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"
21 #include <trace/events/android_fs.h>
23 #define EXT4_XATTR_SYSTEM_DATA "data"
24 #define EXT4_MIN_INLINE_DATA_SIZE ((sizeof(__le32) * EXT4_N_BLOCKS))
25 #define EXT4_INLINE_DOTDOT_OFFSET 2
26 #define EXT4_INLINE_DOTDOT_SIZE 4
28 static int ext4_get_inline_size(struct inode *inode)
30 if (EXT4_I(inode)->i_inline_off)
31 return EXT4_I(inode)->i_inline_size;
36 static int get_max_inline_xattr_value_size(struct inode *inode,
37 struct ext4_iloc *iloc)
39 struct ext4_xattr_ibody_header *header;
40 struct ext4_xattr_entry *entry;
41 struct ext4_inode *raw_inode;
44 min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
45 EXT4_GOOD_OLD_INODE_SIZE -
46 EXT4_I(inode)->i_extra_isize -
47 sizeof(struct ext4_xattr_ibody_header);
50 * We need to subtract another sizeof(__u32) since an in-inode xattr
51 * needs an empty 4 bytes to indicate the gap between the xattr entry
52 * and the name/value pair.
54 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
55 return EXT4_XATTR_SIZE(min_offs -
56 EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
57 EXT4_XATTR_ROUND - sizeof(__u32));
59 raw_inode = ext4_raw_inode(iloc);
60 header = IHDR(inode, raw_inode);
61 entry = IFIRST(header);
63 /* Compute min_offs. */
64 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
65 if (!entry->e_value_block && entry->e_value_size) {
66 size_t offs = le16_to_cpu(entry->e_value_offs);
72 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
74 if (EXT4_I(inode)->i_inline_off) {
75 entry = (struct ext4_xattr_entry *)
76 ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
78 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
82 free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
84 if (free > EXT4_XATTR_ROUND)
85 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
94 * Get the maximum size we now can store in an inode.
95 * If we can't find the space for a xattr entry, don't use the space
96 * of the extents since we have no space to indicate the inline data.
98 int ext4_get_max_inline_size(struct inode *inode)
100 int error, max_inline_size;
101 struct ext4_iloc iloc;
103 if (EXT4_I(inode)->i_extra_isize == 0)
106 error = ext4_get_inode_loc(inode, &iloc);
108 ext4_error_inode(inode, __func__, __LINE__, 0,
109 "can't get inode location %lu",
114 down_read(&EXT4_I(inode)->xattr_sem);
115 max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
116 up_read(&EXT4_I(inode)->xattr_sem);
120 if (!max_inline_size)
123 return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
127 * this function does not take xattr_sem, which is OK because it is
128 * currently only used in a code path coming form ext4_iget, before
129 * the new inode has been unlocked
131 int ext4_find_inline_data_nolock(struct inode *inode)
133 struct ext4_xattr_ibody_find is = {
134 .s = { .not_found = -ENODATA, },
136 struct ext4_xattr_info i = {
137 .name_index = EXT4_XATTR_INDEX_SYSTEM,
138 .name = EXT4_XATTR_SYSTEM_DATA,
142 if (EXT4_I(inode)->i_extra_isize == 0)
145 error = ext4_get_inode_loc(inode, &is.iloc);
149 error = ext4_xattr_ibody_find(inode, &i, &is);
153 if (!is.s.not_found) {
154 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
155 (void *)ext4_raw_inode(&is.iloc));
156 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
157 le32_to_cpu(is.s.here->e_value_size);
158 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
165 static int ext4_read_inline_data(struct inode *inode, void *buffer,
167 struct ext4_iloc *iloc)
169 struct ext4_xattr_entry *entry;
170 struct ext4_xattr_ibody_header *header;
172 struct ext4_inode *raw_inode;
177 BUG_ON(len > EXT4_I(inode)->i_inline_size);
179 cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
180 len : EXT4_MIN_INLINE_DATA_SIZE;
182 raw_inode = ext4_raw_inode(iloc);
183 memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
191 header = IHDR(inode, raw_inode);
192 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
193 EXT4_I(inode)->i_inline_off);
194 len = min_t(unsigned int, len,
195 (unsigned int)le32_to_cpu(entry->e_value_size));
198 (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
206 * write the buffer to the inline inode.
207 * If 'create' is set, we don't need to do the extra copy in the xattr
208 * value since it is already handled by ext4_xattr_ibody_inline_set.
209 * That saves us one memcpy.
211 static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
212 void *buffer, loff_t pos, unsigned int len)
214 struct ext4_xattr_entry *entry;
215 struct ext4_xattr_ibody_header *header;
216 struct ext4_inode *raw_inode;
219 BUG_ON(!EXT4_I(inode)->i_inline_off);
220 BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
222 raw_inode = ext4_raw_inode(iloc);
225 if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
226 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
227 EXT4_MIN_INLINE_DATA_SIZE - pos : len;
228 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
238 pos -= EXT4_MIN_INLINE_DATA_SIZE;
239 header = IHDR(inode, raw_inode);
240 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
241 EXT4_I(inode)->i_inline_off);
243 memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
247 static int ext4_create_inline_data(handle_t *handle,
248 struct inode *inode, unsigned len)
252 struct ext4_xattr_ibody_find is = {
253 .s = { .not_found = -ENODATA, },
255 struct ext4_xattr_info i = {
256 .name_index = EXT4_XATTR_INDEX_SYSTEM,
257 .name = EXT4_XATTR_SYSTEM_DATA,
260 error = ext4_get_inode_loc(inode, &is.iloc);
264 BUFFER_TRACE(is.iloc.bh, "get_write_access");
265 error = ext4_journal_get_write_access(handle, is.iloc.bh);
269 if (len > EXT4_MIN_INLINE_DATA_SIZE) {
270 value = EXT4_ZERO_XATTR_VALUE;
271 len -= EXT4_MIN_INLINE_DATA_SIZE;
277 /* Insert the the xttr entry. */
281 error = ext4_xattr_ibody_find(inode, &i, &is);
285 BUG_ON(!is.s.not_found);
287 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
289 if (error == -ENOSPC)
290 ext4_clear_inode_state(inode,
291 EXT4_STATE_MAY_INLINE_DATA);
295 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
296 0, EXT4_MIN_INLINE_DATA_SIZE);
298 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
299 (void *)ext4_raw_inode(&is.iloc));
300 EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
301 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
302 ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
304 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
311 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
316 struct ext4_xattr_ibody_find is = {
317 .s = { .not_found = -ENODATA, },
319 struct ext4_xattr_info i = {
320 .name_index = EXT4_XATTR_INDEX_SYSTEM,
321 .name = EXT4_XATTR_SYSTEM_DATA,
324 /* If the old space is ok, write the data directly. */
325 if (len <= EXT4_I(inode)->i_inline_size)
328 error = ext4_get_inode_loc(inode, &is.iloc);
332 error = ext4_xattr_ibody_find(inode, &i, &is);
336 BUG_ON(is.s.not_found);
338 len -= EXT4_MIN_INLINE_DATA_SIZE;
339 value = kzalloc(len, GFP_NOFS);
345 error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
347 if (error == -ENODATA)
350 BUFFER_TRACE(is.iloc.bh, "get_write_access");
351 error = ext4_journal_get_write_access(handle, is.iloc.bh);
355 /* Update the xttr entry. */
359 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
363 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
364 (void *)ext4_raw_inode(&is.iloc));
365 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
366 le32_to_cpu(is.s.here->e_value_size);
367 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
369 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
377 static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
380 int ret, size, no_expand;
381 struct ext4_inode_info *ei = EXT4_I(inode);
383 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
386 size = ext4_get_max_inline_size(inode);
390 ext4_write_lock_xattr(inode, &no_expand);
392 if (ei->i_inline_off)
393 ret = ext4_update_inline_data(handle, inode, len);
395 ret = ext4_create_inline_data(handle, inode, len);
397 ext4_write_unlock_xattr(inode, &no_expand);
401 static int ext4_destroy_inline_data_nolock(handle_t *handle,
404 struct ext4_inode_info *ei = EXT4_I(inode);
405 struct ext4_xattr_ibody_find is = {
406 .s = { .not_found = 0, },
408 struct ext4_xattr_info i = {
409 .name_index = EXT4_XATTR_INDEX_SYSTEM,
410 .name = EXT4_XATTR_SYSTEM_DATA,
416 if (!ei->i_inline_off)
419 error = ext4_get_inode_loc(inode, &is.iloc);
423 error = ext4_xattr_ibody_find(inode, &i, &is);
427 BUFFER_TRACE(is.iloc.bh, "get_write_access");
428 error = ext4_journal_get_write_access(handle, is.iloc.bh);
432 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
436 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
437 0, EXT4_MIN_INLINE_DATA_SIZE);
438 memset(ei->i_data, 0, EXT4_MIN_INLINE_DATA_SIZE);
440 if (ext4_has_feature_extents(inode->i_sb)) {
441 if (S_ISDIR(inode->i_mode) ||
442 S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
443 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
444 ext4_ext_tree_init(handle, inode);
447 ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
450 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
452 EXT4_I(inode)->i_inline_off = 0;
453 EXT4_I(inode)->i_inline_size = 0;
454 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
457 if (error == -ENODATA)
462 static int ext4_read_inline_page(struct inode *inode, struct page *page)
467 struct ext4_iloc iloc;
469 BUG_ON(!PageLocked(page));
470 BUG_ON(!ext4_has_inline_data(inode));
473 if (!EXT4_I(inode)->i_inline_off) {
474 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
479 ret = ext4_get_inode_loc(inode, &iloc);
483 len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
484 kaddr = kmap_atomic(page);
485 ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
486 flush_dcache_page(page);
487 kunmap_atomic(kaddr);
488 zero_user_segment(page, len, PAGE_SIZE);
489 SetPageUptodate(page);
496 int ext4_readpage_inline(struct inode *inode, struct page *page)
500 down_read(&EXT4_I(inode)->xattr_sem);
501 if (!ext4_has_inline_data(inode)) {
502 up_read(&EXT4_I(inode)->xattr_sem);
506 if (trace_android_fs_dataread_start_enabled()) {
507 char *path, pathbuf[MAX_TRACE_PATHBUF_LEN];
509 path = android_fstrace_get_pathname(pathbuf,
510 MAX_TRACE_PATHBUF_LEN,
512 trace_android_fs_dataread_start(inode, page_offset(page),
513 PAGE_SIZE, current->pid,
514 path, current->comm);
518 * Current inline data can only exist in the 1st page,
519 * So for all the other pages, just set them uptodate.
522 ret = ext4_read_inline_page(inode, page);
523 else if (!PageUptodate(page)) {
524 zero_user_segment(page, 0, PAGE_SIZE);
525 SetPageUptodate(page);
528 trace_android_fs_dataread_end(inode, page_offset(page), PAGE_SIZE);
530 up_read(&EXT4_I(inode)->xattr_sem);
533 return ret >= 0 ? 0 : ret;
536 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
540 int ret, needed_blocks, no_expand;
541 handle_t *handle = NULL;
542 int retries = 0, sem_held = 0;
543 struct page *page = NULL;
545 struct ext4_iloc iloc;
547 if (!ext4_has_inline_data(inode)) {
549 * clear the flag so that no new write
550 * will trap here again.
552 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
556 needed_blocks = ext4_writepage_trans_blocks(inode);
558 ret = ext4_get_inode_loc(inode, &iloc);
563 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
564 if (IS_ERR(handle)) {
565 ret = PTR_ERR(handle);
570 /* We cannot recurse into the filesystem as the transaction is already
572 flags |= AOP_FLAG_NOFS;
574 page = grab_cache_page_write_begin(mapping, 0, flags);
580 ext4_write_lock_xattr(inode, &no_expand);
582 /* If some one has already done this for us, just exit. */
583 if (!ext4_has_inline_data(inode)) {
589 to = ext4_get_inline_size(inode);
590 if (!PageUptodate(page)) {
591 ret = ext4_read_inline_page(inode, page);
596 ret = ext4_destroy_inline_data_nolock(handle, inode);
600 if (ext4_should_dioread_nolock(inode)) {
601 ret = __block_write_begin(page, from, to,
602 ext4_get_block_unwritten);
604 ret = __block_write_begin(page, from, to, ext4_get_block);
606 if (!ret && ext4_should_journal_data(inode)) {
607 ret = ext4_walk_page_buffers(handle, page_buffers(page),
609 do_journal_get_write_access);
616 ext4_orphan_add(handle, inode);
617 ext4_write_unlock_xattr(inode, &no_expand);
619 ext4_journal_stop(handle);
621 ext4_truncate_failed_write(inode);
623 * If truncate failed early the inode might
624 * still be on the orphan list; we need to
625 * make sure the inode is removed from the
626 * orphan list in that case.
629 ext4_orphan_del(NULL, inode);
632 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
636 block_commit_write(page, from, to);
643 ext4_write_unlock_xattr(inode, &no_expand);
645 ext4_journal_stop(handle);
651 * Try to write data in the inode.
652 * If the inode has inline data, check whether the new write can be
653 * in the inode also. If not, create the page the handle, move the data
654 * to the page make it update and let the later codes create extent for it.
656 int ext4_try_to_write_inline_data(struct address_space *mapping,
658 loff_t pos, unsigned len,
665 struct ext4_iloc iloc;
667 if (pos + len > ext4_get_max_inline_size(inode))
670 ret = ext4_get_inode_loc(inode, &iloc);
675 * The possible write could happen in the inode,
676 * so try to reserve the space in inode first.
678 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
679 if (IS_ERR(handle)) {
680 ret = PTR_ERR(handle);
685 ret = ext4_prepare_inline_data(handle, inode, pos + len);
686 if (ret && ret != -ENOSPC)
689 /* We don't have space in inline inode, so convert it to extent. */
690 if (ret == -ENOSPC) {
691 ext4_journal_stop(handle);
696 ret = ext4_journal_get_write_access(handle, iloc.bh);
700 flags |= AOP_FLAG_NOFS;
702 page = grab_cache_page_write_begin(mapping, 0, flags);
709 down_read(&EXT4_I(inode)->xattr_sem);
710 if (!ext4_has_inline_data(inode)) {
717 if (!PageUptodate(page)) {
718 ret = ext4_read_inline_page(inode, page);
726 up_read(&EXT4_I(inode)->xattr_sem);
728 if (handle && (ret != 1))
729 ext4_journal_stop(handle);
733 return ext4_convert_inline_data_to_extent(mapping,
737 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
738 unsigned copied, struct page *page)
742 struct ext4_iloc iloc;
744 if (unlikely(copied < len)) {
745 if (!PageUptodate(page)) {
751 ret = ext4_get_inode_loc(inode, &iloc);
753 ext4_std_error(inode->i_sb, ret);
758 ext4_write_lock_xattr(inode, &no_expand);
759 BUG_ON(!ext4_has_inline_data(inode));
761 kaddr = kmap_atomic(page);
762 ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
763 kunmap_atomic(kaddr);
764 SetPageUptodate(page);
765 /* clear page dirty so that writepages wouldn't work for us. */
766 ClearPageDirty(page);
768 ext4_write_unlock_xattr(inode, &no_expand);
770 mark_inode_dirty(inode);
776 ext4_journalled_write_inline_data(struct inode *inode,
782 struct ext4_iloc iloc;
784 ret = ext4_get_inode_loc(inode, &iloc);
786 ext4_std_error(inode->i_sb, ret);
790 ext4_write_lock_xattr(inode, &no_expand);
791 kaddr = kmap_atomic(page);
792 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
793 kunmap_atomic(kaddr);
794 ext4_write_unlock_xattr(inode, &no_expand);
800 * Try to make the page cache and handle ready for the inline data case.
801 * We can call this function in 2 cases:
802 * 1. The inode is created and the first write exceeds inline size. We can
803 * clear the inode state safely.
804 * 2. The inode has inline data, then we need to read the data, make it
805 * update and dirty so that ext4_da_writepages can handle it. We don't
806 * need to start the journal since the file's metatdata isn't changed now.
808 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
813 int ret = 0, inline_size;
816 page = grab_cache_page_write_begin(mapping, 0, flags);
820 down_read(&EXT4_I(inode)->xattr_sem);
821 if (!ext4_has_inline_data(inode)) {
822 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
826 inline_size = ext4_get_inline_size(inode);
828 if (!PageUptodate(page)) {
829 ret = ext4_read_inline_page(inode, page);
834 ret = __block_write_begin(page, 0, inline_size,
835 ext4_da_get_block_prep);
837 up_read(&EXT4_I(inode)->xattr_sem);
840 ext4_truncate_failed_write(inode);
845 SetPageUptodate(page);
846 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
847 *fsdata = (void *)CONVERT_INLINE_DATA;
850 up_read(&EXT4_I(inode)->xattr_sem);
859 * Prepare the write for the inline data.
860 * If the the data can be written into the inode, we just read
861 * the page and make it uptodate, and start the journal.
862 * Otherwise read the page, makes it dirty so that it can be
863 * handle in writepages(the i_disksize update is left to the
864 * normal ext4_da_write_end).
866 int ext4_da_write_inline_data_begin(struct address_space *mapping,
868 loff_t pos, unsigned len,
873 int ret, inline_size;
876 struct ext4_iloc iloc;
879 ret = ext4_get_inode_loc(inode, &iloc);
884 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
885 if (IS_ERR(handle)) {
886 ret = PTR_ERR(handle);
890 inline_size = ext4_get_max_inline_size(inode);
893 if (inline_size >= pos + len) {
894 ret = ext4_prepare_inline_data(handle, inode, pos + len);
895 if (ret && ret != -ENOSPC)
900 * We cannot recurse into the filesystem as the transaction
901 * is already started.
903 flags |= AOP_FLAG_NOFS;
905 if (ret == -ENOSPC) {
906 ret = ext4_da_convert_inline_data_to_extent(mapping,
910 ext4_journal_stop(handle);
911 if (ret == -ENOSPC &&
912 ext4_should_retry_alloc(inode->i_sb, &retries))
917 page = grab_cache_page_write_begin(mapping, 0, flags);
923 down_read(&EXT4_I(inode)->xattr_sem);
924 if (!ext4_has_inline_data(inode)) {
926 goto out_release_page;
929 if (!PageUptodate(page)) {
930 ret = ext4_read_inline_page(inode, page);
932 goto out_release_page;
934 ret = ext4_journal_get_write_access(handle, iloc.bh);
936 goto out_release_page;
938 up_read(&EXT4_I(inode)->xattr_sem);
943 up_read(&EXT4_I(inode)->xattr_sem);
947 ext4_journal_stop(handle);
953 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
954 unsigned len, unsigned copied,
959 ret = ext4_write_inline_data_end(inode, pos, len, copied, page);
968 * No need to use i_size_read() here, the i_size
969 * cannot change under us because we hold i_mutex.
971 * But it's important to update i_size while still holding page lock:
972 * page writeout could otherwise come in and zero beyond i_size.
974 if (pos+copied > inode->i_size)
975 i_size_write(inode, pos+copied);
980 * Don't mark the inode dirty under page lock. First, it unnecessarily
981 * makes the holding time of page lock longer. Second, it forces lock
982 * ordering of page lock and transaction start for journaling
985 mark_inode_dirty(inode);
990 #ifdef INLINE_DIR_DEBUG
991 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
992 void *inline_start, int inline_size)
995 unsigned short de_len;
996 struct ext4_dir_entry_2 *de = inline_start;
997 void *dlimit = inline_start + inline_size;
999 trace_printk("inode %lu\n", dir->i_ino);
1001 while ((void *)de < dlimit) {
1002 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
1003 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
1004 offset, de_len, de->name_len, de->name,
1005 de->name_len, le32_to_cpu(de->inode));
1006 if (ext4_check_dir_entry(dir, NULL, de, bh,
1007 inline_start, inline_size, offset))
1011 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1015 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1019 * Add a new entry into a inline dir.
1020 * It will return -ENOSPC if no space is available, and -EIO
1021 * and -EEXIST if directory entry already exists.
1023 static int ext4_add_dirent_to_inline(handle_t *handle,
1024 struct ext4_filename *fname,
1026 struct inode *inode,
1027 struct ext4_iloc *iloc,
1028 void *inline_start, int inline_size)
1031 struct ext4_dir_entry_2 *de;
1033 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1034 inline_size, fname, &de);
1038 BUFFER_TRACE(iloc->bh, "get_write_access");
1039 err = ext4_journal_get_write_access(handle, iloc->bh);
1042 ext4_insert_dentry(inode, de, inline_size, fname);
1044 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1047 * XXX shouldn't update any times until successful
1048 * completion of syscall, but too many callers depend
1051 * XXX similarly, too many callers depend on
1052 * ext4_new_inode() setting the times, but error
1053 * recovery deletes the inode, so the worst that can
1054 * happen is that the times are slightly out of date
1055 * and/or different from the directory change time.
1057 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1058 ext4_update_dx_flag(dir);
1060 ext4_mark_inode_dirty(handle, dir);
1064 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1065 struct ext4_iloc *iloc)
1067 struct ext4_xattr_entry *entry;
1068 struct ext4_xattr_ibody_header *header;
1070 BUG_ON(!EXT4_I(inode)->i_inline_off);
1072 header = IHDR(inode, ext4_raw_inode(iloc));
1073 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1074 EXT4_I(inode)->i_inline_off);
1076 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1079 /* Set the final de to cover the whole block. */
1080 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1082 struct ext4_dir_entry_2 *de, *prev_de;
1086 de = (struct ext4_dir_entry_2 *)de_buf;
1088 limit = de_buf + old_size;
1091 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1093 de = (struct ext4_dir_entry_2 *)de_buf;
1094 } while (de_buf < limit);
1096 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1097 old_size, new_size);
1099 /* this is just created, so create an empty entry. */
1101 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1105 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1106 struct ext4_iloc *iloc)
1109 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1110 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1112 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1115 ret = ext4_update_inline_data(handle, dir,
1116 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1120 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1121 EXT4_I(dir)->i_inline_size -
1122 EXT4_MIN_INLINE_DATA_SIZE);
1123 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1127 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1128 struct ext4_iloc *iloc,
1129 void *buf, int inline_size)
1131 ext4_create_inline_data(handle, inode, inline_size);
1132 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1133 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1136 static int ext4_finish_convert_inline_dir(handle_t *handle,
1137 struct inode *inode,
1138 struct buffer_head *dir_block,
1142 int err, csum_size = 0, header_size = 0;
1143 struct ext4_dir_entry_2 *de;
1144 struct ext4_dir_entry_tail *t;
1145 void *target = dir_block->b_data;
1148 * First create "." and ".." and then copy the dir information
1149 * back to the block.
1151 de = (struct ext4_dir_entry_2 *)target;
1152 de = ext4_init_dot_dotdot(inode, de,
1153 inode->i_sb->s_blocksize, csum_size,
1154 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1155 header_size = (void *)de - target;
1157 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1158 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1160 if (ext4_has_metadata_csum(inode->i_sb))
1161 csum_size = sizeof(struct ext4_dir_entry_tail);
1163 inode->i_size = inode->i_sb->s_blocksize;
1164 i_size_write(inode, inode->i_sb->s_blocksize);
1165 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1166 ext4_update_final_de(dir_block->b_data,
1167 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1168 inode->i_sb->s_blocksize - csum_size);
1171 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1172 inode->i_sb->s_blocksize);
1173 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1175 set_buffer_uptodate(dir_block);
1176 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1179 set_buffer_verified(dir_block);
1180 return ext4_mark_inode_dirty(handle, inode);
1183 static int ext4_convert_inline_data_nolock(handle_t *handle,
1184 struct inode *inode,
1185 struct ext4_iloc *iloc)
1189 struct buffer_head *data_bh = NULL;
1190 struct ext4_map_blocks map;
1193 inline_size = ext4_get_inline_size(inode);
1194 buf = kmalloc(inline_size, GFP_NOFS);
1200 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1205 * Make sure the inline directory entries pass checks before we try to
1206 * convert them, so that we avoid touching stuff that needs fsck.
1208 if (S_ISDIR(inode->i_mode)) {
1209 error = ext4_check_all_de(inode, iloc->bh,
1210 buf + EXT4_INLINE_DOTDOT_SIZE,
1211 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1216 error = ext4_destroy_inline_data_nolock(handle, inode);
1223 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1226 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1231 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1237 lock_buffer(data_bh);
1238 error = ext4_journal_get_create_access(handle, data_bh);
1240 unlock_buffer(data_bh);
1244 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1246 if (!S_ISDIR(inode->i_mode)) {
1247 memcpy(data_bh->b_data, buf, inline_size);
1248 set_buffer_uptodate(data_bh);
1249 error = ext4_handle_dirty_metadata(handle,
1252 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1256 unlock_buffer(data_bh);
1259 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1268 * Try to add the new entry to the inline data.
1269 * If succeeds, return 0. If not, extended the inline dir and copied data to
1270 * the new created block.
1272 int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1273 struct inode *dir, struct inode *inode)
1275 int ret, inline_size, no_expand;
1277 struct ext4_iloc iloc;
1279 ret = ext4_get_inode_loc(dir, &iloc);
1283 ext4_write_lock_xattr(dir, &no_expand);
1284 if (!ext4_has_inline_data(dir))
1287 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1288 EXT4_INLINE_DOTDOT_SIZE;
1289 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1291 ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1292 inline_start, inline_size);
1296 /* check whether it can be inserted to inline xattr space. */
1297 inline_size = EXT4_I(dir)->i_inline_size -
1298 EXT4_MIN_INLINE_DATA_SIZE;
1300 /* Try to use the xattr space.*/
1301 ret = ext4_update_inline_dir(handle, dir, &iloc);
1302 if (ret && ret != -ENOSPC)
1305 inline_size = EXT4_I(dir)->i_inline_size -
1306 EXT4_MIN_INLINE_DATA_SIZE;
1310 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1312 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1313 inode, &iloc, inline_start,
1321 * The inline space is filled up, so create a new block for it.
1322 * As the extent tree will be created, we have to save the inline
1325 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1328 ext4_mark_inode_dirty(handle, dir);
1329 ext4_write_unlock_xattr(dir, &no_expand);
1335 * This function fills a red-black tree with information from an
1336 * inlined dir. It returns the number directory entries loaded
1337 * into the tree. If there is an error it is returned in err.
1339 int htree_inlinedir_to_tree(struct file *dir_file,
1340 struct inode *dir, ext4_lblk_t block,
1341 struct dx_hash_info *hinfo,
1342 __u32 start_hash, __u32 start_minor_hash,
1343 int *has_inline_data)
1345 int err = 0, count = 0;
1346 unsigned int parent_ino;
1348 struct ext4_dir_entry_2 *de;
1349 struct inode *inode = file_inode(dir_file);
1350 int ret, inline_size = 0;
1351 struct ext4_iloc iloc;
1352 void *dir_buf = NULL;
1353 struct ext4_dir_entry_2 fake;
1354 struct fscrypt_str tmp_str;
1356 ret = ext4_get_inode_loc(inode, &iloc);
1360 down_read(&EXT4_I(inode)->xattr_sem);
1361 if (!ext4_has_inline_data(inode)) {
1362 up_read(&EXT4_I(inode)->xattr_sem);
1363 *has_inline_data = 0;
1367 inline_size = ext4_get_inline_size(inode);
1368 dir_buf = kmalloc(inline_size, GFP_NOFS);
1371 up_read(&EXT4_I(inode)->xattr_sem);
1375 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1376 up_read(&EXT4_I(inode)->xattr_sem);
1381 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1382 while (pos < inline_size) {
1384 * As inlined dir doesn't store any information about '.' and
1385 * only the inode number of '..' is stored, we have to handle
1389 fake.inode = cpu_to_le32(inode->i_ino);
1391 strcpy(fake.name, ".");
1392 fake.rec_len = ext4_rec_len_to_disk(
1393 EXT4_DIR_REC_LEN(fake.name_len),
1395 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1397 pos = EXT4_INLINE_DOTDOT_OFFSET;
1398 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1399 fake.inode = cpu_to_le32(parent_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_SIZE;
1409 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1410 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1411 if (ext4_check_dir_entry(inode, dir_file, de,
1413 inline_size, pos)) {
1419 ext4fs_dirhash(de->name, de->name_len, hinfo);
1420 if ((hinfo->hash < start_hash) ||
1421 ((hinfo->hash == start_hash) &&
1422 (hinfo->minor_hash < start_minor_hash)))
1426 tmp_str.name = de->name;
1427 tmp_str.len = de->name_len;
1428 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1429 hinfo->minor_hash, de, &tmp_str);
1444 * So this function is called when the volume is mkfsed with
1445 * dir_index disabled. In order to keep f_pos persistent
1446 * after we convert from an inlined dir to a blocked based,
1447 * we just pretend that we are a normal dir and return the
1448 * offset as if '.' and '..' really take place.
1451 int ext4_read_inline_dir(struct file *file,
1452 struct dir_context *ctx,
1453 int *has_inline_data)
1455 unsigned int offset, parent_ino;
1457 struct ext4_dir_entry_2 *de;
1458 struct super_block *sb;
1459 struct inode *inode = file_inode(file);
1460 int ret, inline_size = 0;
1461 struct ext4_iloc iloc;
1462 void *dir_buf = NULL;
1463 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1465 ret = ext4_get_inode_loc(inode, &iloc);
1469 down_read(&EXT4_I(inode)->xattr_sem);
1470 if (!ext4_has_inline_data(inode)) {
1471 up_read(&EXT4_I(inode)->xattr_sem);
1472 *has_inline_data = 0;
1476 inline_size = ext4_get_inline_size(inode);
1477 dir_buf = kmalloc(inline_size, GFP_NOFS);
1480 up_read(&EXT4_I(inode)->xattr_sem);
1484 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1485 up_read(&EXT4_I(inode)->xattr_sem);
1491 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1495 * dotdot_offset and dotdot_size is the real offset and
1496 * size for ".." and "." if the dir is block based while
1497 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1498 * So we will use extra_offset and extra_size to indicate them
1499 * during the inline dir iteration.
1501 dotdot_offset = EXT4_DIR_REC_LEN(1);
1502 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1503 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1504 extra_size = extra_offset + inline_size;
1507 * If the version has changed since the last call to
1508 * readdir(2), then we might be pointing to an invalid
1509 * dirent right now. Scan from the start of the inline
1512 if (file->f_version != inode->i_version) {
1513 for (i = 0; i < extra_size && i < offset;) {
1515 * "." is with offset 0 and
1516 * ".." is dotdot_offset.
1521 } else if (i == dotdot_offset) {
1525 /* for other entry, the real offset in
1526 * the buf has to be tuned accordingly.
1528 de = (struct ext4_dir_entry_2 *)
1529 (dir_buf + i - extra_offset);
1530 /* It's too expensive to do a full
1531 * dirent test each time round this
1532 * loop, but we do have to test at
1533 * least that it is non-zero. A
1534 * failure will be detected in the
1535 * dirent test below. */
1536 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1537 < EXT4_DIR_REC_LEN(1))
1539 i += ext4_rec_len_from_disk(de->rec_len,
1544 file->f_version = inode->i_version;
1547 while (ctx->pos < extra_size) {
1548 if (ctx->pos == 0) {
1549 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1551 ctx->pos = dotdot_offset;
1555 if (ctx->pos == dotdot_offset) {
1556 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1558 ctx->pos = dotdot_size;
1562 de = (struct ext4_dir_entry_2 *)
1563 (dir_buf + ctx->pos - extra_offset);
1564 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1565 extra_size, ctx->pos))
1567 if (le32_to_cpu(de->inode)) {
1568 if (!dir_emit(ctx, de->name, de->name_len,
1569 le32_to_cpu(de->inode),
1570 get_dtype(sb, de->file_type)))
1573 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1581 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1582 struct ext4_dir_entry_2 **parent_de,
1585 struct ext4_iloc iloc;
1587 *retval = ext4_get_inode_loc(inode, &iloc);
1591 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1597 * Try to create the inline data for the new dir.
1598 * If it succeeds, return 0, otherwise return the error.
1599 * In case of ENOSPC, the caller should create the normal disk layout dir.
1601 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1602 struct inode *inode)
1604 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1605 struct ext4_iloc iloc;
1606 struct ext4_dir_entry_2 *de;
1608 ret = ext4_get_inode_loc(inode, &iloc);
1612 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1617 * For inline dir, we only save the inode information for the ".."
1618 * and create a fake dentry to cover the left space.
1620 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1621 de->inode = cpu_to_le32(parent->i_ino);
1622 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1624 de->rec_len = ext4_rec_len_to_disk(
1625 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1627 set_nlink(inode, 2);
1628 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1634 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1635 struct ext4_filename *fname,
1636 const struct qstr *d_name,
1637 struct ext4_dir_entry_2 **res_dir,
1638 int *has_inline_data)
1641 struct ext4_iloc iloc;
1645 if (ext4_get_inode_loc(dir, &iloc))
1648 down_read(&EXT4_I(dir)->xattr_sem);
1649 if (!ext4_has_inline_data(dir)) {
1650 *has_inline_data = 0;
1654 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1655 EXT4_INLINE_DOTDOT_SIZE;
1656 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1657 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1658 dir, fname, d_name, 0, res_dir);
1664 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1667 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1668 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1670 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1671 dir, fname, d_name, 0, res_dir);
1679 up_read(&EXT4_I(dir)->xattr_sem);
1683 int ext4_delete_inline_entry(handle_t *handle,
1685 struct ext4_dir_entry_2 *de_del,
1686 struct buffer_head *bh,
1687 int *has_inline_data)
1689 int err, inline_size, no_expand;
1690 struct ext4_iloc iloc;
1693 err = ext4_get_inode_loc(dir, &iloc);
1697 ext4_write_lock_xattr(dir, &no_expand);
1698 if (!ext4_has_inline_data(dir)) {
1699 *has_inline_data = 0;
1703 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1704 EXT4_MIN_INLINE_DATA_SIZE) {
1705 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1706 EXT4_INLINE_DOTDOT_SIZE;
1707 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1708 EXT4_INLINE_DOTDOT_SIZE;
1710 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1711 inline_size = ext4_get_inline_size(dir) -
1712 EXT4_MIN_INLINE_DATA_SIZE;
1715 BUFFER_TRACE(bh, "get_write_access");
1716 err = ext4_journal_get_write_access(handle, bh);
1720 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1721 inline_start, inline_size, 0);
1725 err = ext4_mark_inode_dirty(handle, dir);
1729 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1731 ext4_write_unlock_xattr(dir, &no_expand);
1734 ext4_std_error(dir->i_sb, err);
1739 * Get the inline dentry at offset.
1741 static inline struct ext4_dir_entry_2 *
1742 ext4_get_inline_entry(struct inode *inode,
1743 struct ext4_iloc *iloc,
1744 unsigned int offset,
1745 void **inline_start,
1750 BUG_ON(offset > ext4_get_inline_size(inode));
1752 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1753 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1754 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1756 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1757 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1758 *inline_size = ext4_get_inline_size(inode) -
1759 EXT4_MIN_INLINE_DATA_SIZE;
1763 *inline_start = inline_pos;
1764 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1767 bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1769 int err, inline_size;
1770 struct ext4_iloc iloc;
1772 unsigned int offset;
1773 struct ext4_dir_entry_2 *de;
1776 err = ext4_get_inode_loc(dir, &iloc);
1778 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1783 down_read(&EXT4_I(dir)->xattr_sem);
1784 if (!ext4_has_inline_data(dir)) {
1785 *has_inline_data = 0;
1789 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1790 if (!le32_to_cpu(de->inode)) {
1791 ext4_warning(dir->i_sb,
1792 "bad inline directory (dir #%lu) - no `..'",
1798 offset = EXT4_INLINE_DOTDOT_SIZE;
1799 while (offset < dir->i_size) {
1800 de = ext4_get_inline_entry(dir, &iloc, offset,
1801 &inline_pos, &inline_size);
1802 if (ext4_check_dir_entry(dir, NULL, de,
1803 iloc.bh, inline_pos,
1804 inline_size, offset)) {
1805 ext4_warning(dir->i_sb,
1806 "bad inline directory (dir #%lu) - "
1807 "inode %u, rec_len %u, name_len %d"
1809 dir->i_ino, le32_to_cpu(de->inode),
1810 le16_to_cpu(de->rec_len), de->name_len,
1815 if (le32_to_cpu(de->inode)) {
1819 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1823 up_read(&EXT4_I(dir)->xattr_sem);
1828 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1832 ext4_write_lock_xattr(inode, &no_expand);
1833 ret = ext4_destroy_inline_data_nolock(handle, inode);
1834 ext4_write_unlock_xattr(inode, &no_expand);
1839 int ext4_inline_data_fiemap(struct inode *inode,
1840 struct fiemap_extent_info *fieinfo,
1841 int *has_inline, __u64 start, __u64 len)
1845 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1848 struct ext4_iloc iloc;
1850 down_read(&EXT4_I(inode)->xattr_sem);
1851 if (!ext4_has_inline_data(inode)) {
1855 inline_len = min_t(size_t, ext4_get_inline_size(inode),
1856 i_size_read(inode));
1857 if (start >= inline_len)
1859 if (start + len < inline_len)
1860 inline_len = start + len;
1861 inline_len -= start;
1863 error = ext4_get_inode_loc(inode, &iloc);
1867 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1868 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1869 physical += offsetof(struct ext4_inode, i_block);
1872 error = fiemap_fill_next_extent(fieinfo, start, physical,
1876 up_read(&EXT4_I(inode)->xattr_sem);
1877 return (error < 0 ? error : 0);
1881 * Called during xattr set, and if we can sparse space 'needed',
1882 * just create the extent tree evict the data to the outer block.
1884 * We use jbd2 instead of page cache to move data to the 1st block
1885 * so that the whole transaction can be committed as a whole and
1886 * the data isn't lost because of the delayed page cache write.
1888 int ext4_try_to_evict_inline_data(handle_t *handle,
1889 struct inode *inode,
1893 struct ext4_xattr_entry *entry;
1894 struct ext4_inode *raw_inode;
1895 struct ext4_iloc iloc;
1897 error = ext4_get_inode_loc(inode, &iloc);
1901 raw_inode = ext4_raw_inode(&iloc);
1902 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1903 EXT4_I(inode)->i_inline_off);
1904 if (EXT4_XATTR_LEN(entry->e_name_len) +
1905 EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1910 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1916 void ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1919 int inline_size, value_len, needed_blocks, no_expand;
1922 struct ext4_xattr_ibody_find is = {
1923 .s = { .not_found = -ENODATA, },
1925 struct ext4_xattr_info i = {
1926 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1927 .name = EXT4_XATTR_SYSTEM_DATA,
1931 needed_blocks = ext4_writepage_trans_blocks(inode);
1932 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1936 ext4_write_lock_xattr(inode, &no_expand);
1937 if (!ext4_has_inline_data(inode)) {
1939 ext4_journal_stop(handle);
1943 if (ext4_orphan_add(handle, inode))
1946 if (ext4_get_inode_loc(inode, &is.iloc))
1949 down_write(&EXT4_I(inode)->i_data_sem);
1950 i_size = inode->i_size;
1951 inline_size = ext4_get_inline_size(inode);
1952 EXT4_I(inode)->i_disksize = i_size;
1954 if (i_size < inline_size) {
1955 /* Clear the content in the xattr space. */
1956 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1957 if (ext4_xattr_ibody_find(inode, &i, &is))
1960 BUG_ON(is.s.not_found);
1962 value_len = le32_to_cpu(is.s.here->e_value_size);
1963 value = kmalloc(value_len, GFP_NOFS);
1967 if (ext4_xattr_ibody_get(inode, i.name_index, i.name,
1972 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1973 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1974 if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is))
1978 /* Clear the content within i_blocks. */
1979 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1980 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1981 memset(p + i_size, 0,
1982 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1985 EXT4_I(inode)->i_inline_size = i_size <
1986 EXT4_MIN_INLINE_DATA_SIZE ?
1987 EXT4_MIN_INLINE_DATA_SIZE : i_size;
1991 up_write(&EXT4_I(inode)->i_data_sem);
1994 ext4_write_unlock_xattr(inode, &no_expand);
1997 ext4_orphan_del(handle, inode);
1999 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
2000 ext4_mark_inode_dirty(handle, inode);
2002 ext4_handle_sync(handle);
2004 ext4_journal_stop(handle);
2008 int ext4_convert_inline_data(struct inode *inode)
2010 int error, needed_blocks, no_expand;
2012 struct ext4_iloc iloc;
2014 if (!ext4_has_inline_data(inode)) {
2015 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
2019 needed_blocks = ext4_writepage_trans_blocks(inode);
2022 error = ext4_get_inode_loc(inode, &iloc);
2026 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2027 if (IS_ERR(handle)) {
2028 error = PTR_ERR(handle);
2032 ext4_write_lock_xattr(inode, &no_expand);
2033 if (ext4_has_inline_data(inode))
2034 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2035 ext4_write_unlock_xattr(inode, &no_expand);
2036 ext4_journal_stop(handle);