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_block && 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 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 BUG_ON(!EXT4_I(inode)->i_inline_off);
219 BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
221 raw_inode = ext4_raw_inode(iloc);
224 if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
225 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
226 EXT4_MIN_INLINE_DATA_SIZE - pos : len;
227 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
237 pos -= EXT4_MIN_INLINE_DATA_SIZE;
238 header = IHDR(inode, raw_inode);
239 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
240 EXT4_I(inode)->i_inline_off);
242 memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
246 static int ext4_create_inline_data(handle_t *handle,
247 struct inode *inode, unsigned len)
251 struct ext4_xattr_ibody_find is = {
252 .s = { .not_found = -ENODATA, },
254 struct ext4_xattr_info i = {
255 .name_index = EXT4_XATTR_INDEX_SYSTEM,
256 .name = EXT4_XATTR_SYSTEM_DATA,
259 error = ext4_get_inode_loc(inode, &is.iloc);
263 BUFFER_TRACE(is.iloc.bh, "get_write_access");
264 error = ext4_journal_get_write_access(handle, is.iloc.bh);
268 if (len > EXT4_MIN_INLINE_DATA_SIZE) {
269 value = EXT4_ZERO_XATTR_VALUE;
270 len -= EXT4_MIN_INLINE_DATA_SIZE;
276 /* Insert the the xttr entry. */
280 error = ext4_xattr_ibody_find(inode, &i, &is);
284 BUG_ON(!is.s.not_found);
286 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
288 if (error == -ENOSPC)
289 ext4_clear_inode_state(inode,
290 EXT4_STATE_MAY_INLINE_DATA);
294 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
295 0, EXT4_MIN_INLINE_DATA_SIZE);
297 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
298 (void *)ext4_raw_inode(&is.iloc));
299 EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
300 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
301 ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
303 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
310 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
315 struct ext4_xattr_ibody_find is = {
316 .s = { .not_found = -ENODATA, },
318 struct ext4_xattr_info i = {
319 .name_index = EXT4_XATTR_INDEX_SYSTEM,
320 .name = EXT4_XATTR_SYSTEM_DATA,
323 /* If the old space is ok, write the data directly. */
324 if (len <= EXT4_I(inode)->i_inline_size)
327 error = ext4_get_inode_loc(inode, &is.iloc);
331 error = ext4_xattr_ibody_find(inode, &i, &is);
335 BUG_ON(is.s.not_found);
337 len -= EXT4_MIN_INLINE_DATA_SIZE;
338 value = kzalloc(len, GFP_NOFS);
344 error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
346 if (error == -ENODATA)
349 BUFFER_TRACE(is.iloc.bh, "get_write_access");
350 error = ext4_journal_get_write_access(handle, is.iloc.bh);
354 /* Update the xttr entry. */
358 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
362 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
363 (void *)ext4_raw_inode(&is.iloc));
364 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
365 le32_to_cpu(is.s.here->e_value_size);
366 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
368 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
376 static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
379 int ret, size, no_expand;
380 struct ext4_inode_info *ei = EXT4_I(inode);
382 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
385 size = ext4_get_max_inline_size(inode);
389 ext4_write_lock_xattr(inode, &no_expand);
391 if (ei->i_inline_off)
392 ret = ext4_update_inline_data(handle, inode, len);
394 ret = ext4_create_inline_data(handle, inode, len);
396 ext4_write_unlock_xattr(inode, &no_expand);
400 static int ext4_destroy_inline_data_nolock(handle_t *handle,
403 struct ext4_inode_info *ei = EXT4_I(inode);
404 struct ext4_xattr_ibody_find is = {
405 .s = { .not_found = 0, },
407 struct ext4_xattr_info i = {
408 .name_index = EXT4_XATTR_INDEX_SYSTEM,
409 .name = EXT4_XATTR_SYSTEM_DATA,
415 if (!ei->i_inline_off)
418 error = ext4_get_inode_loc(inode, &is.iloc);
422 error = ext4_xattr_ibody_find(inode, &i, &is);
426 BUFFER_TRACE(is.iloc.bh, "get_write_access");
427 error = ext4_journal_get_write_access(handle, is.iloc.bh);
431 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
435 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
436 0, EXT4_MIN_INLINE_DATA_SIZE);
437 memset(ei->i_data, 0, EXT4_MIN_INLINE_DATA_SIZE);
439 if (ext4_has_feature_extents(inode->i_sb)) {
440 if (S_ISDIR(inode->i_mode) ||
441 S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
442 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
443 ext4_ext_tree_init(handle, inode);
446 ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
449 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
451 EXT4_I(inode)->i_inline_off = 0;
452 EXT4_I(inode)->i_inline_size = 0;
453 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
456 if (error == -ENODATA)
461 static int ext4_read_inline_page(struct inode *inode, struct page *page)
466 struct ext4_iloc iloc;
468 BUG_ON(!PageLocked(page));
469 BUG_ON(!ext4_has_inline_data(inode));
472 if (!EXT4_I(inode)->i_inline_off) {
473 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
478 ret = ext4_get_inode_loc(inode, &iloc);
482 len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
483 kaddr = kmap_atomic(page);
484 ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
485 flush_dcache_page(page);
486 kunmap_atomic(kaddr);
487 zero_user_segment(page, len, PAGE_CACHE_SIZE);
488 SetPageUptodate(page);
495 int ext4_readpage_inline(struct inode *inode, struct page *page)
499 down_read(&EXT4_I(inode)->xattr_sem);
500 if (!ext4_has_inline_data(inode)) {
501 up_read(&EXT4_I(inode)->xattr_sem);
506 * Current inline data can only exist in the 1st page,
507 * So for all the other pages, just set them uptodate.
510 ret = ext4_read_inline_page(inode, page);
511 else if (!PageUptodate(page)) {
512 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
513 SetPageUptodate(page);
516 up_read(&EXT4_I(inode)->xattr_sem);
519 return ret >= 0 ? 0 : ret;
522 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
526 int ret, needed_blocks, no_expand;
527 handle_t *handle = NULL;
528 int retries = 0, sem_held = 0;
529 struct page *page = NULL;
531 struct ext4_iloc iloc;
533 if (!ext4_has_inline_data(inode)) {
535 * clear the flag so that no new write
536 * will trap here again.
538 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
542 needed_blocks = ext4_writepage_trans_blocks(inode);
544 ret = ext4_get_inode_loc(inode, &iloc);
549 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
550 if (IS_ERR(handle)) {
551 ret = PTR_ERR(handle);
556 /* We cannot recurse into the filesystem as the transaction is already
558 flags |= AOP_FLAG_NOFS;
560 page = grab_cache_page_write_begin(mapping, 0, flags);
566 ext4_write_lock_xattr(inode, &no_expand);
568 /* If some one has already done this for us, just exit. */
569 if (!ext4_has_inline_data(inode)) {
575 to = ext4_get_inline_size(inode);
576 if (!PageUptodate(page)) {
577 ret = ext4_read_inline_page(inode, page);
582 ret = ext4_destroy_inline_data_nolock(handle, inode);
586 if (ext4_should_dioread_nolock(inode))
587 ret = __block_write_begin(page, from, to, ext4_get_block_write);
589 ret = __block_write_begin(page, from, to, ext4_get_block);
591 if (!ret && ext4_should_journal_data(inode)) {
592 ret = ext4_walk_page_buffers(handle, page_buffers(page),
594 do_journal_get_write_access);
599 page_cache_release(page);
601 ext4_orphan_add(handle, inode);
602 ext4_write_unlock_xattr(inode, &no_expand);
604 ext4_journal_stop(handle);
606 ext4_truncate_failed_write(inode);
608 * If truncate failed early the inode might
609 * still be on the orphan list; we need to
610 * make sure the inode is removed from the
611 * orphan list in that case.
614 ext4_orphan_del(NULL, inode);
617 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
621 block_commit_write(page, from, to);
625 page_cache_release(page);
628 ext4_write_unlock_xattr(inode, &no_expand);
630 ext4_journal_stop(handle);
636 * Try to write data in the inode.
637 * If the inode has inline data, check whether the new write can be
638 * in the inode also. If not, create the page the handle, move the data
639 * to the page make it update and let the later codes create extent for it.
641 int ext4_try_to_write_inline_data(struct address_space *mapping,
643 loff_t pos, unsigned len,
650 struct ext4_iloc iloc;
652 if (pos + len > ext4_get_max_inline_size(inode))
655 ret = ext4_get_inode_loc(inode, &iloc);
660 * The possible write could happen in the inode,
661 * so try to reserve the space in inode first.
663 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
664 if (IS_ERR(handle)) {
665 ret = PTR_ERR(handle);
670 ret = ext4_prepare_inline_data(handle, inode, pos + len);
671 if (ret && ret != -ENOSPC)
674 /* We don't have space in inline inode, so convert it to extent. */
675 if (ret == -ENOSPC) {
676 ext4_journal_stop(handle);
681 ret = ext4_journal_get_write_access(handle, iloc.bh);
685 flags |= AOP_FLAG_NOFS;
687 page = grab_cache_page_write_begin(mapping, 0, flags);
694 down_read(&EXT4_I(inode)->xattr_sem);
695 if (!ext4_has_inline_data(inode)) {
698 page_cache_release(page);
702 if (!PageUptodate(page)) {
703 ret = ext4_read_inline_page(inode, page);
714 up_read(&EXT4_I(inode)->xattr_sem);
716 if (handle && (ret != 1))
717 ext4_journal_stop(handle);
721 return ext4_convert_inline_data_to_extent(mapping,
725 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
726 unsigned copied, struct page *page)
730 struct ext4_iloc iloc;
732 if (unlikely(copied < len)) {
733 if (!PageUptodate(page)) {
739 ret = ext4_get_inode_loc(inode, &iloc);
741 ext4_std_error(inode->i_sb, ret);
746 ext4_write_lock_xattr(inode, &no_expand);
747 BUG_ON(!ext4_has_inline_data(inode));
750 * ei->i_inline_off may have changed since ext4_write_begin()
751 * called ext4_try_to_write_inline_data()
753 (void) ext4_find_inline_data_nolock(inode);
755 kaddr = kmap_atomic(page);
756 ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
757 kunmap_atomic(kaddr);
758 SetPageUptodate(page);
759 /* clear page dirty so that writepages wouldn't work for us. */
760 ClearPageDirty(page);
762 ext4_write_unlock_xattr(inode, &no_expand);
764 mark_inode_dirty(inode);
770 ext4_journalled_write_inline_data(struct inode *inode,
776 struct ext4_iloc iloc;
778 ret = ext4_get_inode_loc(inode, &iloc);
780 ext4_std_error(inode->i_sb, ret);
784 ext4_write_lock_xattr(inode, &no_expand);
785 kaddr = kmap_atomic(page);
786 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
787 kunmap_atomic(kaddr);
788 ext4_write_unlock_xattr(inode, &no_expand);
794 * Try to make the page cache and handle ready for the inline data case.
795 * We can call this function in 2 cases:
796 * 1. The inode is created and the first write exceeds inline size. We can
797 * clear the inode state safely.
798 * 2. The inode has inline data, then we need to read the data, make it
799 * update and dirty so that ext4_da_writepages can handle it. We don't
800 * need to start the journal since the file's metatdata isn't changed now.
802 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
807 int ret = 0, inline_size;
810 page = grab_cache_page_write_begin(mapping, 0, flags);
814 down_read(&EXT4_I(inode)->xattr_sem);
815 if (!ext4_has_inline_data(inode)) {
816 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
820 inline_size = ext4_get_inline_size(inode);
822 if (!PageUptodate(page)) {
823 ret = ext4_read_inline_page(inode, page);
828 ret = __block_write_begin(page, 0, inline_size,
829 ext4_da_get_block_prep);
831 up_read(&EXT4_I(inode)->xattr_sem);
833 page_cache_release(page);
834 ext4_truncate_failed_write(inode);
839 SetPageUptodate(page);
840 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
841 *fsdata = (void *)CONVERT_INLINE_DATA;
844 up_read(&EXT4_I(inode)->xattr_sem);
847 page_cache_release(page);
853 * Prepare the write for the inline data.
854 * If the the data can be written into the inode, we just read
855 * the page and make it uptodate, and start the journal.
856 * Otherwise read the page, makes it dirty so that it can be
857 * handle in writepages(the i_disksize update is left to the
858 * normal ext4_da_write_end).
860 int ext4_da_write_inline_data_begin(struct address_space *mapping,
862 loff_t pos, unsigned len,
867 int ret, inline_size;
870 struct ext4_iloc iloc;
873 ret = ext4_get_inode_loc(inode, &iloc);
878 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
879 if (IS_ERR(handle)) {
880 ret = PTR_ERR(handle);
884 inline_size = ext4_get_max_inline_size(inode);
887 if (inline_size >= pos + len) {
888 ret = ext4_prepare_inline_data(handle, inode, pos + len);
889 if (ret && ret != -ENOSPC)
894 * We cannot recurse into the filesystem as the transaction
895 * is already started.
897 flags |= AOP_FLAG_NOFS;
899 if (ret == -ENOSPC) {
900 ext4_journal_stop(handle);
901 ret = ext4_da_convert_inline_data_to_extent(mapping,
905 if (ret == -ENOSPC &&
906 ext4_should_retry_alloc(inode->i_sb, &retries))
911 page = grab_cache_page_write_begin(mapping, 0, flags);
917 down_read(&EXT4_I(inode)->xattr_sem);
918 if (!ext4_has_inline_data(inode)) {
920 goto out_release_page;
923 if (!PageUptodate(page)) {
924 ret = ext4_read_inline_page(inode, page);
926 goto out_release_page;
928 ret = ext4_journal_get_write_access(handle, iloc.bh);
930 goto out_release_page;
932 up_read(&EXT4_I(inode)->xattr_sem);
937 up_read(&EXT4_I(inode)->xattr_sem);
939 page_cache_release(page);
941 ext4_journal_stop(handle);
947 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
948 unsigned len, unsigned copied,
953 ret = ext4_write_inline_data_end(inode, pos, len, copied, page);
962 * No need to use i_size_read() here, the i_size
963 * cannot change under us because we hold i_mutex.
965 * But it's important to update i_size while still holding page lock:
966 * page writeout could otherwise come in and zero beyond i_size.
968 if (pos+copied > inode->i_size)
969 i_size_write(inode, pos+copied);
971 page_cache_release(page);
974 * Don't mark the inode dirty under page lock. First, it unnecessarily
975 * makes the holding time of page lock longer. Second, it forces lock
976 * ordering of page lock and transaction start for journaling
979 mark_inode_dirty(inode);
984 #ifdef INLINE_DIR_DEBUG
985 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
986 void *inline_start, int inline_size)
989 unsigned short de_len;
990 struct ext4_dir_entry_2 *de = inline_start;
991 void *dlimit = inline_start + inline_size;
993 trace_printk("inode %lu\n", dir->i_ino);
995 while ((void *)de < dlimit) {
996 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
997 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
998 offset, de_len, de->name_len, de->name,
999 de->name_len, le32_to_cpu(de->inode));
1000 if (ext4_check_dir_entry(dir, NULL, de, bh,
1001 inline_start, inline_size, offset))
1005 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1009 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1013 * Add a new entry into a inline dir.
1014 * It will return -ENOSPC if no space is available, and -EIO
1015 * and -EEXIST if directory entry already exists.
1017 static int ext4_add_dirent_to_inline(handle_t *handle,
1018 struct ext4_filename *fname,
1019 struct dentry *dentry,
1020 struct inode *inode,
1021 struct ext4_iloc *iloc,
1022 void *inline_start, int inline_size)
1024 struct inode *dir = d_inode(dentry->d_parent);
1026 struct ext4_dir_entry_2 *de;
1028 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1029 inline_size, fname, &de);
1033 BUFFER_TRACE(iloc->bh, "get_write_access");
1034 err = ext4_journal_get_write_access(handle, iloc->bh);
1037 ext4_insert_dentry(dir, inode, de, inline_size, fname);
1039 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1042 * XXX shouldn't update any times until successful
1043 * completion of syscall, but too many callers depend
1046 * XXX similarly, too many callers depend on
1047 * ext4_new_inode() setting the times, but error
1048 * recovery deletes the inode, so the worst that can
1049 * happen is that the times are slightly out of date
1050 * and/or different from the directory change time.
1052 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1053 ext4_update_dx_flag(dir);
1055 ext4_mark_inode_dirty(handle, dir);
1059 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1060 struct ext4_iloc *iloc)
1062 struct ext4_xattr_entry *entry;
1063 struct ext4_xattr_ibody_header *header;
1065 BUG_ON(!EXT4_I(inode)->i_inline_off);
1067 header = IHDR(inode, ext4_raw_inode(iloc));
1068 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1069 EXT4_I(inode)->i_inline_off);
1071 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1074 /* Set the final de to cover the whole block. */
1075 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1077 struct ext4_dir_entry_2 *de, *prev_de;
1081 de = (struct ext4_dir_entry_2 *)de_buf;
1083 limit = de_buf + old_size;
1086 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1088 de = (struct ext4_dir_entry_2 *)de_buf;
1089 } while (de_buf < limit);
1091 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1092 old_size, new_size);
1094 /* this is just created, so create an empty entry. */
1096 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1100 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1101 struct ext4_iloc *iloc)
1104 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1105 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1107 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1110 ret = ext4_update_inline_data(handle, dir,
1111 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1115 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1116 EXT4_I(dir)->i_inline_size -
1117 EXT4_MIN_INLINE_DATA_SIZE);
1118 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1122 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1123 struct ext4_iloc *iloc,
1124 void *buf, int inline_size)
1126 ext4_create_inline_data(handle, inode, inline_size);
1127 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1128 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1131 static int ext4_finish_convert_inline_dir(handle_t *handle,
1132 struct inode *inode,
1133 struct buffer_head *dir_block,
1137 int err, csum_size = 0, header_size = 0;
1138 struct ext4_dir_entry_2 *de;
1139 struct ext4_dir_entry_tail *t;
1140 void *target = dir_block->b_data;
1143 * First create "." and ".." and then copy the dir information
1144 * back to the block.
1146 de = (struct ext4_dir_entry_2 *)target;
1147 de = ext4_init_dot_dotdot(inode, de,
1148 inode->i_sb->s_blocksize, csum_size,
1149 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1150 header_size = (void *)de - target;
1152 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1153 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1155 if (ext4_has_metadata_csum(inode->i_sb))
1156 csum_size = sizeof(struct ext4_dir_entry_tail);
1158 inode->i_size = inode->i_sb->s_blocksize;
1159 i_size_write(inode, inode->i_sb->s_blocksize);
1160 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1161 ext4_update_final_de(dir_block->b_data,
1162 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1163 inode->i_sb->s_blocksize - csum_size);
1166 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1167 inode->i_sb->s_blocksize);
1168 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1170 set_buffer_uptodate(dir_block);
1171 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1174 set_buffer_verified(dir_block);
1175 return ext4_mark_inode_dirty(handle, inode);
1178 static int ext4_convert_inline_data_nolock(handle_t *handle,
1179 struct inode *inode,
1180 struct ext4_iloc *iloc)
1184 struct buffer_head *data_bh = NULL;
1185 struct ext4_map_blocks map;
1188 inline_size = ext4_get_inline_size(inode);
1189 buf = kmalloc(inline_size, GFP_NOFS);
1195 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1200 * Make sure the inline directory entries pass checks before we try to
1201 * convert them, so that we avoid touching stuff that needs fsck.
1203 if (S_ISDIR(inode->i_mode)) {
1204 error = ext4_check_all_de(inode, iloc->bh,
1205 buf + EXT4_INLINE_DOTDOT_SIZE,
1206 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1211 error = ext4_destroy_inline_data_nolock(handle, inode);
1218 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1221 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1226 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1232 lock_buffer(data_bh);
1233 error = ext4_journal_get_create_access(handle, data_bh);
1235 unlock_buffer(data_bh);
1239 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1241 if (!S_ISDIR(inode->i_mode)) {
1242 memcpy(data_bh->b_data, buf, inline_size);
1243 set_buffer_uptodate(data_bh);
1244 error = ext4_handle_dirty_metadata(handle,
1247 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1251 unlock_buffer(data_bh);
1254 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1263 * Try to add the new entry to the inline data.
1264 * If succeeds, return 0. If not, extended the inline dir and copied data to
1265 * the new created block.
1267 int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1268 struct dentry *dentry, struct inode *inode)
1270 int ret, inline_size, no_expand;
1272 struct ext4_iloc iloc;
1273 struct inode *dir = d_inode(dentry->d_parent);
1275 ret = ext4_get_inode_loc(dir, &iloc);
1279 ext4_write_lock_xattr(dir, &no_expand);
1280 if (!ext4_has_inline_data(dir))
1283 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1284 EXT4_INLINE_DOTDOT_SIZE;
1285 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1287 ret = ext4_add_dirent_to_inline(handle, fname, dentry, inode, &iloc,
1288 inline_start, inline_size);
1292 /* check whether it can be inserted to inline xattr space. */
1293 inline_size = EXT4_I(dir)->i_inline_size -
1294 EXT4_MIN_INLINE_DATA_SIZE;
1296 /* Try to use the xattr space.*/
1297 ret = ext4_update_inline_dir(handle, dir, &iloc);
1298 if (ret && ret != -ENOSPC)
1301 inline_size = EXT4_I(dir)->i_inline_size -
1302 EXT4_MIN_INLINE_DATA_SIZE;
1306 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1308 ret = ext4_add_dirent_to_inline(handle, fname, dentry,
1309 inode, &iloc, inline_start,
1317 * The inline space is filled up, so create a new block for it.
1318 * As the extent tree will be created, we have to save the inline
1321 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1324 ext4_mark_inode_dirty(handle, dir);
1325 ext4_write_unlock_xattr(dir, &no_expand);
1331 * This function fills a red-black tree with information from an
1332 * inlined dir. It returns the number directory entries loaded
1333 * into the tree. If there is an error it is returned in err.
1335 int htree_inlinedir_to_tree(struct file *dir_file,
1336 struct inode *dir, ext4_lblk_t block,
1337 struct dx_hash_info *hinfo,
1338 __u32 start_hash, __u32 start_minor_hash,
1339 int *has_inline_data)
1341 int err = 0, count = 0;
1342 unsigned int parent_ino;
1344 struct ext4_dir_entry_2 *de;
1345 struct inode *inode = file_inode(dir_file);
1346 int ret, inline_size = 0;
1347 struct ext4_iloc iloc;
1348 void *dir_buf = NULL;
1349 struct ext4_dir_entry_2 fake;
1350 struct ext4_str tmp_str;
1352 ret = ext4_get_inode_loc(inode, &iloc);
1356 down_read(&EXT4_I(inode)->xattr_sem);
1357 if (!ext4_has_inline_data(inode)) {
1358 up_read(&EXT4_I(inode)->xattr_sem);
1359 *has_inline_data = 0;
1363 inline_size = ext4_get_inline_size(inode);
1364 dir_buf = kmalloc(inline_size, GFP_NOFS);
1367 up_read(&EXT4_I(inode)->xattr_sem);
1371 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1372 up_read(&EXT4_I(inode)->xattr_sem);
1377 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1378 while (pos < inline_size) {
1380 * As inlined dir doesn't store any information about '.' and
1381 * only the inode number of '..' is stored, we have to handle
1385 fake.inode = cpu_to_le32(inode->i_ino);
1387 strcpy(fake.name, ".");
1388 fake.rec_len = ext4_rec_len_to_disk(
1389 EXT4_DIR_REC_LEN(fake.name_len),
1391 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1393 pos = EXT4_INLINE_DOTDOT_OFFSET;
1394 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1395 fake.inode = cpu_to_le32(parent_ino);
1397 strcpy(fake.name, "..");
1398 fake.rec_len = ext4_rec_len_to_disk(
1399 EXT4_DIR_REC_LEN(fake.name_len),
1401 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1403 pos = EXT4_INLINE_DOTDOT_SIZE;
1405 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1406 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1407 if (ext4_check_dir_entry(inode, dir_file, de,
1409 inline_size, pos)) {
1415 ext4fs_dirhash(de->name, de->name_len, hinfo);
1416 if ((hinfo->hash < start_hash) ||
1417 ((hinfo->hash == start_hash) &&
1418 (hinfo->minor_hash < start_minor_hash)))
1422 tmp_str.name = de->name;
1423 tmp_str.len = de->name_len;
1424 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1425 hinfo->minor_hash, de, &tmp_str);
1440 * So this function is called when the volume is mkfsed with
1441 * dir_index disabled. In order to keep f_pos persistent
1442 * after we convert from an inlined dir to a blocked based,
1443 * we just pretend that we are a normal dir and return the
1444 * offset as if '.' and '..' really take place.
1447 int ext4_read_inline_dir(struct file *file,
1448 struct dir_context *ctx,
1449 int *has_inline_data)
1451 unsigned int offset, parent_ino;
1453 struct ext4_dir_entry_2 *de;
1454 struct super_block *sb;
1455 struct inode *inode = file_inode(file);
1456 int ret, inline_size = 0;
1457 struct ext4_iloc iloc;
1458 void *dir_buf = NULL;
1459 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1461 ret = ext4_get_inode_loc(inode, &iloc);
1465 down_read(&EXT4_I(inode)->xattr_sem);
1466 if (!ext4_has_inline_data(inode)) {
1467 up_read(&EXT4_I(inode)->xattr_sem);
1468 *has_inline_data = 0;
1472 inline_size = ext4_get_inline_size(inode);
1473 dir_buf = kmalloc(inline_size, GFP_NOFS);
1476 up_read(&EXT4_I(inode)->xattr_sem);
1480 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1481 up_read(&EXT4_I(inode)->xattr_sem);
1487 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1491 * dotdot_offset and dotdot_size is the real offset and
1492 * size for ".." and "." if the dir is block based while
1493 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1494 * So we will use extra_offset and extra_size to indicate them
1495 * during the inline dir iteration.
1497 dotdot_offset = EXT4_DIR_REC_LEN(1);
1498 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1499 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1500 extra_size = extra_offset + inline_size;
1503 * If the version has changed since the last call to
1504 * readdir(2), then we might be pointing to an invalid
1505 * dirent right now. Scan from the start of the inline
1508 if (file->f_version != inode->i_version) {
1509 for (i = 0; i < extra_size && i < offset;) {
1511 * "." is with offset 0 and
1512 * ".." is dotdot_offset.
1517 } else if (i == dotdot_offset) {
1521 /* for other entry, the real offset in
1522 * the buf has to be tuned accordingly.
1524 de = (struct ext4_dir_entry_2 *)
1525 (dir_buf + i - extra_offset);
1526 /* It's too expensive to do a full
1527 * dirent test each time round this
1528 * loop, but we do have to test at
1529 * least that it is non-zero. A
1530 * failure will be detected in the
1531 * dirent test below. */
1532 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1533 < EXT4_DIR_REC_LEN(1))
1535 i += ext4_rec_len_from_disk(de->rec_len,
1540 file->f_version = inode->i_version;
1543 while (ctx->pos < extra_size) {
1544 if (ctx->pos == 0) {
1545 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1547 ctx->pos = dotdot_offset;
1551 if (ctx->pos == dotdot_offset) {
1552 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1554 ctx->pos = dotdot_size;
1558 de = (struct ext4_dir_entry_2 *)
1559 (dir_buf + ctx->pos - extra_offset);
1560 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1561 extra_size, ctx->pos))
1563 if (le32_to_cpu(de->inode)) {
1564 if (!dir_emit(ctx, de->name, de->name_len,
1565 le32_to_cpu(de->inode),
1566 get_dtype(sb, de->file_type)))
1569 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1577 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1578 struct ext4_dir_entry_2 **parent_de,
1581 struct ext4_iloc iloc;
1583 *retval = ext4_get_inode_loc(inode, &iloc);
1587 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1593 * Try to create the inline data for the new dir.
1594 * If it succeeds, return 0, otherwise return the error.
1595 * In case of ENOSPC, the caller should create the normal disk layout dir.
1597 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1598 struct inode *inode)
1600 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1601 struct ext4_iloc iloc;
1602 struct ext4_dir_entry_2 *de;
1604 ret = ext4_get_inode_loc(inode, &iloc);
1608 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1613 * For inline dir, we only save the inode information for the ".."
1614 * and create a fake dentry to cover the left space.
1616 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1617 de->inode = cpu_to_le32(parent->i_ino);
1618 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1620 de->rec_len = ext4_rec_len_to_disk(
1621 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1623 set_nlink(inode, 2);
1624 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1630 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1631 struct ext4_filename *fname,
1632 const struct qstr *d_name,
1633 struct ext4_dir_entry_2 **res_dir,
1634 int *has_inline_data)
1637 struct ext4_iloc iloc;
1641 if (ext4_get_inode_loc(dir, &iloc))
1644 down_read(&EXT4_I(dir)->xattr_sem);
1645 if (!ext4_has_inline_data(dir)) {
1646 *has_inline_data = 0;
1650 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1651 EXT4_INLINE_DOTDOT_SIZE;
1652 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1653 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1654 dir, fname, d_name, 0, res_dir);
1660 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1663 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1664 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1666 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1667 dir, fname, d_name, 0, res_dir);
1675 up_read(&EXT4_I(dir)->xattr_sem);
1679 int ext4_delete_inline_entry(handle_t *handle,
1681 struct ext4_dir_entry_2 *de_del,
1682 struct buffer_head *bh,
1683 int *has_inline_data)
1685 int err, inline_size, no_expand;
1686 struct ext4_iloc iloc;
1689 err = ext4_get_inode_loc(dir, &iloc);
1693 ext4_write_lock_xattr(dir, &no_expand);
1694 if (!ext4_has_inline_data(dir)) {
1695 *has_inline_data = 0;
1699 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1700 EXT4_MIN_INLINE_DATA_SIZE) {
1701 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1702 EXT4_INLINE_DOTDOT_SIZE;
1703 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1704 EXT4_INLINE_DOTDOT_SIZE;
1706 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1707 inline_size = ext4_get_inline_size(dir) -
1708 EXT4_MIN_INLINE_DATA_SIZE;
1711 BUFFER_TRACE(bh, "get_write_access");
1712 err = ext4_journal_get_write_access(handle, bh);
1716 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1717 inline_start, inline_size, 0);
1721 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1722 err = ext4_mark_inode_dirty(handle, dir);
1726 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1728 ext4_write_unlock_xattr(dir, &no_expand);
1731 ext4_std_error(dir->i_sb, err);
1736 * Get the inline dentry at offset.
1738 static inline struct ext4_dir_entry_2 *
1739 ext4_get_inline_entry(struct inode *inode,
1740 struct ext4_iloc *iloc,
1741 unsigned int offset,
1742 void **inline_start,
1747 BUG_ON(offset > ext4_get_inline_size(inode));
1749 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1750 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1751 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1753 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1754 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1755 *inline_size = ext4_get_inline_size(inode) -
1756 EXT4_MIN_INLINE_DATA_SIZE;
1760 *inline_start = inline_pos;
1761 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1764 int empty_inline_dir(struct inode *dir, int *has_inline_data)
1766 int err, inline_size;
1767 struct ext4_iloc iloc;
1770 unsigned int offset;
1771 struct ext4_dir_entry_2 *de;
1774 err = ext4_get_inode_loc(dir, &iloc);
1776 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1781 down_read(&EXT4_I(dir)->xattr_sem);
1782 if (!ext4_has_inline_data(dir)) {
1783 *has_inline_data = 0;
1787 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1788 if (!le32_to_cpu(de->inode)) {
1789 ext4_warning(dir->i_sb,
1790 "bad inline directory (dir #%lu) - no `..'",
1796 inline_len = ext4_get_inline_size(dir);
1797 offset = EXT4_INLINE_DOTDOT_SIZE;
1798 while (offset < inline_len) {
1799 de = ext4_get_inline_entry(dir, &iloc, offset,
1800 &inline_pos, &inline_size);
1801 if (ext4_check_dir_entry(dir, NULL, de,
1802 iloc.bh, inline_pos,
1803 inline_size, offset)) {
1804 ext4_warning(dir->i_sb,
1805 "bad inline directory (dir #%lu) - "
1806 "inode %u, rec_len %u, name_len %d"
1808 dir->i_ino, le32_to_cpu(de->inode),
1809 le16_to_cpu(de->rec_len), de->name_len,
1814 if (le32_to_cpu(de->inode)) {
1818 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1822 up_read(&EXT4_I(dir)->xattr_sem);
1827 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1831 ext4_write_lock_xattr(inode, &no_expand);
1832 ret = ext4_destroy_inline_data_nolock(handle, inode);
1833 ext4_write_unlock_xattr(inode, &no_expand);
1838 int ext4_inline_data_fiemap(struct inode *inode,
1839 struct fiemap_extent_info *fieinfo,
1840 int *has_inline, __u64 start, __u64 len)
1844 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1847 struct ext4_iloc iloc;
1849 down_read(&EXT4_I(inode)->xattr_sem);
1850 if (!ext4_has_inline_data(inode)) {
1854 inline_len = min_t(size_t, ext4_get_inline_size(inode),
1855 i_size_read(inode));
1856 if (start >= inline_len)
1858 if (start + len < inline_len)
1859 inline_len = start + len;
1860 inline_len -= start;
1862 error = ext4_get_inode_loc(inode, &iloc);
1866 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1867 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1868 physical += offsetof(struct ext4_inode, i_block);
1872 up_read(&EXT4_I(inode)->xattr_sem);
1874 error = fiemap_fill_next_extent(fieinfo, start, physical,
1876 return (error < 0 ? error : 0);
1879 void ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1882 int inline_size, value_len, needed_blocks, no_expand;
1885 struct ext4_xattr_ibody_find is = {
1886 .s = { .not_found = -ENODATA, },
1888 struct ext4_xattr_info i = {
1889 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1890 .name = EXT4_XATTR_SYSTEM_DATA,
1894 needed_blocks = ext4_writepage_trans_blocks(inode);
1895 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1899 ext4_write_lock_xattr(inode, &no_expand);
1900 if (!ext4_has_inline_data(inode)) {
1901 ext4_write_unlock_xattr(inode, &no_expand);
1903 ext4_journal_stop(handle);
1907 if (ext4_orphan_add(handle, inode))
1910 if (ext4_get_inode_loc(inode, &is.iloc))
1913 down_write(&EXT4_I(inode)->i_data_sem);
1914 i_size = inode->i_size;
1915 inline_size = ext4_get_inline_size(inode);
1916 EXT4_I(inode)->i_disksize = i_size;
1918 if (i_size < inline_size) {
1919 /* Clear the content in the xattr space. */
1920 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1921 if (ext4_xattr_ibody_find(inode, &i, &is))
1924 BUG_ON(is.s.not_found);
1926 value_len = le32_to_cpu(is.s.here->e_value_size);
1927 value = kmalloc(value_len, GFP_NOFS);
1931 if (ext4_xattr_ibody_get(inode, i.name_index, i.name,
1936 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1937 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1938 if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is))
1942 /* Clear the content within i_blocks. */
1943 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1944 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1945 memset(p + i_size, 0,
1946 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1949 EXT4_I(inode)->i_inline_size = i_size <
1950 EXT4_MIN_INLINE_DATA_SIZE ?
1951 EXT4_MIN_INLINE_DATA_SIZE : i_size;
1955 up_write(&EXT4_I(inode)->i_data_sem);
1958 ext4_write_unlock_xattr(inode, &no_expand);
1961 ext4_orphan_del(handle, inode);
1963 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1964 ext4_mark_inode_dirty(handle, inode);
1966 ext4_handle_sync(handle);
1968 ext4_journal_stop(handle);
1972 int ext4_convert_inline_data(struct inode *inode)
1974 int error, needed_blocks, no_expand;
1976 struct ext4_iloc iloc;
1978 if (!ext4_has_inline_data(inode)) {
1979 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1983 needed_blocks = ext4_writepage_trans_blocks(inode);
1986 error = ext4_get_inode_loc(inode, &iloc);
1990 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1991 if (IS_ERR(handle)) {
1992 error = PTR_ERR(handle);
1996 ext4_write_lock_xattr(inode, &no_expand);
1997 if (ext4_has_inline_data(inode))
1998 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1999 ext4_write_unlock_xattr(inode, &no_expand);
2000 ext4_journal_stop(handle);