(inode->i_ino);
1402 strcpy(fake.name, ".");
1403 fake.rec_len = ext4_rec_len_to_disk(
1404 EXT4_DIR_REC_LEN(fake.name_len),
1406 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1408 pos = EXT4_INLINE_DOTDOT_OFFSET;
1409 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1410 fake.inode = cpu_to_le32(parent_ino);
1412 strcpy(fake.name, "..");
1413 fake.rec_len = ext4_rec_len_to_disk(
1414 EXT4_DIR_REC_LEN(fake.name_len),
1416 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1418 pos = EXT4_INLINE_DOTDOT_SIZE;
1420 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1421 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1422 if (ext4_check_dir_entry(inode, dir_file, de,
1424 inline_size, pos)) {
1430 ext4fs_dirhash(de->name, de->name_len, hinfo);
1431 if ((hinfo->hash < start_hash) ||
1432 ((hinfo->hash == start_hash) &&
1433 (hinfo->minor_hash < start_minor_hash)))
1437 tmp_str.name = de->name;
1438 tmp_str.len = de->name_len;
1439 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1440 hinfo->minor_hash, de, &tmp_str);
1455 * So this function is called when the volume is mkfsed with
1456 * dir_index disabled. In order to keep f_pos persistent
1457 * after we convert from an inlined dir to a blocked based,
1458 * we just pretend that we are a normal dir and return the
1459 * offset as if '.' and '..' really take place.
1462 int ext4_read_inline_dir(struct file *file,
1463 struct dir_context *ctx,
1464 int *has_inline_data)
1466 unsigned int offset, parent_ino;
1468 struct ext4_dir_entry_2 *de;
1469 struct super_block *sb;
1470 struct inode *inode = file_inode(file);
1471 int ret, inline_size = 0;
1472 struct ext4_iloc iloc;
1473 void *dir_buf = NULL;
1474 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1476 ret = ext4_get_inode_loc(inode, &iloc);
1480 down_read(&EXT4_I(inode)->xattr_sem);
1481 if (!ext4_has_inline_data(inode)) {
1482 up_read(&EXT4_I(inode)->xattr_sem);
1483 *has_inline_data = 0;
1487 inline_size = ext4_get_inline_size(inode);
1488 dir_buf = kmalloc(inline_size, GFP_NOFS);
1491 up_read(&EXT4_I(inode)->xattr_sem);
1495 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1496 up_read(&EXT4_I(inode)->xattr_sem);
1502 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1506 * dotdot_offset and dotdot_size is the real offset and
1507 * size for ".." and "." if the dir is block based while
1508 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1509 * So we will use extra_offset and extra_size to indicate them
1510 * during the inline dir iteration.
1512 dotdot_offset = EXT4_DIR_REC_LEN(1);
1513 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1514 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1515 extra_size = extra_offset + inline_size;
1518 * If the version has changed since the last call to
1519 * readdir(2), then we might be pointing to an invalid
1520 * dirent right now. Scan from the start of the inline
1523 if (file->f_version != inode->i_version) {
1524 for (i = 0; i < extra_size && i < offset;) {
1526 * "." is with offset 0 and
1527 * ".." is dotdot_offset.
1532 } else if (i == dotdot_offset) {
1536 /* for other entry, the real offset in
1537 * the buf has to be tuned accordingly.
1539 de = (struct ext4_dir_entry_2 *)
1540 (dir_buf + i - extra_offset);
1541 /* It's too expensive to do a full
1542 * dirent test each time round this
1543 * loop, but we do have to test at
1544 * least that it is non-zero. A
1545 * failure will be detected in the
1546 * dirent test below. */
1547 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1548 < EXT4_DIR_REC_LEN(1))
1550 i += ext4_rec_len_from_disk(de->rec_len,
1555 file->f_version = inode->i_version;
1558 while (ctx->pos < extra_size) {
1559 if (ctx->pos == 0) {
1560 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1562 ctx->pos = dotdot_offset;
1566 if (ctx->pos == dotdot_offset) {
1567 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1569 ctx->pos = dotdot_size;
1573 de = (struct ext4_dir_entry_2 *)
1574 (dir_buf + ctx->pos - extra_offset);
1575 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1576 extra_size, ctx->pos))
1578 if (le32_to_cpu(de->inode)) {
1579 if (!dir_emit(ctx, de->name, de->name_len,
1580 le32_to_cpu(de->inode),
1581 get_dtype(sb, de->file_type)))
1584 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1592 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1593 struct ext4_dir_entry_2 **parent_de,
1596 struct ext4_iloc iloc;
1598 *retval = ext4_get_inode_loc(inode, &iloc);
1602 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1608 * Try to create the inline data for the new dir.
1609 * If it succeeds, return 0, otherwise return the error.
1610 * In case of ENOSPC, the caller should create the normal disk layout dir.
1612 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1613 struct inode *inode)
1615 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1616 struct ext4_iloc iloc;
1617 struct ext4_dir_entry_2 *de;
1619 ret = ext4_get_inode_loc(inode, &iloc);
1623 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1628 * For inline dir, we only save the inode information for the ".."
1629 * and create a fake dentry to cover the left space.
1631 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1632 de->inode = cpu_to_le32(parent->i_ino);
1633 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1635 de->rec_len = ext4_rec_len_to_disk(
1636 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1638 set_nlink(inode, 2);
1639 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1645 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1646 struct ext4_filename *fname,
1647 struct ext4_dir_entry_2 **res_dir,
1648 int *has_inline_data)
1651 struct ext4_iloc iloc;
1655 if (ext4_get_inode_loc(dir, &iloc))
1658 down_read(&EXT4_I(dir)->xattr_sem);
1659 if (!ext4_has_inline_data(dir)) {
1660 *has_inline_data = 0;
1664 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1665 EXT4_INLINE_DOTDOT_SIZE;
1666 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1667 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1668 dir, fname, 0, res_dir);
1674 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1677 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1678 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1680 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1681 dir, fname, 0, res_dir);
1689 up_read(&EXT4_I(dir)->xattr_sem);
1693 int ext4_delete_inline_entry(handle_t *handle,
1695 struct ext4_dir_entry_2 *de_del,
1696 struct buffer_head *bh,
1697 int *has_inline_data)
1699 int err, inline_size, no_expand;
1700 struct ext4_iloc iloc;
1703 err = ext4_get_inode_loc(dir, &iloc);
1707 ext4_write_lock_xattr(dir, &no_expand);
1708 if (!ext4_has_inline_data(dir)) {
1709 *has_inline_data = 0;
1713 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1714 EXT4_MIN_INLINE_DATA_SIZE) {
1715 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1716 EXT4_INLINE_DOTDOT_SIZE;
1717 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1718 EXT4_INLINE_DOTDOT_SIZE;
1720 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1721 inline_size = ext4_get_inline_size(dir) -
1722 EXT4_MIN_INLINE_DATA_SIZE;
1725 BUFFER_TRACE(bh, "get_write_access");
1726 err = ext4_journal_get_write_access(handle, bh);
1730 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1731 inline_start, inline_size, 0);
1735 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1737 ext4_write_unlock_xattr(dir, &no_expand);
1738 if (likely(err == 0))
1739 err = ext4_mark_inode_dirty(handle, dir);
1742 ext4_std_error(dir->i_sb, err);
1747 * Get the inline dentry at offset.
1749 static inline struct ext4_dir_entry_2 *
1750 ext4_get_inline_entry(struct inode *inode,
1751 struct ext4_iloc *iloc,
1752 unsigned int offset,
1753 void **inline_start,
1758 BUG_ON(offset > ext4_get_inline_size(inode));
1760 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1761 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1762 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1764 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1765 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1766 *inline_size = ext4_get_inline_size(inode) -
1767 EXT4_MIN_INLINE_DATA_SIZE;
1771 *inline_start = inline_pos;
1772 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1775 bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1777 int err, inline_size;
1778 struct ext4_iloc iloc;
1781 unsigned int offset;
1782 struct ext4_dir_entry_2 *de;
1785 err = ext4_get_inode_loc(dir, &iloc);
1787 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1792 down_read(&EXT4_I(dir)->xattr_sem);
1793 if (!ext4_has_inline_data(dir)) {
1794 *has_inline_data = 0;
1798 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1799 if (!le32_to_cpu(de->inode)) {
1800 ext4_warning(dir->i_sb,
1801 "bad inline directory (dir #%lu) - no `..'",
1807 inline_len = ext4_get_inline_size(dir);
1808 offset = EXT4_INLINE_DOTDOT_SIZE;
1809 while (offset < inline_len) {
1810 de = ext4_get_inline_entry(dir, &iloc, offset,
1811 &inline_pos, &inline_size);
1812 if (ext4_check_dir_entry(dir, NULL, de,
1813 iloc.bh, inline_pos,
1814 inline_size, offset)) {
1815 ext4_warning(dir->i_sb,
1816 "bad inline directory (dir #%lu) - "
1817 "inode %u, rec_len %u, name_len %d"
1819 dir->i_ino, le32_to_cpu(de->inode),
1820 le16_to_cpu(de->rec_len), de->name_len,
1825 if (le32_to_cpu(de->inode)) {
1829 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1833 up_read(&EXT4_I(dir)->xattr_sem);
1838 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1842 ext4_write_lock_xattr(inode, &no_expand);
1843 ret = ext4_destroy_inline_data_nolock(handle, inode);
1844 ext4_write_unlock_xattr(inode, &no_expand);
1849 int ext4_inline_data_fiemap(struct inode *inode,
1850 struct fiemap_extent_info *fieinfo,
1851 int *has_inline, __u64 start, __u64 len)
1855 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1858 struct ext4_iloc iloc;
1860 down_read(&EXT4_I(inode)->xattr_sem);
1861 if (!ext4_has_inline_data(inode)) {
1865 inline_len = min_t(size_t, ext4_get_inline_size(inode),
1866 i_size_read(inode));
1867 if (start >= inline_len)
1869 if (start + len < inline_len)
1870 inline_len = start + len;
1871 inline_len -= start;
1873 error = ext4_get_inode_loc(inode, &iloc);
1877 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1878 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1879 physical += offsetof(struct ext4_inode, i_block);
1883 up_read(&EXT4_I(inode)->xattr_sem);
1885 error = fiemap_fill_next_extent(fieinfo, start, physical,
1887 return (error < 0 ? error : 0);
1890 int ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1893 int inline_size, value_len, needed_blocks, no_expand, err = 0;
1896 struct ext4_xattr_ibody_find is = {
1897 .s = { .not_found = -ENODATA, },
1899 struct ext4_xattr_info i = {
1900 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1901 .name = EXT4_XATTR_SYSTEM_DATA,
1905 needed_blocks = ext4_writepage_trans_blocks(inode);
1906 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1908 return PTR_ERR(handle);
1910 ext4_write_lock_xattr(inode, &no_expand);
1911 if (!ext4_has_inline_data(inode)) {
1912 ext4_write_unlock_xattr(inode, &no_expand);
1914 ext4_journal_stop(handle);
1918 if ((err = ext4_orphan_add(handle, inode)) != 0)
1921 if ((err = ext4_get_inode_loc(inode, &is.iloc)) != 0)
1924 down_write(&EXT4_I(inode)->i_data_sem);
1925 i_size = inode->i_size;
1926 inline_size = ext4_get_inline_size(inode);
1927 EXT4_I(inode)->i_disksize = i_size;
1929 if (i_size < inline_size) {
1930 /* Clear the content in the xattr space. */
1931 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1932 if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0)
1935 BUG_ON(is.s.not_found);
1937 value_len = le32_to_cpu(is.s.here->e_value_size);
1938 value = kmalloc(value_len, GFP_NOFS);
1944 err = ext4_xattr_ibody_get(inode, i.name_index,
1945 i.name, value, value_len);
1950 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1951 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1952 err = ext4_xattr_ibody_inline_set(handle, inode,
1958 /* Clear the content within i_blocks. */
1959 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1960 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1961 memset(p + i_size, 0,
1962 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1965 EXT4_I(inode)->i_inline_size = i_size <
1966 EXT4_MIN_INLINE_DATA_SIZE ?
1967 EXT4_MIN_INLINE_DATA_SIZE : i_size;
1971 up_write(&EXT4_I(inode)->i_data_sem);
1974 ext4_write_unlock_xattr(inode, &no_expand);
1977 ext4_orphan_del(handle, inode);
1980 inode->i_mtime = inode->i_ctime = current_time(inode);
1981 err = ext4_mark_inode_dirty(handle, inode);
1983 ext4_handle_sync(handle);
1985 ext4_journal_stop(handle);
1989 int ext4_convert_inline_data(struct inode *inode)
1991 int error, needed_blocks, no_expand;
1993 struct ext4_iloc iloc;
1995 if (!ext4_has_inline_data(inode)) {
1996 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1998 } else if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2000 * Inode has inline data but EXT4_STATE_MAY_INLINE_DATA is
2001 * cleared. This means we are in the middle of moving of
2002 * inline data to delay allocated block. Just force writeout
2003 * here to finish conversion.
2005 error = filemap_flush(inode->i_mapping);
2008 if (!ext4_has_inline_data(inode))
2012 needed_blocks = ext4_writepage_trans_blocks(inode);
2015 error = ext4_get_inode_loc(inode, &iloc);
2019 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2020 if (IS_ERR(handle)) {
2021 error = PTR_ERR(handle);
2025 ext4_write_lock_xattr(inode, &no_expand);
2026 if (ext4_has_inline_data(inode))
2027 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2028 ext4_write_unlock_xattr(inode, &no_expand);
2029 ext4_journal_stop(handle);