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/iversion.h>
11 #include "ext4_jbd2.h"
16 #define EXT4_XATTR_SYSTEM_DATA "data"
17 #define EXT4_MIN_INLINE_DATA_SIZE ((sizeof(__le32) * EXT4_N_BLOCKS))
18 #define EXT4_INLINE_DOTDOT_OFFSET 2
19 #define EXT4_INLINE_DOTDOT_SIZE 4
21 static int ext4_get_inline_size(struct inode *inode)
23 if (EXT4_I(inode)->i_inline_off)
24 return EXT4_I(inode)->i_inline_size;
29 static int get_max_inline_xattr_value_size(struct inode *inode,
30 struct ext4_iloc *iloc)
32 struct ext4_xattr_ibody_header *header;
33 struct ext4_xattr_entry *entry;
34 struct ext4_inode *raw_inode;
38 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
41 min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
42 EXT4_GOOD_OLD_INODE_SIZE -
43 EXT4_I(inode)->i_extra_isize -
44 sizeof(struct ext4_xattr_ibody_header);
47 * We need to subtract another sizeof(__u32) since an in-inode xattr
48 * needs an empty 4 bytes to indicate the gap between the xattr entry
49 * and the name/value pair.
51 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
52 return EXT4_XATTR_SIZE(min_offs -
53 EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
54 EXT4_XATTR_ROUND - sizeof(__u32));
56 raw_inode = ext4_raw_inode(iloc);
57 header = IHDR(inode, raw_inode);
58 entry = IFIRST(header);
59 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
61 /* Compute min_offs. */
62 while (!IS_LAST_ENTRY(entry)) {
63 void *next = EXT4_XATTR_NEXT(entry);
66 EXT4_ERROR_INODE(inode,
67 "corrupt xattr in inline inode");
70 if (!entry->e_value_inum && entry->e_value_size) {
71 size_t offs = le16_to_cpu(entry->e_value_offs);
78 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
80 if (EXT4_I(inode)->i_inline_off) {
81 entry = (struct ext4_xattr_entry *)
82 ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
84 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
88 free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
90 if (free > EXT4_XATTR_ROUND)
91 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
100 * Get the maximum size we now can store in an inode.
101 * If we can't find the space for a xattr entry, don't use the space
102 * of the extents since we have no space to indicate the inline data.
104 int ext4_get_max_inline_size(struct inode *inode)
106 int error, max_inline_size;
107 struct ext4_iloc iloc;
109 if (EXT4_I(inode)->i_extra_isize == 0)
112 error = ext4_get_inode_loc(inode, &iloc);
114 ext4_error_inode(inode, __func__, __LINE__, 0,
115 "can't get inode location %lu",
120 down_read(&EXT4_I(inode)->xattr_sem);
121 max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
122 up_read(&EXT4_I(inode)->xattr_sem);
126 if (!max_inline_size)
129 return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
133 * this function does not take xattr_sem, which is OK because it is
134 * currently only used in a code path coming form ext4_iget, before
135 * the new inode has been unlocked
137 int ext4_find_inline_data_nolock(struct inode *inode)
139 struct ext4_xattr_ibody_find is = {
140 .s = { .not_found = -ENODATA, },
142 struct ext4_xattr_info i = {
143 .name_index = EXT4_XATTR_INDEX_SYSTEM,
144 .name = EXT4_XATTR_SYSTEM_DATA,
148 if (EXT4_I(inode)->i_extra_isize == 0)
151 error = ext4_get_inode_loc(inode, &is.iloc);
155 error = ext4_xattr_ibody_find(inode, &i, &is);
159 if (!is.s.not_found) {
160 if (is.s.here->e_value_inum) {
161 EXT4_ERROR_INODE(inode, "inline data xattr refers "
162 "to an external xattr inode");
163 error = -EFSCORRUPTED;
166 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
167 (void *)ext4_raw_inode(&is.iloc));
168 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
169 le32_to_cpu(is.s.here->e_value_size);
176 static int ext4_read_inline_data(struct inode *inode, void *buffer,
178 struct ext4_iloc *iloc)
180 struct ext4_xattr_entry *entry;
181 struct ext4_xattr_ibody_header *header;
183 struct ext4_inode *raw_inode;
188 BUG_ON(len > EXT4_I(inode)->i_inline_size);
190 cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
191 len : EXT4_MIN_INLINE_DATA_SIZE;
193 raw_inode = ext4_raw_inode(iloc);
194 memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
202 header = IHDR(inode, raw_inode);
203 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
204 EXT4_I(inode)->i_inline_off);
205 len = min_t(unsigned int, len,
206 (unsigned int)le32_to_cpu(entry->e_value_size));
209 (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
217 * write the buffer to the inline inode.
218 * If 'create' is set, we don't need to do the extra copy in the xattr
219 * value since it is already handled by ext4_xattr_ibody_set.
220 * That saves us one memcpy.
222 static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
223 void *buffer, loff_t pos, unsigned int len)
225 struct ext4_xattr_entry *entry;
226 struct ext4_xattr_ibody_header *header;
227 struct ext4_inode *raw_inode;
230 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
233 BUG_ON(!EXT4_I(inode)->i_inline_off);
234 BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
236 raw_inode = ext4_raw_inode(iloc);
239 if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
240 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
241 EXT4_MIN_INLINE_DATA_SIZE - pos : len;
242 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
252 pos -= EXT4_MIN_INLINE_DATA_SIZE;
253 header = IHDR(inode, raw_inode);
254 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
255 EXT4_I(inode)->i_inline_off);
257 memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
261 static int ext4_create_inline_data(handle_t *handle,
262 struct inode *inode, unsigned len)
266 struct ext4_xattr_ibody_find is = {
267 .s = { .not_found = -ENODATA, },
269 struct ext4_xattr_info i = {
270 .name_index = EXT4_XATTR_INDEX_SYSTEM,
271 .name = EXT4_XATTR_SYSTEM_DATA,
274 error = ext4_get_inode_loc(inode, &is.iloc);
278 BUFFER_TRACE(is.iloc.bh, "get_write_access");
279 error = ext4_journal_get_write_access(handle, is.iloc.bh);
283 if (len > EXT4_MIN_INLINE_DATA_SIZE) {
284 value = EXT4_ZERO_XATTR_VALUE;
285 len -= EXT4_MIN_INLINE_DATA_SIZE;
291 /* Insert the the xttr entry. */
295 error = ext4_xattr_ibody_find(inode, &i, &is);
299 BUG_ON(!is.s.not_found);
301 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
303 if (error == -ENOSPC)
304 ext4_clear_inode_state(inode,
305 EXT4_STATE_MAY_INLINE_DATA);
309 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
310 0, EXT4_MIN_INLINE_DATA_SIZE);
312 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
313 (void *)ext4_raw_inode(&is.iloc));
314 EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
315 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
316 ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
318 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
325 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
330 struct ext4_xattr_ibody_find is = {
331 .s = { .not_found = -ENODATA, },
333 struct ext4_xattr_info i = {
334 .name_index = EXT4_XATTR_INDEX_SYSTEM,
335 .name = EXT4_XATTR_SYSTEM_DATA,
338 /* If the old space is ok, write the data directly. */
339 if (len <= EXT4_I(inode)->i_inline_size)
342 error = ext4_get_inode_loc(inode, &is.iloc);
346 error = ext4_xattr_ibody_find(inode, &i, &is);
350 BUG_ON(is.s.not_found);
352 len -= EXT4_MIN_INLINE_DATA_SIZE;
353 value = kzalloc(len, GFP_NOFS);
359 error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
364 BUFFER_TRACE(is.iloc.bh, "get_write_access");
365 error = ext4_journal_get_write_access(handle, is.iloc.bh);
369 /* Update the xttr entry. */
373 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
377 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
378 (void *)ext4_raw_inode(&is.iloc));
379 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
380 le32_to_cpu(is.s.here->e_value_size);
381 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
383 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
391 static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
394 int ret, size, no_expand;
395 struct ext4_inode_info *ei = EXT4_I(inode);
397 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
400 size = ext4_get_max_inline_size(inode);
404 ext4_write_lock_xattr(inode, &no_expand);
406 if (ei->i_inline_off)
407 ret = ext4_update_inline_data(handle, inode, len);
409 ret = ext4_create_inline_data(handle, inode, len);
411 ext4_write_unlock_xattr(inode, &no_expand);
415 static int ext4_destroy_inline_data_nolock(handle_t *handle,
418 struct ext4_inode_info *ei = EXT4_I(inode);
419 struct ext4_xattr_ibody_find is = {
420 .s = { .not_found = 0, },
422 struct ext4_xattr_info i = {
423 .name_index = EXT4_XATTR_INDEX_SYSTEM,
424 .name = EXT4_XATTR_SYSTEM_DATA,
430 if (!ei->i_inline_off)
433 error = ext4_get_inode_loc(inode, &is.iloc);
437 error = ext4_xattr_ibody_find(inode, &i, &is);
441 BUFFER_TRACE(is.iloc.bh, "get_write_access");
442 error = ext4_journal_get_write_access(handle, is.iloc.bh);
446 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
450 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
451 0, EXT4_MIN_INLINE_DATA_SIZE);
452 memset(ei->i_data, 0, EXT4_MIN_INLINE_DATA_SIZE);
454 if (ext4_has_feature_extents(inode->i_sb)) {
455 if (S_ISDIR(inode->i_mode) ||
456 S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
457 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
458 ext4_ext_tree_init(handle, inode);
461 ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
464 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
466 EXT4_I(inode)->i_inline_off = 0;
467 EXT4_I(inode)->i_inline_size = 0;
468 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
471 if (error == -ENODATA)
476 static int ext4_read_inline_page(struct inode *inode, struct page *page)
481 struct ext4_iloc iloc;
483 BUG_ON(!PageLocked(page));
484 BUG_ON(!ext4_has_inline_data(inode));
487 if (!EXT4_I(inode)->i_inline_off) {
488 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
493 ret = ext4_get_inode_loc(inode, &iloc);
497 len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
498 kaddr = kmap_atomic(page);
499 ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
500 flush_dcache_page(page);
501 kunmap_atomic(kaddr);
502 zero_user_segment(page, len, PAGE_SIZE);
503 SetPageUptodate(page);
510 int ext4_readpage_inline(struct inode *inode, struct page *page)
514 down_read(&EXT4_I(inode)->xattr_sem);
515 if (!ext4_has_inline_data(inode)) {
516 up_read(&EXT4_I(inode)->xattr_sem);
521 * Current inline data can only exist in the 1st page,
522 * So for all the other pages, just set them uptodate.
525 ret = ext4_read_inline_page(inode, page);
526 else if (!PageUptodate(page)) {
527 zero_user_segment(page, 0, PAGE_SIZE);
528 SetPageUptodate(page);
531 up_read(&EXT4_I(inode)->xattr_sem);
534 return ret >= 0 ? 0 : ret;
537 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
541 int ret, needed_blocks, no_expand;
542 handle_t *handle = NULL;
543 int retries = 0, sem_held = 0;
544 struct page *page = NULL;
546 struct ext4_iloc iloc;
548 if (!ext4_has_inline_data(inode)) {
550 * clear the flag so that no new write
551 * will trap here again.
553 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
557 needed_blocks = ext4_writepage_trans_blocks(inode);
559 ret = ext4_get_inode_loc(inode, &iloc);
564 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
565 if (IS_ERR(handle)) {
566 ret = PTR_ERR(handle);
571 /* We cannot recurse into the filesystem as the transaction is already
573 flags |= AOP_FLAG_NOFS;
575 page = grab_cache_page_write_begin(mapping, 0, flags);
581 ext4_write_lock_xattr(inode, &no_expand);
583 /* If some one has already done this for us, just exit. */
584 if (!ext4_has_inline_data(inode)) {
590 to = ext4_get_inline_size(inode);
591 if (!PageUptodate(page)) {
592 ret = ext4_read_inline_page(inode, page);
597 ret = ext4_destroy_inline_data_nolock(handle, inode);
601 if (ext4_should_dioread_nolock(inode)) {
602 ret = __block_write_begin(page, from, to,
603 ext4_get_block_unwritten);
605 ret = __block_write_begin(page, from, to, ext4_get_block);
607 if (!ret && ext4_should_journal_data(inode)) {
608 ret = ext4_walk_page_buffers(handle, page_buffers(page),
610 do_journal_get_write_access);
617 ext4_orphan_add(handle, inode);
618 ext4_write_unlock_xattr(inode, &no_expand);
620 ext4_journal_stop(handle);
622 ext4_truncate_failed_write(inode);
624 * If truncate failed early the inode might
625 * still be on the orphan list; we need to
626 * make sure the inode is removed from the
627 * orphan list in that case.
630 ext4_orphan_del(NULL, inode);
633 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
637 block_commit_write(page, from, to);
644 ext4_write_unlock_xattr(inode, &no_expand);
646 ext4_journal_stop(handle);
652 * Try to write data in the inode.
653 * If the inode has inline data, check whether the new write can be
654 * in the inode also. If not, create the page the handle, move the data
655 * to the page make it update and let the later codes create extent for it.
657 int ext4_try_to_write_inline_data(struct address_space *mapping,
659 loff_t pos, unsigned len,
666 struct ext4_iloc iloc;
668 if (pos + len > ext4_get_max_inline_size(inode))
671 ret = ext4_get_inode_loc(inode, &iloc);
676 * The possible write could happen in the inode,
677 * so try to reserve the space in inode first.
679 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
680 if (IS_ERR(handle)) {
681 ret = PTR_ERR(handle);
686 ret = ext4_prepare_inline_data(handle, inode, pos + len);
687 if (ret && ret != -ENOSPC)
690 /* We don't have space in inline inode, so convert it to extent. */
691 if (ret == -ENOSPC) {
692 ext4_journal_stop(handle);
697 ret = ext4_journal_get_write_access(handle, iloc.bh);
701 flags |= AOP_FLAG_NOFS;
703 page = grab_cache_page_write_begin(mapping, 0, flags);
710 down_read(&EXT4_I(inode)->xattr_sem);
711 if (!ext4_has_inline_data(inode)) {
718 if (!PageUptodate(page)) {
719 ret = ext4_read_inline_page(inode, page);
730 up_read(&EXT4_I(inode)->xattr_sem);
732 if (handle && (ret != 1))
733 ext4_journal_stop(handle);
737 return ext4_convert_inline_data_to_extent(mapping,
741 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
742 unsigned copied, struct page *page)
746 struct ext4_iloc iloc;
748 if (unlikely(copied < len) && !PageUptodate(page))
751 ret = ext4_get_inode_loc(inode, &iloc);
753 ext4_std_error(inode->i_sb, ret);
757 ext4_write_lock_xattr(inode, &no_expand);
758 BUG_ON(!ext4_has_inline_data(inode));
761 * ei->i_inline_off may have changed since ext4_write_begin()
762 * called ext4_try_to_write_inline_data()
764 (void) ext4_find_inline_data_nolock(inode);
766 kaddr = kmap_atomic(page);
767 ext4_write_inline_data(inode, &iloc, kaddr, pos, copied);
768 kunmap_atomic(kaddr);
769 SetPageUptodate(page);
770 /* clear page dirty so that writepages wouldn't work for us. */
771 ClearPageDirty(page);
773 ext4_write_unlock_xattr(inode, &no_expand);
775 mark_inode_dirty(inode);
781 ext4_journalled_write_inline_data(struct inode *inode,
787 struct ext4_iloc iloc;
789 ret = ext4_get_inode_loc(inode, &iloc);
791 ext4_std_error(inode->i_sb, ret);
795 ext4_write_lock_xattr(inode, &no_expand);
796 kaddr = kmap_atomic(page);
797 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
798 kunmap_atomic(kaddr);
799 ext4_write_unlock_xattr(inode, &no_expand);
805 * Try to make the page cache and handle ready for the inline data case.
806 * We can call this function in 2 cases:
807 * 1. The inode is created and the first write exceeds inline size. We can
808 * clear the inode state safely.
809 * 2. The inode has inline data, then we need to read the data, make it
810 * update and dirty so that ext4_da_writepages can handle it. We don't
811 * need to start the journal since the file's metatdata isn't changed now.
813 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
818 int ret = 0, inline_size;
821 page = grab_cache_page_write_begin(mapping, 0, flags);
825 down_read(&EXT4_I(inode)->xattr_sem);
826 if (!ext4_has_inline_data(inode)) {
827 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
831 inline_size = ext4_get_inline_size(inode);
833 if (!PageUptodate(page)) {
834 ret = ext4_read_inline_page(inode, page);
839 ret = __block_write_begin(page, 0, inline_size,
840 ext4_da_get_block_prep);
842 up_read(&EXT4_I(inode)->xattr_sem);
845 ext4_truncate_failed_write(inode);
850 SetPageUptodate(page);
851 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
852 *fsdata = (void *)CONVERT_INLINE_DATA;
855 up_read(&EXT4_I(inode)->xattr_sem);
864 * Prepare the write for the inline data.
865 * If the the data can be written into the inode, we just read
866 * the page and make it uptodate, and start the journal.
867 * Otherwise read the page, makes it dirty so that it can be
868 * handle in writepages(the i_disksize update is left to the
869 * normal ext4_da_write_end).
871 int ext4_da_write_inline_data_begin(struct address_space *mapping,
873 loff_t pos, unsigned len,
878 int ret, inline_size;
881 struct ext4_iloc iloc;
884 ret = ext4_get_inode_loc(inode, &iloc);
889 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
890 if (IS_ERR(handle)) {
891 ret = PTR_ERR(handle);
895 inline_size = ext4_get_max_inline_size(inode);
898 if (inline_size >= pos + len) {
899 ret = ext4_prepare_inline_data(handle, inode, pos + len);
900 if (ret && ret != -ENOSPC)
905 * We cannot recurse into the filesystem as the transaction
906 * is already started.
908 flags |= AOP_FLAG_NOFS;
910 if (ret == -ENOSPC) {
911 ext4_journal_stop(handle);
912 ret = ext4_da_convert_inline_data_to_extent(mapping,
916 if (ret == -ENOSPC &&
917 ext4_should_retry_alloc(inode->i_sb, &retries))
922 page = grab_cache_page_write_begin(mapping, 0, flags);
928 down_read(&EXT4_I(inode)->xattr_sem);
929 if (!ext4_has_inline_data(inode)) {
931 goto out_release_page;
934 if (!PageUptodate(page)) {
935 ret = ext4_read_inline_page(inode, page);
937 goto out_release_page;
939 ret = ext4_journal_get_write_access(handle, iloc.bh);
941 goto out_release_page;
943 up_read(&EXT4_I(inode)->xattr_sem);
948 up_read(&EXT4_I(inode)->xattr_sem);
952 ext4_journal_stop(handle);
958 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
959 unsigned len, unsigned copied,
964 ret = ext4_write_inline_data_end(inode, pos, len, copied, page);
973 * No need to use i_size_read() here, the i_size
974 * cannot change under us because we hold i_mutex.
976 * But it's important to update i_size while still holding page lock:
977 * page writeout could otherwise come in and zero beyond i_size.
979 if (pos+copied > inode->i_size)
980 i_size_write(inode, pos+copied);
985 * Don't mark the inode dirty under page lock. First, it unnecessarily
986 * makes the holding time of page lock longer. Second, it forces lock
987 * ordering of page lock and transaction start for journaling
990 mark_inode_dirty(inode);
995 #ifdef INLINE_DIR_DEBUG
996 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
997 void *inline_start, int inline_size)
1000 unsigned short de_len;
1001 struct ext4_dir_entry_2 *de = inline_start;
1002 void *dlimit = inline_start + inline_size;
1004 trace_printk("inode %lu\n", dir->i_ino);
1006 while ((void *)de < dlimit) {
1007 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
1008 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
1009 offset, de_len, de->name_len, de->name,
1010 de->name_len, le32_to_cpu(de->inode));
1011 if (ext4_check_dir_entry(dir, NULL, de, bh,
1012 inline_start, inline_size, offset))
1016 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1020 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1024 * Add a new entry into a inline dir.
1025 * It will return -ENOSPC if no space is available, and -EIO
1026 * and -EEXIST if directory entry already exists.
1028 static int ext4_add_dirent_to_inline(handle_t *handle,
1029 struct ext4_filename *fname,
1031 struct inode *inode,
1032 struct ext4_iloc *iloc,
1033 void *inline_start, int inline_size)
1036 struct ext4_dir_entry_2 *de;
1038 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1039 inline_size, fname, &de);
1043 BUFFER_TRACE(iloc->bh, "get_write_access");
1044 err = ext4_journal_get_write_access(handle, iloc->bh);
1047 ext4_insert_dentry(inode, de, inline_size, fname);
1049 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1052 * XXX shouldn't update any times until successful
1053 * completion of syscall, but too many callers depend
1056 * XXX similarly, too many callers depend on
1057 * ext4_new_inode() setting the times, but error
1058 * recovery deletes the inode, so the worst that can
1059 * happen is that the times are slightly out of date
1060 * and/or different from the directory change time.
1062 dir->i_mtime = dir->i_ctime = current_time(dir);
1063 ext4_update_dx_flag(dir);
1064 inode_inc_iversion(dir);
1068 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1069 struct ext4_iloc *iloc)
1071 struct ext4_xattr_entry *entry;
1072 struct ext4_xattr_ibody_header *header;
1074 BUG_ON(!EXT4_I(inode)->i_inline_off);
1076 header = IHDR(inode, ext4_raw_inode(iloc));
1077 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1078 EXT4_I(inode)->i_inline_off);
1080 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1083 /* Set the final de to cover the whole block. */
1084 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1086 struct ext4_dir_entry_2 *de, *prev_de;
1090 de = (struct ext4_dir_entry_2 *)de_buf;
1092 limit = de_buf + old_size;
1095 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1097 de = (struct ext4_dir_entry_2 *)de_buf;
1098 } while (de_buf < limit);
1100 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1101 old_size, new_size);
1103 /* this is just created, so create an empty entry. */
1105 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1109 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1110 struct ext4_iloc *iloc)
1113 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1114 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1116 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1119 ret = ext4_update_inline_data(handle, dir,
1120 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1124 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1125 EXT4_I(dir)->i_inline_size -
1126 EXT4_MIN_INLINE_DATA_SIZE);
1127 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1131 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1132 struct ext4_iloc *iloc,
1133 void *buf, int inline_size)
1137 ret = ext4_create_inline_data(handle, inode, inline_size);
1139 ext4_msg(inode->i_sb, KERN_EMERG,
1140 "error restoring inline_data for inode -- potential data loss! (inode %lu, error %d)",
1144 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1145 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1148 static int ext4_finish_convert_inline_dir(handle_t *handle,
1149 struct inode *inode,
1150 struct buffer_head *dir_block,
1154 int err, csum_size = 0, header_size = 0;
1155 struct ext4_dir_entry_2 *de;
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 ext4_initialize_dirent_tail(dir_block,
1183 inode->i_sb->s_blocksize);
1184 set_buffer_uptodate(dir_block);
1185 unlock_buffer(dir_block);
1186 err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
1189 set_buffer_verified(dir_block);
1190 return ext4_mark_inode_dirty(handle, inode);
1193 static int ext4_convert_inline_data_nolock(handle_t *handle,
1194 struct inode *inode,
1195 struct ext4_iloc *iloc)
1199 struct buffer_head *data_bh = NULL;
1200 struct ext4_map_blocks map;
1203 inline_size = ext4_get_inline_size(inode);
1204 buf = kmalloc(inline_size, GFP_NOFS);
1210 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1215 * Make sure the inline directory entries pass checks before we try to
1216 * convert them, so that we avoid touching stuff that needs fsck.
1218 if (S_ISDIR(inode->i_mode)) {
1219 error = ext4_check_all_de(inode, iloc->bh,
1220 buf + EXT4_INLINE_DOTDOT_SIZE,
1221 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1226 error = ext4_destroy_inline_data_nolock(handle, inode);
1233 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1236 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1241 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1247 lock_buffer(data_bh);
1248 error = ext4_journal_get_create_access(handle, data_bh);
1250 unlock_buffer(data_bh);
1254 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1256 if (!S_ISDIR(inode->i_mode)) {
1257 memcpy(data_bh->b_data, buf, inline_size);
1258 set_buffer_uptodate(data_bh);
1259 unlock_buffer(data_bh);
1260 error = ext4_handle_dirty_metadata(handle,
1263 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1269 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1278 * Try to add the new entry to the inline data.
1279 * If succeeds, return 0. If not, extended the inline dir and copied data to
1280 * the new created block.
1282 int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1283 struct inode *dir, struct inode *inode)
1285 int ret, inline_size, no_expand;
1287 struct ext4_iloc iloc;
1289 ret = ext4_get_inode_loc(dir, &iloc);
1293 ext4_write_lock_xattr(dir, &no_expand);
1294 if (!ext4_has_inline_data(dir))
1297 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1298 EXT4_INLINE_DOTDOT_SIZE;
1299 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1301 ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1302 inline_start, inline_size);
1306 /* check whether it can be inserted to inline xattr space. */
1307 inline_size = EXT4_I(dir)->i_inline_size -
1308 EXT4_MIN_INLINE_DATA_SIZE;
1310 /* Try to use the xattr space.*/
1311 ret = ext4_update_inline_dir(handle, dir, &iloc);
1312 if (ret && ret != -ENOSPC)
1315 inline_size = EXT4_I(dir)->i_inline_size -
1316 EXT4_MIN_INLINE_DATA_SIZE;
1320 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1322 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1323 inode, &iloc, inline_start,
1331 * The inline space is filled up, so create a new block for it.
1332 * As the extent tree will be created, we have to save the inline
1335 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1338 ext4_write_unlock_xattr(dir, &no_expand);
1339 ext4_mark_inode_dirty(handle, dir);
1345 * This function fills a red-black tree with information from an
1346 * inlined dir. It returns the number directory entries loaded
1347 * into the tree. If there is an error it is returned in err.
1349 int ext4_inlinedir_to_tree(struct file *dir_file,
1350 struct inode *dir, ext4_lblk_t block,
1351 struct dx_hash_info *hinfo,
1352 __u32 start_hash, __u32 start_minor_hash,
1353 int *has_inline_data)
1355 int err = 0, count = 0;
1356 unsigned int parent_ino;
1358 struct ext4_dir_entry_2 *de;
1359 struct inode *inode = file_inode(dir_file);
1360 int ret, inline_size = 0;
1361 struct ext4_iloc iloc;
1362 void *dir_buf = NULL;
1363 struct ext4_dir_entry_2 fake;
1364 struct fscrypt_str tmp_str;
1366 ret = ext4_get_inode_loc(inode, &iloc);
1370 down_read(&EXT4_I(inode)->xattr_sem);
1371 if (!ext4_has_inline_data(inode)) {
1372 up_read(&EXT4_I(inode)->xattr_sem);
1373 *has_inline_data = 0;
1377 inline_size = ext4_get_inline_size(inode);
1378 dir_buf = kmalloc(inline_size, GFP_NOFS);
1381 up_read(&EXT4_I(inode)->xattr_sem);
1385 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1386 up_read(&EXT4_I(inode)->xattr_sem);
1391 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1392 while (pos < inline_size) {
1394 * As inlined dir doesn't store any information about '.' and
1395 * only the inode number of '..' is stored, we have to handle
1399 fake.inode = cpu_to_le32(inode->i_ino);
1401 strcpy(fake.name, ".");
1402 fake.rec_len = ext4_rec_len_to_disk(
1403 EXT4_DIR_REC_LEN(fake.name_len),
1405 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1407 pos = EXT4_INLINE_DOTDOT_OFFSET;
1408 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1409 fake.inode = cpu_to_le32(parent_ino);
1411 strcpy(fake.name, "..");
1412 fake.rec_len = ext4_rec_len_to_disk(
1413 EXT4_DIR_REC_LEN(fake.name_len),
1415 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1417 pos = EXT4_INLINE_DOTDOT_SIZE;
1419 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1420 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1421 if (ext4_check_dir_entry(inode, dir_file, de,
1423 inline_size, pos)) {
1429 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);
1512 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
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))
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 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1592 struct ext4_dir_entry_2 **parent_de,
1595 struct ext4_iloc iloc;
1597 *retval = ext4_get_inode_loc(inode, &iloc);
1601 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1607 * Try to create the inline data for the new dir.
1608 * If it succeeds, return 0, otherwise return the error.
1609 * In case of ENOSPC, the caller should create the normal disk layout dir.
1611 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1612 struct inode *inode)
1614 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1615 struct ext4_iloc iloc;
1616 struct ext4_dir_entry_2 *de;
1618 ret = ext4_get_inode_loc(inode, &iloc);
1622 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1627 * For inline dir, we only save the inode information for the ".."
1628 * and create a fake dentry to cover the left space.
1630 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1631 de->inode = cpu_to_le32(parent->i_ino);
1632 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1634 de->rec_len = ext4_rec_len_to_disk(
1635 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1637 set_nlink(inode, 2);
1638 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1644 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1645 struct ext4_filename *fname,
1646 struct ext4_dir_entry_2 **res_dir,
1647 int *has_inline_data)
1650 struct ext4_iloc iloc;
1654 if (ext4_get_inode_loc(dir, &iloc))
1657 down_read(&EXT4_I(dir)->xattr_sem);
1658 if (!ext4_has_inline_data(dir)) {
1659 *has_inline_data = 0;
1663 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1664 EXT4_INLINE_DOTDOT_SIZE;
1665 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1666 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1667 dir, fname, 0, res_dir);
1673 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1676 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1677 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1679 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1680 dir, fname, 0, res_dir);
1688 up_read(&EXT4_I(dir)->xattr_sem);
1692 int ext4_delete_inline_entry(handle_t *handle,
1694 struct ext4_dir_entry_2 *de_del,
1695 struct buffer_head *bh,
1696 int *has_inline_data)
1698 int err, inline_size, no_expand;
1699 struct ext4_iloc iloc;
1702 err = ext4_get_inode_loc(dir, &iloc);
1706 ext4_write_lock_xattr(dir, &no_expand);
1707 if (!ext4_has_inline_data(dir)) {
1708 *has_inline_data = 0;
1712 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1713 EXT4_MIN_INLINE_DATA_SIZE) {
1714 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1715 EXT4_INLINE_DOTDOT_SIZE;
1716 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1717 EXT4_INLINE_DOTDOT_SIZE;
1719 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1720 inline_size = ext4_get_inline_size(dir) -
1721 EXT4_MIN_INLINE_DATA_SIZE;
1724 BUFFER_TRACE(bh, "get_write_access");
1725 err = ext4_journal_get_write_access(handle, bh);
1729 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1730 inline_start, inline_size, 0);
1734 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1736 ext4_write_unlock_xattr(dir, &no_expand);
1737 if (likely(err == 0))
1738 err = ext4_mark_inode_dirty(handle, dir);
1741 ext4_std_error(dir->i_sb, err);
1746 * Get the inline dentry at offset.
1748 static inline struct ext4_dir_entry_2 *
1749 ext4_get_inline_entry(struct inode *inode,
1750 struct ext4_iloc *iloc,
1751 unsigned int offset,
1752 void **inline_start,
1757 BUG_ON(offset > ext4_get_inline_size(inode));
1759 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1760 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1761 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1763 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1764 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1765 *inline_size = ext4_get_inline_size(inode) -
1766 EXT4_MIN_INLINE_DATA_SIZE;
1770 *inline_start = inline_pos;
1771 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1774 bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1776 int err, inline_size;
1777 struct ext4_iloc iloc;
1780 unsigned int offset;
1781 struct ext4_dir_entry_2 *de;
1784 err = ext4_get_inode_loc(dir, &iloc);
1786 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1791 down_read(&EXT4_I(dir)->xattr_sem);
1792 if (!ext4_has_inline_data(dir)) {
1793 *has_inline_data = 0;
1797 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1798 if (!le32_to_cpu(de->inode)) {
1799 ext4_warning(dir->i_sb,
1800 "bad inline directory (dir #%lu) - no `..'",
1806 inline_len = ext4_get_inline_size(dir);
1807 offset = EXT4_INLINE_DOTDOT_SIZE;
1808 while (offset < inline_len) {
1809 de = ext4_get_inline_entry(dir, &iloc, offset,
1810 &inline_pos, &inline_size);
1811 if (ext4_check_dir_entry(dir, NULL, de,
1812 iloc.bh, inline_pos,
1813 inline_size, offset)) {
1814 ext4_warning(dir->i_sb,
1815 "bad inline directory (dir #%lu) - "
1816 "inode %u, rec_len %u, name_len %d"
1818 dir->i_ino, le32_to_cpu(de->inode),
1819 le16_to_cpu(de->rec_len), de->name_len,
1824 if (le32_to_cpu(de->inode)) {
1828 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1832 up_read(&EXT4_I(dir)->xattr_sem);
1837 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1841 ext4_write_lock_xattr(inode, &no_expand);
1842 ret = ext4_destroy_inline_data_nolock(handle, inode);
1843 ext4_write_unlock_xattr(inode, &no_expand);
1848 int ext4_inline_data_iomap(struct inode *inode, struct iomap *iomap)
1851 int error = -EAGAIN;
1852 struct ext4_iloc iloc;
1854 down_read(&EXT4_I(inode)->xattr_sem);
1855 if (!ext4_has_inline_data(inode))
1858 error = ext4_get_inode_loc(inode, &iloc);
1862 addr = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1863 addr += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1864 addr += offsetof(struct ext4_inode, i_block);
1870 iomap->length = min_t(loff_t, ext4_get_inline_size(inode),
1871 i_size_read(inode));
1872 iomap->type = IOMAP_INLINE;
1876 up_read(&EXT4_I(inode)->xattr_sem);
1880 int ext4_inline_data_fiemap(struct inode *inode,
1881 struct fiemap_extent_info *fieinfo,
1882 int *has_inline, __u64 start, __u64 len)
1886 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1889 struct ext4_iloc iloc;
1891 down_read(&EXT4_I(inode)->xattr_sem);
1892 if (!ext4_has_inline_data(inode)) {
1896 inline_len = min_t(size_t, ext4_get_inline_size(inode),
1897 i_size_read(inode));
1898 if (start >= inline_len)
1900 if (start + len < inline_len)
1901 inline_len = start + len;
1902 inline_len -= start;
1904 error = ext4_get_inode_loc(inode, &iloc);
1908 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1909 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1910 physical += offsetof(struct ext4_inode, i_block);
1914 up_read(&EXT4_I(inode)->xattr_sem);
1916 error = fiemap_fill_next_extent(fieinfo, start, physical,
1918 return (error < 0 ? error : 0);
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) {
1961 /* Clear the content in the xattr space. */
1962 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1963 if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0)
1966 BUG_ON(is.s.not_found);
1968 value_len = le32_to_cpu(is.s.here->e_value_size);
1969 value = kmalloc(value_len, GFP_NOFS);
1975 err = ext4_xattr_ibody_get(inode, i.name_index,
1976 i.name, value, value_len);
1981 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1982 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1983 err = ext4_xattr_ibody_set(handle, inode, &i, &is);
1988 /* Clear the content within i_blocks. */
1989 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1990 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1991 memset(p + i_size, 0,
1992 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1995 EXT4_I(inode)->i_inline_size = i_size <
1996 EXT4_MIN_INLINE_DATA_SIZE ?
1997 EXT4_MIN_INLINE_DATA_SIZE : i_size;
2001 up_write(&EXT4_I(inode)->i_data_sem);
2004 ext4_write_unlock_xattr(inode, &no_expand);
2007 ext4_orphan_del(handle, inode);
2010 inode->i_mtime = inode->i_ctime = current_time(inode);
2011 err = ext4_mark_inode_dirty(handle, inode);
2013 ext4_handle_sync(handle);
2015 ext4_journal_stop(handle);
2019 int ext4_convert_inline_data(struct inode *inode)
2021 int error, needed_blocks, no_expand;
2023 struct ext4_iloc iloc;
2025 if (!ext4_has_inline_data(inode)) {
2026 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
2028 } else if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2030 * Inode has inline data but EXT4_STATE_MAY_INLINE_DATA is
2031 * cleared. This means we are in the middle of moving of
2032 * inline data to delay allocated block. Just force writeout
2033 * here to finish conversion.
2035 error = filemap_flush(inode->i_mapping);
2038 if (!ext4_has_inline_data(inode))
2042 needed_blocks = ext4_writepage_trans_blocks(inode);
2045 error = ext4_get_inode_loc(inode, &iloc);
2049 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2050 if (IS_ERR(handle)) {
2051 error = PTR_ERR(handle);
2055 ext4_write_lock_xattr(inode, &no_expand);
2056 if (ext4_has_inline_data(inode))
2057 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2058 ext4_write_unlock_xattr(inode, &no_expand);
2059 ext4_journal_stop(handle);