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;
37 min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
38 EXT4_GOOD_OLD_INODE_SIZE -
39 EXT4_I(inode)->i_extra_isize -
40 sizeof(struct ext4_xattr_ibody_header);
43 * We need to subtract another sizeof(__u32) since an in-inode xattr
44 * needs an empty 4 bytes to indicate the gap between the xattr entry
45 * and the name/value pair.
47 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
48 return EXT4_XATTR_SIZE(min_offs -
49 EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
50 EXT4_XATTR_ROUND - sizeof(__u32));
52 raw_inode = ext4_raw_inode(iloc);
53 header = IHDR(inode, raw_inode);
54 entry = IFIRST(header);
56 /* Compute min_offs. */
57 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
58 if (!entry->e_value_inum && entry->e_value_size) {
59 size_t offs = le16_to_cpu(entry->e_value_offs);
65 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
67 if (EXT4_I(inode)->i_inline_off) {
68 entry = (struct ext4_xattr_entry *)
69 ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
71 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
75 free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
77 if (free > EXT4_XATTR_ROUND)
78 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
87 * Get the maximum size we now can store in an inode.
88 * If we can't find the space for a xattr entry, don't use the space
89 * of the extents since we have no space to indicate the inline data.
91 int ext4_get_max_inline_size(struct inode *inode)
93 int error, max_inline_size;
94 struct ext4_iloc iloc;
96 if (EXT4_I(inode)->i_extra_isize == 0)
99 error = ext4_get_inode_loc(inode, &iloc);
101 ext4_error_inode(inode, __func__, __LINE__, 0,
102 "can't get inode location %lu",
107 down_read(&EXT4_I(inode)->xattr_sem);
108 max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
109 up_read(&EXT4_I(inode)->xattr_sem);
113 if (!max_inline_size)
116 return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
120 * this function does not take xattr_sem, which is OK because it is
121 * currently only used in a code path coming form ext4_iget, before
122 * the new inode has been unlocked
124 int ext4_find_inline_data_nolock(struct inode *inode)
126 struct ext4_xattr_ibody_find is = {
127 .s = { .not_found = -ENODATA, },
129 struct ext4_xattr_info i = {
130 .name_index = EXT4_XATTR_INDEX_SYSTEM,
131 .name = EXT4_XATTR_SYSTEM_DATA,
135 if (EXT4_I(inode)->i_extra_isize == 0)
138 error = ext4_get_inode_loc(inode, &is.iloc);
142 error = ext4_xattr_ibody_find(inode, &i, &is);
146 if (!is.s.not_found) {
147 if (is.s.here->e_value_inum) {
148 EXT4_ERROR_INODE(inode, "inline data xattr refers "
149 "to an external xattr inode");
150 error = -EFSCORRUPTED;
153 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
154 (void *)ext4_raw_inode(&is.iloc));
155 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
156 le32_to_cpu(is.s.here->e_value_size);
157 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
164 static int ext4_read_inline_data(struct inode *inode, void *buffer,
166 struct ext4_iloc *iloc)
168 struct ext4_xattr_entry *entry;
169 struct ext4_xattr_ibody_header *header;
171 struct ext4_inode *raw_inode;
176 BUG_ON(len > EXT4_I(inode)->i_inline_size);
178 cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
179 len : EXT4_MIN_INLINE_DATA_SIZE;
181 raw_inode = ext4_raw_inode(iloc);
182 memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
190 header = IHDR(inode, raw_inode);
191 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
192 EXT4_I(inode)->i_inline_off);
193 len = min_t(unsigned int, len,
194 (unsigned int)le32_to_cpu(entry->e_value_size));
197 (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
205 * write the buffer to the inline inode.
206 * If 'create' is set, we don't need to do the extra copy in the xattr
207 * value since it is already handled by ext4_xattr_ibody_inline_set.
208 * That saves us one memcpy.
210 static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
211 void *buffer, loff_t pos, unsigned int len)
213 struct ext4_xattr_entry *entry;
214 struct ext4_xattr_ibody_header *header;
215 struct ext4_inode *raw_inode;
218 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
221 BUG_ON(!EXT4_I(inode)->i_inline_off);
222 BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
224 raw_inode = ext4_raw_inode(iloc);
227 if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
228 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
229 EXT4_MIN_INLINE_DATA_SIZE - pos : len;
230 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
240 pos -= EXT4_MIN_INLINE_DATA_SIZE;
241 header = IHDR(inode, raw_inode);
242 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
243 EXT4_I(inode)->i_inline_off);
245 memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
249 static int ext4_create_inline_data(handle_t *handle,
250 struct inode *inode, unsigned len)
254 struct ext4_xattr_ibody_find is = {
255 .s = { .not_found = -ENODATA, },
257 struct ext4_xattr_info i = {
258 .name_index = EXT4_XATTR_INDEX_SYSTEM,
259 .name = EXT4_XATTR_SYSTEM_DATA,
262 error = ext4_get_inode_loc(inode, &is.iloc);
266 BUFFER_TRACE(is.iloc.bh, "get_write_access");
267 error = ext4_journal_get_write_access(handle, is.iloc.bh);
271 if (len > EXT4_MIN_INLINE_DATA_SIZE) {
272 value = EXT4_ZERO_XATTR_VALUE;
273 len -= EXT4_MIN_INLINE_DATA_SIZE;
279 /* Insert the the xttr entry. */
283 error = ext4_xattr_ibody_find(inode, &i, &is);
287 BUG_ON(!is.s.not_found);
289 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
291 if (error == -ENOSPC)
292 ext4_clear_inode_state(inode,
293 EXT4_STATE_MAY_INLINE_DATA);
297 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
298 0, EXT4_MIN_INLINE_DATA_SIZE);
300 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
301 (void *)ext4_raw_inode(&is.iloc));
302 EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
303 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
304 ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
306 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
313 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
318 struct ext4_xattr_ibody_find is = {
319 .s = { .not_found = -ENODATA, },
321 struct ext4_xattr_info i = {
322 .name_index = EXT4_XATTR_INDEX_SYSTEM,
323 .name = EXT4_XATTR_SYSTEM_DATA,
326 /* If the old space is ok, write the data directly. */
327 if (len <= EXT4_I(inode)->i_inline_size)
330 error = ext4_get_inode_loc(inode, &is.iloc);
334 error = ext4_xattr_ibody_find(inode, &i, &is);
338 BUG_ON(is.s.not_found);
340 len -= EXT4_MIN_INLINE_DATA_SIZE;
341 value = kzalloc(len, GFP_NOFS);
347 error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
349 if (error == -ENODATA)
352 BUFFER_TRACE(is.iloc.bh, "get_write_access");
353 error = ext4_journal_get_write_access(handle, is.iloc.bh);
357 /* Update the xttr entry. */
361 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
365 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
366 (void *)ext4_raw_inode(&is.iloc));
367 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
368 le32_to_cpu(is.s.here->e_value_size);
369 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
371 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
379 static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
382 int ret, size, no_expand;
383 struct ext4_inode_info *ei = EXT4_I(inode);
385 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
388 size = ext4_get_max_inline_size(inode);
392 ext4_write_lock_xattr(inode, &no_expand);
394 if (ei->i_inline_off)
395 ret = ext4_update_inline_data(handle, inode, len);
397 ret = ext4_create_inline_data(handle, inode, len);
399 ext4_write_unlock_xattr(inode, &no_expand);
403 static int ext4_destroy_inline_data_nolock(handle_t *handle,
406 struct ext4_inode_info *ei = EXT4_I(inode);
407 struct ext4_xattr_ibody_find is = {
408 .s = { .not_found = 0, },
410 struct ext4_xattr_info i = {
411 .name_index = EXT4_XATTR_INDEX_SYSTEM,
412 .name = EXT4_XATTR_SYSTEM_DATA,
418 if (!ei->i_inline_off)
421 error = ext4_get_inode_loc(inode, &is.iloc);
425 error = ext4_xattr_ibody_find(inode, &i, &is);
429 BUFFER_TRACE(is.iloc.bh, "get_write_access");
430 error = ext4_journal_get_write_access(handle, is.iloc.bh);
434 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
438 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
439 0, EXT4_MIN_INLINE_DATA_SIZE);
441 if (ext4_has_feature_extents(inode->i_sb)) {
442 if (S_ISDIR(inode->i_mode) ||
443 S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
444 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
445 ext4_ext_tree_init(handle, inode);
448 ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
451 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
453 EXT4_I(inode)->i_inline_off = 0;
454 EXT4_I(inode)->i_inline_size = 0;
455 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
458 if (error == -ENODATA)
463 static int ext4_read_inline_page(struct inode *inode, struct page *page)
468 struct ext4_iloc iloc;
470 BUG_ON(!PageLocked(page));
471 BUG_ON(!ext4_has_inline_data(inode));
474 if (!EXT4_I(inode)->i_inline_off) {
475 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
480 ret = ext4_get_inode_loc(inode, &iloc);
484 len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
485 kaddr = kmap_atomic(page);
486 ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
487 flush_dcache_page(page);
488 kunmap_atomic(kaddr);
489 zero_user_segment(page, len, PAGE_SIZE);
490 SetPageUptodate(page);
497 int ext4_readpage_inline(struct inode *inode, struct page *page)
501 down_read(&EXT4_I(inode)->xattr_sem);
502 if (!ext4_has_inline_data(inode)) {
503 up_read(&EXT4_I(inode)->xattr_sem);
508 * Current inline data can only exist in the 1st page,
509 * So for all the other pages, just set them uptodate.
512 ret = ext4_read_inline_page(inode, page);
513 else if (!PageUptodate(page)) {
514 zero_user_segment(page, 0, PAGE_SIZE);
515 SetPageUptodate(page);
518 up_read(&EXT4_I(inode)->xattr_sem);
521 return ret >= 0 ? 0 : ret;
524 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
528 int ret, needed_blocks, no_expand;
529 handle_t *handle = NULL;
530 int retries = 0, sem_held = 0;
531 struct page *page = NULL;
533 struct ext4_iloc iloc;
535 if (!ext4_has_inline_data(inode)) {
537 * clear the flag so that no new write
538 * will trap here again.
540 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
544 needed_blocks = ext4_writepage_trans_blocks(inode);
546 ret = ext4_get_inode_loc(inode, &iloc);
551 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
552 if (IS_ERR(handle)) {
553 ret = PTR_ERR(handle);
558 /* We cannot recurse into the filesystem as the transaction is already
560 flags |= AOP_FLAG_NOFS;
562 page = grab_cache_page_write_begin(mapping, 0, flags);
568 ext4_write_lock_xattr(inode, &no_expand);
570 /* If some one has already done this for us, just exit. */
571 if (!ext4_has_inline_data(inode)) {
577 to = ext4_get_inline_size(inode);
578 if (!PageUptodate(page)) {
579 ret = ext4_read_inline_page(inode, page);
584 ret = ext4_destroy_inline_data_nolock(handle, inode);
588 if (ext4_should_dioread_nolock(inode)) {
589 ret = __block_write_begin(page, from, to,
590 ext4_get_block_unwritten);
592 ret = __block_write_begin(page, from, to, ext4_get_block);
594 if (!ret && ext4_should_journal_data(inode)) {
595 ret = ext4_walk_page_buffers(handle, page_buffers(page),
597 do_journal_get_write_access);
604 ext4_orphan_add(handle, inode);
605 ext4_write_unlock_xattr(inode, &no_expand);
607 ext4_journal_stop(handle);
609 ext4_truncate_failed_write(inode);
611 * If truncate failed early the inode might
612 * still be on the orphan list; we need to
613 * make sure the inode is removed from the
614 * orphan list in that case.
617 ext4_orphan_del(NULL, inode);
620 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
624 block_commit_write(page, from, to);
631 ext4_write_unlock_xattr(inode, &no_expand);
633 ext4_journal_stop(handle);
639 * Try to write data in the inode.
640 * If the inode has inline data, check whether the new write can be
641 * in the inode also. If not, create the page the handle, move the data
642 * to the page make it update and let the later codes create extent for it.
644 int ext4_try_to_write_inline_data(struct address_space *mapping,
646 loff_t pos, unsigned len,
653 struct ext4_iloc iloc;
655 if (pos + len > ext4_get_max_inline_size(inode))
658 ret = ext4_get_inode_loc(inode, &iloc);
663 * The possible write could happen in the inode,
664 * so try to reserve the space in inode first.
666 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
667 if (IS_ERR(handle)) {
668 ret = PTR_ERR(handle);
673 ret = ext4_prepare_inline_data(handle, inode, pos + len);
674 if (ret && ret != -ENOSPC)
677 /* We don't have space in inline inode, so convert it to extent. */
678 if (ret == -ENOSPC) {
679 ext4_journal_stop(handle);
684 flags |= AOP_FLAG_NOFS;
686 page = grab_cache_page_write_begin(mapping, 0, flags);
693 down_read(&EXT4_I(inode)->xattr_sem);
694 if (!ext4_has_inline_data(inode)) {
701 if (!PageUptodate(page)) {
702 ret = ext4_read_inline_page(inode, page);
710 up_read(&EXT4_I(inode)->xattr_sem);
713 ext4_journal_stop(handle);
717 return ext4_convert_inline_data_to_extent(mapping,
721 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
722 unsigned copied, struct page *page)
726 struct ext4_iloc iloc;
728 if (unlikely(copied < len)) {
729 if (!PageUptodate(page)) {
735 ret = ext4_get_inode_loc(inode, &iloc);
737 ext4_std_error(inode->i_sb, ret);
742 ext4_write_lock_xattr(inode, &no_expand);
743 BUG_ON(!ext4_has_inline_data(inode));
745 kaddr = kmap_atomic(page);
746 ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
747 kunmap_atomic(kaddr);
748 SetPageUptodate(page);
749 /* clear page dirty so that writepages wouldn't work for us. */
750 ClearPageDirty(page);
752 ext4_write_unlock_xattr(inode, &no_expand);
759 ext4_journalled_write_inline_data(struct inode *inode,
765 struct ext4_iloc iloc;
767 ret = ext4_get_inode_loc(inode, &iloc);
769 ext4_std_error(inode->i_sb, ret);
773 ext4_write_lock_xattr(inode, &no_expand);
774 kaddr = kmap_atomic(page);
775 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
776 kunmap_atomic(kaddr);
777 ext4_write_unlock_xattr(inode, &no_expand);
783 * Try to make the page cache and handle ready for the inline data case.
784 * We can call this function in 2 cases:
785 * 1. The inode is created and the first write exceeds inline size. We can
786 * clear the inode state safely.
787 * 2. The inode has inline data, then we need to read the data, make it
788 * update and dirty so that ext4_da_writepages can handle it. We don't
789 * need to start the journal since the file's metatdata isn't changed now.
791 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
796 int ret = 0, inline_size;
799 page = grab_cache_page_write_begin(mapping, 0, flags);
803 down_read(&EXT4_I(inode)->xattr_sem);
804 if (!ext4_has_inline_data(inode)) {
805 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
809 inline_size = ext4_get_inline_size(inode);
811 if (!PageUptodate(page)) {
812 ret = ext4_read_inline_page(inode, page);
817 ret = __block_write_begin(page, 0, inline_size,
818 ext4_da_get_block_prep);
820 up_read(&EXT4_I(inode)->xattr_sem);
823 ext4_truncate_failed_write(inode);
828 SetPageUptodate(page);
829 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
830 *fsdata = (void *)CONVERT_INLINE_DATA;
833 up_read(&EXT4_I(inode)->xattr_sem);
842 * Prepare the write for the inline data.
843 * If the the data can be written into the inode, we just read
844 * the page and make it uptodate, and start the journal.
845 * Otherwise read the page, makes it dirty so that it can be
846 * handle in writepages(the i_disksize update is left to the
847 * normal ext4_da_write_end).
849 int ext4_da_write_inline_data_begin(struct address_space *mapping,
851 loff_t pos, unsigned len,
856 int ret, inline_size;
859 struct ext4_iloc iloc;
862 ret = ext4_get_inode_loc(inode, &iloc);
867 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
868 if (IS_ERR(handle)) {
869 ret = PTR_ERR(handle);
873 inline_size = ext4_get_max_inline_size(inode);
876 if (inline_size >= pos + len) {
877 ret = ext4_prepare_inline_data(handle, inode, pos + len);
878 if (ret && ret != -ENOSPC)
883 * We cannot recurse into the filesystem as the transaction
884 * is already started.
886 flags |= AOP_FLAG_NOFS;
888 if (ret == -ENOSPC) {
889 ret = ext4_da_convert_inline_data_to_extent(mapping,
893 ext4_journal_stop(handle);
894 if (ret == -ENOSPC &&
895 ext4_should_retry_alloc(inode->i_sb, &retries))
901 page = grab_cache_page_write_begin(mapping, 0, flags);
907 down_read(&EXT4_I(inode)->xattr_sem);
908 if (!ext4_has_inline_data(inode)) {
910 goto out_release_page;
913 if (!PageUptodate(page)) {
914 ret = ext4_read_inline_page(inode, page);
916 goto out_release_page;
919 up_read(&EXT4_I(inode)->xattr_sem);
924 up_read(&EXT4_I(inode)->xattr_sem);
928 ext4_journal_stop(handle);
934 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
935 unsigned len, unsigned copied,
938 int i_size_changed = 0;
941 ret = ext4_write_inline_data_end(inode, pos, len, copied, page);
950 * No need to use i_size_read() here, the i_size
951 * cannot change under us because we hold i_mutex.
953 * But it's important to update i_size while still holding page lock:
954 * page writeout could otherwise come in and zero beyond i_size.
956 if (pos+copied > inode->i_size) {
957 i_size_write(inode, pos+copied);
964 * Don't mark the inode dirty under page lock. First, it unnecessarily
965 * makes the holding time of page lock longer. Second, it forces lock
966 * ordering of page lock and transaction start for journaling
970 mark_inode_dirty(inode);
975 #ifdef INLINE_DIR_DEBUG
976 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
977 void *inline_start, int inline_size)
980 unsigned short de_len;
981 struct ext4_dir_entry_2 *de = inline_start;
982 void *dlimit = inline_start + inline_size;
984 trace_printk("inode %lu\n", dir->i_ino);
986 while ((void *)de < dlimit) {
987 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
988 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
989 offset, de_len, de->name_len, de->name,
990 de->name_len, le32_to_cpu(de->inode));
991 if (ext4_check_dir_entry(dir, NULL, de, bh,
992 inline_start, inline_size, offset))
996 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1000 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1004 * Add a new entry into a inline dir.
1005 * It will return -ENOSPC if no space is available, and -EIO
1006 * and -EEXIST if directory entry already exists.
1008 static int ext4_add_dirent_to_inline(handle_t *handle,
1009 struct ext4_filename *fname,
1011 struct inode *inode,
1012 struct ext4_iloc *iloc,
1013 void *inline_start, int inline_size)
1016 struct ext4_dir_entry_2 *de;
1018 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1019 inline_size, fname, &de);
1023 BUFFER_TRACE(iloc->bh, "get_write_access");
1024 err = ext4_journal_get_write_access(handle, iloc->bh);
1027 ext4_insert_dentry(inode, de, inline_size, fname);
1029 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1032 * XXX shouldn't update any times until successful
1033 * completion of syscall, but too many callers depend
1036 * XXX similarly, too many callers depend on
1037 * ext4_new_inode() setting the times, but error
1038 * recovery deletes the inode, so the worst that can
1039 * happen is that the times are slightly out of date
1040 * and/or different from the directory change time.
1042 dir->i_mtime = dir->i_ctime = current_time(dir);
1043 ext4_update_dx_flag(dir);
1044 inode_inc_iversion(dir);
1048 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1049 struct ext4_iloc *iloc)
1051 struct ext4_xattr_entry *entry;
1052 struct ext4_xattr_ibody_header *header;
1054 BUG_ON(!EXT4_I(inode)->i_inline_off);
1056 header = IHDR(inode, ext4_raw_inode(iloc));
1057 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1058 EXT4_I(inode)->i_inline_off);
1060 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1063 /* Set the final de to cover the whole block. */
1064 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1066 struct ext4_dir_entry_2 *de, *prev_de;
1070 de = (struct ext4_dir_entry_2 *)de_buf;
1072 limit = de_buf + old_size;
1075 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1077 de = (struct ext4_dir_entry_2 *)de_buf;
1078 } while (de_buf < limit);
1080 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1081 old_size, new_size);
1083 /* this is just created, so create an empty entry. */
1085 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1089 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1090 struct ext4_iloc *iloc)
1093 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1094 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1096 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1099 ret = ext4_update_inline_data(handle, dir,
1100 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1104 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1105 EXT4_I(dir)->i_inline_size -
1106 EXT4_MIN_INLINE_DATA_SIZE);
1107 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1111 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1112 struct ext4_iloc *iloc,
1113 void *buf, int inline_size)
1115 ext4_create_inline_data(handle, inode, inline_size);
1116 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1117 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1120 static int ext4_finish_convert_inline_dir(handle_t *handle,
1121 struct inode *inode,
1122 struct buffer_head *dir_block,
1126 int err, csum_size = 0, header_size = 0;
1127 struct ext4_dir_entry_2 *de;
1128 struct ext4_dir_entry_tail *t;
1129 void *target = dir_block->b_data;
1132 * First create "." and ".." and then copy the dir information
1133 * back to the block.
1135 de = (struct ext4_dir_entry_2 *)target;
1136 de = ext4_init_dot_dotdot(inode, de,
1137 inode->i_sb->s_blocksize, csum_size,
1138 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1139 header_size = (void *)de - target;
1141 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1142 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1144 if (ext4_has_metadata_csum(inode->i_sb))
1145 csum_size = sizeof(struct ext4_dir_entry_tail);
1147 inode->i_size = inode->i_sb->s_blocksize;
1148 i_size_write(inode, inode->i_sb->s_blocksize);
1149 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1150 ext4_update_final_de(dir_block->b_data,
1151 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1152 inode->i_sb->s_blocksize - csum_size);
1155 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1156 inode->i_sb->s_blocksize);
1157 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1159 set_buffer_uptodate(dir_block);
1160 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1163 set_buffer_verified(dir_block);
1164 return ext4_mark_inode_dirty(handle, inode);
1167 static int ext4_convert_inline_data_nolock(handle_t *handle,
1168 struct inode *inode,
1169 struct ext4_iloc *iloc)
1173 struct buffer_head *data_bh = NULL;
1174 struct ext4_map_blocks map;
1177 inline_size = ext4_get_inline_size(inode);
1178 buf = kmalloc(inline_size, GFP_NOFS);
1184 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1189 * Make sure the inline directory entries pass checks before we try to
1190 * convert them, so that we avoid touching stuff that needs fsck.
1192 if (S_ISDIR(inode->i_mode)) {
1193 error = ext4_check_all_de(inode, iloc->bh,
1194 buf + EXT4_INLINE_DOTDOT_SIZE,
1195 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1200 error = ext4_destroy_inline_data_nolock(handle, inode);
1207 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1210 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1215 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1221 lock_buffer(data_bh);
1222 error = ext4_journal_get_create_access(handle, data_bh);
1224 unlock_buffer(data_bh);
1228 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1230 if (!S_ISDIR(inode->i_mode)) {
1231 memcpy(data_bh->b_data, buf, inline_size);
1232 set_buffer_uptodate(data_bh);
1233 error = ext4_handle_dirty_metadata(handle,
1236 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1240 unlock_buffer(data_bh);
1243 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1252 * Try to add the new entry to the inline data.
1253 * If succeeds, return 0. If not, extended the inline dir and copied data to
1254 * the new created block.
1256 int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1257 struct inode *dir, struct inode *inode)
1259 int ret, inline_size, no_expand;
1261 struct ext4_iloc iloc;
1263 ret = ext4_get_inode_loc(dir, &iloc);
1267 ext4_write_lock_xattr(dir, &no_expand);
1268 if (!ext4_has_inline_data(dir))
1271 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1272 EXT4_INLINE_DOTDOT_SIZE;
1273 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1275 ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1276 inline_start, inline_size);
1280 /* check whether it can be inserted to inline xattr space. */
1281 inline_size = EXT4_I(dir)->i_inline_size -
1282 EXT4_MIN_INLINE_DATA_SIZE;
1284 /* Try to use the xattr space.*/
1285 ret = ext4_update_inline_dir(handle, dir, &iloc);
1286 if (ret && ret != -ENOSPC)
1289 inline_size = EXT4_I(dir)->i_inline_size -
1290 EXT4_MIN_INLINE_DATA_SIZE;
1294 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1296 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1297 inode, &iloc, inline_start,
1305 * The inline space is filled up, so create a new block for it.
1306 * As the extent tree will be created, we have to save the inline
1309 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1312 ext4_write_unlock_xattr(dir, &no_expand);
1313 ext4_mark_inode_dirty(handle, dir);
1319 * This function fills a red-black tree with information from an
1320 * inlined dir. It returns the number directory entries loaded
1321 * into the tree. If there is an error it is returned in err.
1323 int htree_inlinedir_to_tree(struct file *dir_file,
1324 struct inode *dir, ext4_lblk_t block,
1325 struct dx_hash_info *hinfo,
1326 __u32 start_hash, __u32 start_minor_hash,
1327 int *has_inline_data)
1329 int err = 0, count = 0;
1330 unsigned int parent_ino;
1332 struct ext4_dir_entry_2 *de;
1333 struct inode *inode = file_inode(dir_file);
1334 int ret, inline_size = 0;
1335 struct ext4_iloc iloc;
1336 void *dir_buf = NULL;
1337 struct ext4_dir_entry_2 fake;
1338 struct fscrypt_str tmp_str;
1340 ret = ext4_get_inode_loc(inode, &iloc);
1344 down_read(&EXT4_I(inode)->xattr_sem);
1345 if (!ext4_has_inline_data(inode)) {
1346 up_read(&EXT4_I(inode)->xattr_sem);
1347 *has_inline_data = 0;
1351 inline_size = ext4_get_inline_size(inode);
1352 dir_buf = kmalloc(inline_size, GFP_NOFS);
1355 up_read(&EXT4_I(inode)->xattr_sem);
1359 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1360 up_read(&EXT4_I(inode)->xattr_sem);
1365 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1366 while (pos < inline_size) {
1368 * As inlined dir doesn't store any information about '.' and
1369 * only the inode number of '..' is stored, we have to handle
1373 fake.inode = cpu_to_le32(inode->i_ino);
1375 strcpy(fake.name, ".");
1376 fake.rec_len = ext4_rec_len_to_disk(
1377 EXT4_DIR_REC_LEN(fake.name_len),
1379 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1381 pos = EXT4_INLINE_DOTDOT_OFFSET;
1382 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1383 fake.inode = cpu_to_le32(parent_ino);
1385 strcpy(fake.name, "..");
1386 fake.rec_len = ext4_rec_len_to_disk(
1387 EXT4_DIR_REC_LEN(fake.name_len),
1389 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1391 pos = EXT4_INLINE_DOTDOT_SIZE;
1393 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1394 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1395 if (ext4_check_dir_entry(inode, dir_file, de,
1397 inline_size, pos)) {
1403 ext4fs_dirhash(de->name, de->name_len, hinfo);
1404 if ((hinfo->hash < start_hash) ||
1405 ((hinfo->hash == start_hash) &&
1406 (hinfo->minor_hash < start_minor_hash)))
1410 tmp_str.name = de->name;
1411 tmp_str.len = de->name_len;
1412 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1413 hinfo->minor_hash, de, &tmp_str);
1428 * So this function is called when the volume is mkfsed with
1429 * dir_index disabled. In order to keep f_pos persistent
1430 * after we convert from an inlined dir to a blocked based,
1431 * we just pretend that we are a normal dir and return the
1432 * offset as if '.' and '..' really take place.
1435 int ext4_read_inline_dir(struct file *file,
1436 struct dir_context *ctx,
1437 int *has_inline_data)
1439 unsigned int offset, parent_ino;
1441 struct ext4_dir_entry_2 *de;
1442 struct super_block *sb;
1443 struct inode *inode = file_inode(file);
1444 int ret, inline_size = 0;
1445 struct ext4_iloc iloc;
1446 void *dir_buf = NULL;
1447 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1449 ret = ext4_get_inode_loc(inode, &iloc);
1453 down_read(&EXT4_I(inode)->xattr_sem);
1454 if (!ext4_has_inline_data(inode)) {
1455 up_read(&EXT4_I(inode)->xattr_sem);
1456 *has_inline_data = 0;
1460 inline_size = ext4_get_inline_size(inode);
1461 dir_buf = kmalloc(inline_size, GFP_NOFS);
1464 up_read(&EXT4_I(inode)->xattr_sem);
1468 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1469 up_read(&EXT4_I(inode)->xattr_sem);
1475 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1479 * dotdot_offset and dotdot_size is the real offset and
1480 * size for ".." and "." if the dir is block based while
1481 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1482 * So we will use extra_offset and extra_size to indicate them
1483 * during the inline dir iteration.
1485 dotdot_offset = EXT4_DIR_REC_LEN(1);
1486 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1487 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1488 extra_size = extra_offset + inline_size;
1491 * If the version has changed since the last call to
1492 * readdir(2), then we might be pointing to an invalid
1493 * dirent right now. Scan from the start of the inline
1496 if (!inode_eq_iversion(inode, file->f_version)) {
1497 for (i = 0; i < extra_size && i < offset;) {
1499 * "." is with offset 0 and
1500 * ".." is dotdot_offset.
1505 } else if (i == dotdot_offset) {
1509 /* for other entry, the real offset in
1510 * the buf has to be tuned accordingly.
1512 de = (struct ext4_dir_entry_2 *)
1513 (dir_buf + i - extra_offset);
1514 /* It's too expensive to do a full
1515 * dirent test each time round this
1516 * loop, but we do have to test at
1517 * least that it is non-zero. A
1518 * failure will be detected in the
1519 * dirent test below. */
1520 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1521 < EXT4_DIR_REC_LEN(1))
1523 i += ext4_rec_len_from_disk(de->rec_len,
1528 file->f_version = inode_query_iversion(inode);
1531 while (ctx->pos < extra_size) {
1532 if (ctx->pos == 0) {
1533 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1535 ctx->pos = dotdot_offset;
1539 if (ctx->pos == dotdot_offset) {
1540 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1542 ctx->pos = dotdot_size;
1546 de = (struct ext4_dir_entry_2 *)
1547 (dir_buf + ctx->pos - extra_offset);
1548 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1549 extra_size, ctx->pos))
1551 if (le32_to_cpu(de->inode)) {
1552 if (!dir_emit(ctx, de->name, de->name_len,
1553 le32_to_cpu(de->inode),
1554 get_dtype(sb, de->file_type)))
1557 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1565 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1566 struct ext4_dir_entry_2 **parent_de,
1569 struct ext4_iloc iloc;
1571 *retval = ext4_get_inode_loc(inode, &iloc);
1575 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1581 * Try to create the inline data for the new dir.
1582 * If it succeeds, return 0, otherwise return the error.
1583 * In case of ENOSPC, the caller should create the normal disk layout dir.
1585 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1586 struct inode *inode)
1588 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1589 struct ext4_iloc iloc;
1590 struct ext4_dir_entry_2 *de;
1592 ret = ext4_get_inode_loc(inode, &iloc);
1596 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1601 * For inline dir, we only save the inode information for the ".."
1602 * and create a fake dentry to cover the left space.
1604 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1605 de->inode = cpu_to_le32(parent->i_ino);
1606 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1608 de->rec_len = ext4_rec_len_to_disk(
1609 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1611 set_nlink(inode, 2);
1612 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1618 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1619 struct ext4_filename *fname,
1620 struct ext4_dir_entry_2 **res_dir,
1621 int *has_inline_data)
1624 struct ext4_iloc iloc;
1628 if (ext4_get_inode_loc(dir, &iloc))
1631 down_read(&EXT4_I(dir)->xattr_sem);
1632 if (!ext4_has_inline_data(dir)) {
1633 *has_inline_data = 0;
1637 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1638 EXT4_INLINE_DOTDOT_SIZE;
1639 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1640 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1641 dir, fname, 0, res_dir);
1647 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1650 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1651 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1653 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1654 dir, fname, 0, res_dir);
1662 up_read(&EXT4_I(dir)->xattr_sem);
1666 int ext4_delete_inline_entry(handle_t *handle,
1668 struct ext4_dir_entry_2 *de_del,
1669 struct buffer_head *bh,
1670 int *has_inline_data)
1672 int err, inline_size, no_expand;
1673 struct ext4_iloc iloc;
1676 err = ext4_get_inode_loc(dir, &iloc);
1680 ext4_write_lock_xattr(dir, &no_expand);
1681 if (!ext4_has_inline_data(dir)) {
1682 *has_inline_data = 0;
1686 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1687 EXT4_MIN_INLINE_DATA_SIZE) {
1688 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1689 EXT4_INLINE_DOTDOT_SIZE;
1690 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1691 EXT4_INLINE_DOTDOT_SIZE;
1693 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1694 inline_size = ext4_get_inline_size(dir) -
1695 EXT4_MIN_INLINE_DATA_SIZE;
1698 BUFFER_TRACE(bh, "get_write_access");
1699 err = ext4_journal_get_write_access(handle, bh);
1703 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1704 inline_start, inline_size, 0);
1708 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1710 ext4_write_unlock_xattr(dir, &no_expand);
1711 if (likely(err == 0))
1712 err = ext4_mark_inode_dirty(handle, dir);
1715 ext4_std_error(dir->i_sb, err);
1720 * Get the inline dentry at offset.
1722 static inline struct ext4_dir_entry_2 *
1723 ext4_get_inline_entry(struct inode *inode,
1724 struct ext4_iloc *iloc,
1725 unsigned int offset,
1726 void **inline_start,
1731 BUG_ON(offset > ext4_get_inline_size(inode));
1733 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1734 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1735 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1737 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1738 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1739 *inline_size = ext4_get_inline_size(inode) -
1740 EXT4_MIN_INLINE_DATA_SIZE;
1744 *inline_start = inline_pos;
1745 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1748 bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1750 int err, inline_size;
1751 struct ext4_iloc iloc;
1753 unsigned int offset;
1754 struct ext4_dir_entry_2 *de;
1757 err = ext4_get_inode_loc(dir, &iloc);
1759 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1764 down_read(&EXT4_I(dir)->xattr_sem);
1765 if (!ext4_has_inline_data(dir)) {
1766 *has_inline_data = 0;
1770 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1771 if (!le32_to_cpu(de->inode)) {
1772 ext4_warning(dir->i_sb,
1773 "bad inline directory (dir #%lu) - no `..'",
1779 offset = EXT4_INLINE_DOTDOT_SIZE;
1780 while (offset < dir->i_size) {
1781 de = ext4_get_inline_entry(dir, &iloc, offset,
1782 &inline_pos, &inline_size);
1783 if (ext4_check_dir_entry(dir, NULL, de,
1784 iloc.bh, inline_pos,
1785 inline_size, offset)) {
1786 ext4_warning(dir->i_sb,
1787 "bad inline directory (dir #%lu) - "
1788 "inode %u, rec_len %u, name_len %d"
1790 dir->i_ino, le32_to_cpu(de->inode),
1791 le16_to_cpu(de->rec_len), de->name_len,
1796 if (le32_to_cpu(de->inode)) {
1800 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1804 up_read(&EXT4_I(dir)->xattr_sem);
1809 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1813 ext4_write_lock_xattr(inode, &no_expand);
1814 ret = ext4_destroy_inline_data_nolock(handle, inode);
1815 ext4_write_unlock_xattr(inode, &no_expand);
1820 int ext4_inline_data_iomap(struct inode *inode, struct iomap *iomap)
1823 int error = -EAGAIN;
1824 struct ext4_iloc iloc;
1826 down_read(&EXT4_I(inode)->xattr_sem);
1827 if (!ext4_has_inline_data(inode))
1830 error = ext4_get_inode_loc(inode, &iloc);
1834 addr = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1835 addr += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1836 addr += offsetof(struct ext4_inode, i_block);
1842 iomap->length = min_t(loff_t, ext4_get_inline_size(inode),
1843 i_size_read(inode));
1844 iomap->type = IOMAP_INLINE;
1848 up_read(&EXT4_I(inode)->xattr_sem);
1852 int ext4_inline_data_fiemap(struct inode *inode,
1853 struct fiemap_extent_info *fieinfo,
1854 int *has_inline, __u64 start, __u64 len)
1858 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1861 struct ext4_iloc iloc;
1863 down_read(&EXT4_I(inode)->xattr_sem);
1864 if (!ext4_has_inline_data(inode)) {
1868 inline_len = min_t(size_t, ext4_get_inline_size(inode),
1869 i_size_read(inode));
1870 if (start >= inline_len)
1872 if (start + len < inline_len)
1873 inline_len = start + len;
1874 inline_len -= start;
1876 error = ext4_get_inode_loc(inode, &iloc);
1880 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1881 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1882 physical += offsetof(struct ext4_inode, i_block);
1885 error = fiemap_fill_next_extent(fieinfo, start, physical,
1889 up_read(&EXT4_I(inode)->xattr_sem);
1890 return (error < 0 ? error : 0);
1894 * Called during xattr set, and if we can sparse space 'needed',
1895 * just create the extent tree evict the data to the outer block.
1897 * We use jbd2 instead of page cache to move data to the 1st block
1898 * so that the whole transaction can be committed as a whole and
1899 * the data isn't lost because of the delayed page cache write.
1901 int ext4_try_to_evict_inline_data(handle_t *handle,
1902 struct inode *inode,
1906 struct ext4_xattr_entry *entry;
1907 struct ext4_inode *raw_inode;
1908 struct ext4_iloc iloc;
1910 error = ext4_get_inode_loc(inode, &iloc);
1914 raw_inode = ext4_raw_inode(&iloc);
1915 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1916 EXT4_I(inode)->i_inline_off);
1917 if (EXT4_XATTR_LEN(entry->e_name_len) +
1918 EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1923 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1929 int ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1932 int inline_size, value_len, needed_blocks, no_expand, err = 0;
1935 struct ext4_xattr_ibody_find is = {
1936 .s = { .not_found = -ENODATA, },
1938 struct ext4_xattr_info i = {
1939 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1940 .name = EXT4_XATTR_SYSTEM_DATA,
1944 needed_blocks = ext4_writepage_trans_blocks(inode);
1945 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1947 return PTR_ERR(handle);
1949 ext4_write_lock_xattr(inode, &no_expand);
1950 if (!ext4_has_inline_data(inode)) {
1952 ext4_journal_stop(handle);
1956 if ((err = ext4_orphan_add(handle, inode)) != 0)
1959 if ((err = ext4_get_inode_loc(inode, &is.iloc)) != 0)
1962 down_write(&EXT4_I(inode)->i_data_sem);
1963 i_size = inode->i_size;
1964 inline_size = ext4_get_inline_size(inode);
1965 EXT4_I(inode)->i_disksize = i_size;
1967 if (i_size < inline_size) {
1968 /* Clear the content in the xattr space. */
1969 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1970 if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0)
1973 BUG_ON(is.s.not_found);
1975 value_len = le32_to_cpu(is.s.here->e_value_size);
1976 value = kmalloc(value_len, GFP_NOFS);
1982 err = ext4_xattr_ibody_get(inode, i.name_index,
1983 i.name, value, value_len);
1988 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1989 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1990 err = ext4_xattr_ibody_inline_set(handle, inode,
1996 /* Clear the content within i_blocks. */
1997 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1998 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1999 memset(p + i_size, 0,
2000 EXT4_MIN_INLINE_DATA_SIZE - i_size);
2003 EXT4_I(inode)->i_inline_size = i_size <
2004 EXT4_MIN_INLINE_DATA_SIZE ?
2005 EXT4_MIN_INLINE_DATA_SIZE : i_size;
2009 up_write(&EXT4_I(inode)->i_data_sem);
2012 ext4_write_unlock_xattr(inode, &no_expand);
2015 ext4_orphan_del(handle, inode);
2018 inode->i_mtime = inode->i_ctime = current_time(inode);
2019 err = ext4_mark_inode_dirty(handle, inode);
2021 ext4_handle_sync(handle);
2023 ext4_journal_stop(handle);
2027 int ext4_convert_inline_data(struct inode *inode)
2029 int error, needed_blocks, no_expand;
2031 struct ext4_iloc iloc;
2033 if (!ext4_has_inline_data(inode)) {
2034 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
2038 needed_blocks = ext4_writepage_trans_blocks(inode);
2041 error = ext4_get_inode_loc(inode, &iloc);
2045 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2046 if (IS_ERR(handle)) {
2047 error = PTR_ERR(handle);
2051 ext4_write_lock_xattr(inode, &no_expand);
2052 if (ext4_has_inline_data(inode))
2053 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2054 ext4_write_unlock_xattr(inode, &no_expand);
2055 ext4_journal_stop(handle);