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,
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 down_write(&EXT4_I(inode)->xattr_sem);
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 up_write(&EXT4_I(inode)->xattr_sem);
402 static int ext4_destroy_inline_data_nolock(handle_t *handle,
405 struct ext4_inode_info *ei = EXT4_I(inode);
406 struct ext4_xattr_ibody_find is = {
407 .s = { .not_found = 0, },
409 struct ext4_xattr_info i = {
410 .name_index = EXT4_XATTR_INDEX_SYSTEM,
411 .name = EXT4_XATTR_SYSTEM_DATA,
417 if (!ei->i_inline_off)
420 error = ext4_get_inode_loc(inode, &is.iloc);
424 error = ext4_xattr_ibody_find(inode, &i, &is);
428 BUFFER_TRACE(is.iloc.bh, "get_write_access");
429 error = ext4_journal_get_write_access(handle, is.iloc.bh);
433 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is);
437 memset((void *)ext4_raw_inode(&is.iloc)->i_block,
438 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_CACHE_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 trace_android_fs_dataread_start(inode, page_offset(page), PAGE_SIZE,
507 current->pid, current->comm);
510 * Current inline data can only exist in the 1st page,
511 * So for all the other pages, just set them uptodate.
514 ret = ext4_read_inline_page(inode, page);
515 else if (!PageUptodate(page)) {
516 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
517 SetPageUptodate(page);
520 trace_android_fs_dataread_end(inode, page_offset(page), PAGE_SIZE);
522 up_read(&EXT4_I(inode)->xattr_sem);
525 return ret >= 0 ? 0 : ret;
528 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
532 int ret, needed_blocks;
533 handle_t *handle = NULL;
534 int retries = 0, sem_held = 0;
535 struct page *page = NULL;
537 struct ext4_iloc iloc;
539 if (!ext4_has_inline_data(inode)) {
541 * clear the flag so that no new write
542 * will trap here again.
544 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
548 needed_blocks = ext4_writepage_trans_blocks(inode);
550 ret = ext4_get_inode_loc(inode, &iloc);
555 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
556 if (IS_ERR(handle)) {
557 ret = PTR_ERR(handle);
562 /* We cannot recurse into the filesystem as the transaction is already
564 flags |= AOP_FLAG_NOFS;
566 page = grab_cache_page_write_begin(mapping, 0, flags);
572 down_write(&EXT4_I(inode)->xattr_sem);
574 /* If some one has already done this for us, just exit. */
575 if (!ext4_has_inline_data(inode)) {
581 to = ext4_get_inline_size(inode);
582 if (!PageUptodate(page)) {
583 ret = ext4_read_inline_page(inode, page);
588 ret = ext4_destroy_inline_data_nolock(handle, inode);
592 if (ext4_should_dioread_nolock(inode))
593 ret = __block_write_begin(page, from, to, ext4_get_block_write);
595 ret = __block_write_begin(page, from, to, ext4_get_block);
597 if (!ret && ext4_should_journal_data(inode)) {
598 ret = ext4_walk_page_buffers(handle, page_buffers(page),
600 do_journal_get_write_access);
605 page_cache_release(page);
607 ext4_orphan_add(handle, inode);
608 up_write(&EXT4_I(inode)->xattr_sem);
610 ext4_journal_stop(handle);
612 ext4_truncate_failed_write(inode);
614 * If truncate failed early the inode might
615 * still be on the orphan list; we need to
616 * make sure the inode is removed from the
617 * orphan list in that case.
620 ext4_orphan_del(NULL, inode);
623 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
627 block_commit_write(page, from, to);
631 page_cache_release(page);
634 up_write(&EXT4_I(inode)->xattr_sem);
636 ext4_journal_stop(handle);
642 * Try to write data in the inode.
643 * If the inode has inline data, check whether the new write can be
644 * in the inode also. If not, create the page the handle, move the data
645 * to the page make it update and let the later codes create extent for it.
647 int ext4_try_to_write_inline_data(struct address_space *mapping,
649 loff_t pos, unsigned len,
656 struct ext4_iloc iloc;
658 if (pos + len > ext4_get_max_inline_size(inode))
661 ret = ext4_get_inode_loc(inode, &iloc);
666 * The possible write could happen in the inode,
667 * so try to reserve the space in inode first.
669 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
670 if (IS_ERR(handle)) {
671 ret = PTR_ERR(handle);
676 ret = ext4_prepare_inline_data(handle, inode, pos + len);
677 if (ret && ret != -ENOSPC)
680 /* We don't have space in inline inode, so convert it to extent. */
681 if (ret == -ENOSPC) {
682 ext4_journal_stop(handle);
687 flags |= AOP_FLAG_NOFS;
689 page = grab_cache_page_write_begin(mapping, 0, flags);
696 down_read(&EXT4_I(inode)->xattr_sem);
697 if (!ext4_has_inline_data(inode)) {
700 page_cache_release(page);
704 if (!PageUptodate(page)) {
705 ret = ext4_read_inline_page(inode, page);
713 up_read(&EXT4_I(inode)->xattr_sem);
716 ext4_journal_stop(handle);
720 return ext4_convert_inline_data_to_extent(mapping,
724 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
725 unsigned copied, struct page *page)
729 struct ext4_iloc iloc;
731 if (unlikely(copied < len)) {
732 if (!PageUptodate(page)) {
738 ret = ext4_get_inode_loc(inode, &iloc);
740 ext4_std_error(inode->i_sb, ret);
745 down_write(&EXT4_I(inode)->xattr_sem);
746 BUG_ON(!ext4_has_inline_data(inode));
748 kaddr = kmap_atomic(page);
749 ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
750 kunmap_atomic(kaddr);
751 SetPageUptodate(page);
752 /* clear page dirty so that writepages wouldn't work for us. */
753 ClearPageDirty(page);
755 up_write(&EXT4_I(inode)->xattr_sem);
762 ext4_journalled_write_inline_data(struct inode *inode,
768 struct ext4_iloc iloc;
770 ret = ext4_get_inode_loc(inode, &iloc);
772 ext4_std_error(inode->i_sb, ret);
776 down_write(&EXT4_I(inode)->xattr_sem);
777 kaddr = kmap_atomic(page);
778 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
779 kunmap_atomic(kaddr);
780 up_write(&EXT4_I(inode)->xattr_sem);
786 * Try to make the page cache and handle ready for the inline data case.
787 * We can call this function in 2 cases:
788 * 1. The inode is created and the first write exceeds inline size. We can
789 * clear the inode state safely.
790 * 2. The inode has inline data, then we need to read the data, make it
791 * update and dirty so that ext4_da_writepages can handle it. We don't
792 * need to start the journal since the file's metatdata isn't changed now.
794 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
799 int ret = 0, inline_size;
802 page = grab_cache_page_write_begin(mapping, 0, flags);
806 down_read(&EXT4_I(inode)->xattr_sem);
807 if (!ext4_has_inline_data(inode)) {
808 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
812 inline_size = ext4_get_inline_size(inode);
814 if (!PageUptodate(page)) {
815 ret = ext4_read_inline_page(inode, page);
820 ret = __block_write_begin(page, 0, inline_size,
821 ext4_da_get_block_prep);
823 up_read(&EXT4_I(inode)->xattr_sem);
825 page_cache_release(page);
826 ext4_truncate_failed_write(inode);
831 SetPageUptodate(page);
832 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
833 *fsdata = (void *)CONVERT_INLINE_DATA;
836 up_read(&EXT4_I(inode)->xattr_sem);
839 page_cache_release(page);
845 * Prepare the write for the inline data.
846 * If the the data can be written into the inode, we just read
847 * the page and make it uptodate, and start the journal.
848 * Otherwise read the page, makes it dirty so that it can be
849 * handle in writepages(the i_disksize update is left to the
850 * normal ext4_da_write_end).
852 int ext4_da_write_inline_data_begin(struct address_space *mapping,
854 loff_t pos, unsigned len,
859 int ret, inline_size;
862 struct ext4_iloc iloc;
865 ret = ext4_get_inode_loc(inode, &iloc);
870 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
871 if (IS_ERR(handle)) {
872 ret = PTR_ERR(handle);
876 inline_size = ext4_get_max_inline_size(inode);
879 if (inline_size >= pos + len) {
880 ret = ext4_prepare_inline_data(handle, inode, pos + len);
881 if (ret && ret != -ENOSPC)
886 * We cannot recurse into the filesystem as the transaction
887 * is already started.
889 flags |= AOP_FLAG_NOFS;
891 if (ret == -ENOSPC) {
892 ret = ext4_da_convert_inline_data_to_extent(mapping,
896 ext4_journal_stop(handle);
897 if (ret == -ENOSPC &&
898 ext4_should_retry_alloc(inode->i_sb, &retries))
904 page = grab_cache_page_write_begin(mapping, 0, flags);
910 down_read(&EXT4_I(inode)->xattr_sem);
911 if (!ext4_has_inline_data(inode)) {
913 goto out_release_page;
916 if (!PageUptodate(page)) {
917 ret = ext4_read_inline_page(inode, page);
919 goto out_release_page;
922 up_read(&EXT4_I(inode)->xattr_sem);
927 up_read(&EXT4_I(inode)->xattr_sem);
929 page_cache_release(page);
931 ext4_journal_stop(handle);
937 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
938 unsigned len, unsigned copied,
941 int i_size_changed = 0;
944 ret = ext4_write_inline_data_end(inode, pos, len, copied, page);
953 * No need to use i_size_read() here, the i_size
954 * cannot change under us because we hold i_mutex.
956 * But it's important to update i_size while still holding page lock:
957 * page writeout could otherwise come in and zero beyond i_size.
959 if (pos+copied > inode->i_size) {
960 i_size_write(inode, pos+copied);
964 page_cache_release(page);
967 * Don't mark the inode dirty under page lock. First, it unnecessarily
968 * makes the holding time of page lock longer. Second, it forces lock
969 * ordering of page lock and transaction start for journaling
973 mark_inode_dirty(inode);
978 #ifdef INLINE_DIR_DEBUG
979 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
980 void *inline_start, int inline_size)
983 unsigned short de_len;
984 struct ext4_dir_entry_2 *de = inline_start;
985 void *dlimit = inline_start + inline_size;
987 trace_printk("inode %lu\n", dir->i_ino);
989 while ((void *)de < dlimit) {
990 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
991 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
992 offset, de_len, de->name_len, de->name,
993 de->name_len, le32_to_cpu(de->inode));
994 if (ext4_check_dir_entry(dir, NULL, de, bh,
995 inline_start, inline_size, offset))
999 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1003 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1007 * Add a new entry into a inline dir.
1008 * It will return -ENOSPC if no space is available, and -EIO
1009 * and -EEXIST if directory entry already exists.
1011 static int ext4_add_dirent_to_inline(handle_t *handle,
1012 struct ext4_filename *fname,
1013 struct dentry *dentry,
1014 struct inode *inode,
1015 struct ext4_iloc *iloc,
1016 void *inline_start, int inline_size)
1018 struct inode *dir = d_inode(dentry->d_parent);
1020 struct ext4_dir_entry_2 *de;
1022 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1023 inline_size, fname, &de);
1027 BUFFER_TRACE(iloc->bh, "get_write_access");
1028 err = ext4_journal_get_write_access(handle, iloc->bh);
1031 ext4_insert_dentry(dir, inode, de, inline_size, fname);
1033 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1036 * XXX shouldn't update any times until successful
1037 * completion of syscall, but too many callers depend
1040 * XXX similarly, too many callers depend on
1041 * ext4_new_inode() setting the times, but error
1042 * recovery deletes the inode, so the worst that can
1043 * happen is that the times are slightly out of date
1044 * and/or different from the directory change time.
1046 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1047 ext4_update_dx_flag(dir);
1049 ext4_mark_inode_dirty(handle, dir);
1053 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1054 struct ext4_iloc *iloc)
1056 struct ext4_xattr_entry *entry;
1057 struct ext4_xattr_ibody_header *header;
1059 BUG_ON(!EXT4_I(inode)->i_inline_off);
1061 header = IHDR(inode, ext4_raw_inode(iloc));
1062 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1063 EXT4_I(inode)->i_inline_off);
1065 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1068 /* Set the final de to cover the whole block. */
1069 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1071 struct ext4_dir_entry_2 *de, *prev_de;
1075 de = (struct ext4_dir_entry_2 *)de_buf;
1077 limit = de_buf + old_size;
1080 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1082 de = (struct ext4_dir_entry_2 *)de_buf;
1083 } while (de_buf < limit);
1085 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1086 old_size, new_size);
1088 /* this is just created, so create an empty entry. */
1090 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1094 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1095 struct ext4_iloc *iloc)
1098 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1099 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1101 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1104 ret = ext4_update_inline_data(handle, dir,
1105 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1109 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1110 EXT4_I(dir)->i_inline_size -
1111 EXT4_MIN_INLINE_DATA_SIZE);
1112 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1116 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1117 struct ext4_iloc *iloc,
1118 void *buf, int inline_size)
1120 ext4_create_inline_data(handle, inode, inline_size);
1121 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1122 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1125 static int ext4_finish_convert_inline_dir(handle_t *handle,
1126 struct inode *inode,
1127 struct buffer_head *dir_block,
1131 int err, csum_size = 0, header_size = 0;
1132 struct ext4_dir_entry_2 *de;
1133 struct ext4_dir_entry_tail *t;
1134 void *target = dir_block->b_data;
1137 * First create "." and ".." and then copy the dir information
1138 * back to the block.
1140 de = (struct ext4_dir_entry_2 *)target;
1141 de = ext4_init_dot_dotdot(inode, de,
1142 inode->i_sb->s_blocksize, csum_size,
1143 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1144 header_size = (void *)de - target;
1146 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1147 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1149 if (ext4_has_metadata_csum(inode->i_sb))
1150 csum_size = sizeof(struct ext4_dir_entry_tail);
1152 inode->i_size = inode->i_sb->s_blocksize;
1153 i_size_write(inode, inode->i_sb->s_blocksize);
1154 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1155 ext4_update_final_de(dir_block->b_data,
1156 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1157 inode->i_sb->s_blocksize - csum_size);
1160 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1161 inode->i_sb->s_blocksize);
1162 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1164 set_buffer_uptodate(dir_block);
1165 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1168 set_buffer_verified(dir_block);
1173 static int ext4_convert_inline_data_nolock(handle_t *handle,
1174 struct inode *inode,
1175 struct ext4_iloc *iloc)
1179 struct buffer_head *data_bh = NULL;
1180 struct ext4_map_blocks map;
1183 inline_size = ext4_get_inline_size(inode);
1184 buf = kmalloc(inline_size, GFP_NOFS);
1190 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1195 * Make sure the inline directory entries pass checks before we try to
1196 * convert them, so that we avoid touching stuff that needs fsck.
1198 if (S_ISDIR(inode->i_mode)) {
1199 error = ext4_check_all_de(inode, iloc->bh,
1200 buf + EXT4_INLINE_DOTDOT_SIZE,
1201 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1206 error = ext4_destroy_inline_data_nolock(handle, inode);
1213 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1216 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1221 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1227 lock_buffer(data_bh);
1228 error = ext4_journal_get_create_access(handle, data_bh);
1230 unlock_buffer(data_bh);
1234 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1236 if (!S_ISDIR(inode->i_mode)) {
1237 memcpy(data_bh->b_data, buf, inline_size);
1238 set_buffer_uptodate(data_bh);
1239 error = ext4_handle_dirty_metadata(handle,
1242 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1246 unlock_buffer(data_bh);
1249 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1258 * Try to add the new entry to the inline data.
1259 * If succeeds, return 0. If not, extended the inline dir and copied data to
1260 * the new created block.
1262 int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1263 struct dentry *dentry, struct inode *inode)
1265 int ret, inline_size;
1267 struct ext4_iloc iloc;
1268 struct inode *dir = d_inode(dentry->d_parent);
1270 ret = ext4_get_inode_loc(dir, &iloc);
1274 down_write(&EXT4_I(dir)->xattr_sem);
1275 if (!ext4_has_inline_data(dir))
1278 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1279 EXT4_INLINE_DOTDOT_SIZE;
1280 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1282 ret = ext4_add_dirent_to_inline(handle, fname, dentry, inode, &iloc,
1283 inline_start, inline_size);
1287 /* check whether it can be inserted to inline xattr space. */
1288 inline_size = EXT4_I(dir)->i_inline_size -
1289 EXT4_MIN_INLINE_DATA_SIZE;
1291 /* Try to use the xattr space.*/
1292 ret = ext4_update_inline_dir(handle, dir, &iloc);
1293 if (ret && ret != -ENOSPC)
1296 inline_size = EXT4_I(dir)->i_inline_size -
1297 EXT4_MIN_INLINE_DATA_SIZE;
1301 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1303 ret = ext4_add_dirent_to_inline(handle, fname, dentry,
1304 inode, &iloc, inline_start,
1312 * The inline space is filled up, so create a new block for it.
1313 * As the extent tree will be created, we have to save the inline
1316 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1319 ext4_mark_inode_dirty(handle, dir);
1320 up_write(&EXT4_I(dir)->xattr_sem);
1326 * This function fills a red-black tree with information from an
1327 * inlined dir. It returns the number directory entries loaded
1328 * into the tree. If there is an error it is returned in err.
1330 int htree_inlinedir_to_tree(struct file *dir_file,
1331 struct inode *dir, ext4_lblk_t block,
1332 struct dx_hash_info *hinfo,
1333 __u32 start_hash, __u32 start_minor_hash,
1334 int *has_inline_data)
1336 int err = 0, count = 0;
1337 unsigned int parent_ino;
1339 struct ext4_dir_entry_2 *de;
1340 struct inode *inode = file_inode(dir_file);
1341 int ret, inline_size = 0;
1342 struct ext4_iloc iloc;
1343 void *dir_buf = NULL;
1344 struct ext4_dir_entry_2 fake;
1345 struct ext4_str tmp_str;
1347 ret = ext4_get_inode_loc(inode, &iloc);
1351 down_read(&EXT4_I(inode)->xattr_sem);
1352 if (!ext4_has_inline_data(inode)) {
1353 up_read(&EXT4_I(inode)->xattr_sem);
1354 *has_inline_data = 0;
1358 inline_size = ext4_get_inline_size(inode);
1359 dir_buf = kmalloc(inline_size, GFP_NOFS);
1362 up_read(&EXT4_I(inode)->xattr_sem);
1366 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1367 up_read(&EXT4_I(inode)->xattr_sem);
1372 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1373 while (pos < inline_size) {
1375 * As inlined dir doesn't store any information about '.' and
1376 * only the inode number of '..' is stored, we have to handle
1380 fake.inode = cpu_to_le32(inode->i_ino);
1382 strcpy(fake.name, ".");
1383 fake.rec_len = ext4_rec_len_to_disk(
1384 EXT4_DIR_REC_LEN(fake.name_len),
1386 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1388 pos = EXT4_INLINE_DOTDOT_OFFSET;
1389 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1390 fake.inode = cpu_to_le32(parent_ino);
1392 strcpy(fake.name, "..");
1393 fake.rec_len = ext4_rec_len_to_disk(
1394 EXT4_DIR_REC_LEN(fake.name_len),
1396 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1398 pos = EXT4_INLINE_DOTDOT_SIZE;
1400 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1401 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1402 if (ext4_check_dir_entry(inode, dir_file, de,
1404 inline_size, pos)) {
1410 ext4fs_dirhash(de->name, de->name_len, hinfo);
1411 if ((hinfo->hash < start_hash) ||
1412 ((hinfo->hash == start_hash) &&
1413 (hinfo->minor_hash < start_minor_hash)))
1417 tmp_str.name = de->name;
1418 tmp_str.len = de->name_len;
1419 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1420 hinfo->minor_hash, de, &tmp_str);
1435 * So this function is called when the volume is mkfsed with
1436 * dir_index disabled. In order to keep f_pos persistent
1437 * after we convert from an inlined dir to a blocked based,
1438 * we just pretend that we are a normal dir and return the
1439 * offset as if '.' and '..' really take place.
1442 int ext4_read_inline_dir(struct file *file,
1443 struct dir_context *ctx,
1444 int *has_inline_data)
1446 unsigned int offset, parent_ino;
1448 struct ext4_dir_entry_2 *de;
1449 struct super_block *sb;
1450 struct inode *inode = file_inode(file);
1451 int ret, inline_size = 0;
1452 struct ext4_iloc iloc;
1453 void *dir_buf = NULL;
1454 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1456 ret = ext4_get_inode_loc(inode, &iloc);
1460 down_read(&EXT4_I(inode)->xattr_sem);
1461 if (!ext4_has_inline_data(inode)) {
1462 up_read(&EXT4_I(inode)->xattr_sem);
1463 *has_inline_data = 0;
1467 inline_size = ext4_get_inline_size(inode);
1468 dir_buf = kmalloc(inline_size, GFP_NOFS);
1471 up_read(&EXT4_I(inode)->xattr_sem);
1475 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1476 up_read(&EXT4_I(inode)->xattr_sem);
1482 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1486 * dotdot_offset and dotdot_size is the real offset and
1487 * size for ".." and "." if the dir is block based while
1488 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1489 * So we will use extra_offset and extra_size to indicate them
1490 * during the inline dir iteration.
1492 dotdot_offset = EXT4_DIR_REC_LEN(1);
1493 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1494 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1495 extra_size = extra_offset + inline_size;
1498 * If the version has changed since the last call to
1499 * readdir(2), then we might be pointing to an invalid
1500 * dirent right now. Scan from the start of the inline
1503 if (file->f_version != inode->i_version) {
1504 for (i = 0; i < extra_size && i < offset;) {
1506 * "." is with offset 0 and
1507 * ".." is dotdot_offset.
1512 } else if (i == dotdot_offset) {
1516 /* for other entry, the real offset in
1517 * the buf has to be tuned accordingly.
1519 de = (struct ext4_dir_entry_2 *)
1520 (dir_buf + i - extra_offset);
1521 /* It's too expensive to do a full
1522 * dirent test each time round this
1523 * loop, but we do have to test at
1524 * least that it is non-zero. A
1525 * failure will be detected in the
1526 * dirent test below. */
1527 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1528 < EXT4_DIR_REC_LEN(1))
1530 i += ext4_rec_len_from_disk(de->rec_len,
1535 file->f_version = inode->i_version;
1538 while (ctx->pos < extra_size) {
1539 if (ctx->pos == 0) {
1540 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1542 ctx->pos = dotdot_offset;
1546 if (ctx->pos == dotdot_offset) {
1547 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1549 ctx->pos = dotdot_size;
1553 de = (struct ext4_dir_entry_2 *)
1554 (dir_buf + ctx->pos - extra_offset);
1555 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1556 extra_size, ctx->pos))
1558 if (le32_to_cpu(de->inode)) {
1559 if (!dir_emit(ctx, de->name, de->name_len,
1560 le32_to_cpu(de->inode),
1561 get_dtype(sb, de->file_type)))
1564 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1572 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1573 struct ext4_dir_entry_2 **parent_de,
1576 struct ext4_iloc iloc;
1578 *retval = ext4_get_inode_loc(inode, &iloc);
1582 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1588 * Try to create the inline data for the new dir.
1589 * If it succeeds, return 0, otherwise return the error.
1590 * In case of ENOSPC, the caller should create the normal disk layout dir.
1592 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1593 struct inode *inode)
1595 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1596 struct ext4_iloc iloc;
1597 struct ext4_dir_entry_2 *de;
1599 ret = ext4_get_inode_loc(inode, &iloc);
1603 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1608 * For inline dir, we only save the inode information for the ".."
1609 * and create a fake dentry to cover the left space.
1611 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1612 de->inode = cpu_to_le32(parent->i_ino);
1613 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1615 de->rec_len = ext4_rec_len_to_disk(
1616 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1618 set_nlink(inode, 2);
1619 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1625 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1626 struct ext4_filename *fname,
1627 const struct qstr *d_name,
1628 struct ext4_dir_entry_2 **res_dir,
1629 int *has_inline_data)
1632 struct ext4_iloc iloc;
1636 if (ext4_get_inode_loc(dir, &iloc))
1639 down_read(&EXT4_I(dir)->xattr_sem);
1640 if (!ext4_has_inline_data(dir)) {
1641 *has_inline_data = 0;
1645 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1646 EXT4_INLINE_DOTDOT_SIZE;
1647 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1648 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1649 dir, fname, d_name, 0, res_dir);
1655 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1658 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1659 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1661 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1662 dir, fname, d_name, 0, res_dir);
1670 up_read(&EXT4_I(dir)->xattr_sem);
1674 int ext4_delete_inline_entry(handle_t *handle,
1676 struct ext4_dir_entry_2 *de_del,
1677 struct buffer_head *bh,
1678 int *has_inline_data)
1680 int err, inline_size;
1681 struct ext4_iloc iloc;
1684 err = ext4_get_inode_loc(dir, &iloc);
1688 down_write(&EXT4_I(dir)->xattr_sem);
1689 if (!ext4_has_inline_data(dir)) {
1690 *has_inline_data = 0;
1694 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1695 EXT4_MIN_INLINE_DATA_SIZE) {
1696 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1697 EXT4_INLINE_DOTDOT_SIZE;
1698 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1699 EXT4_INLINE_DOTDOT_SIZE;
1701 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1702 inline_size = ext4_get_inline_size(dir) -
1703 EXT4_MIN_INLINE_DATA_SIZE;
1706 BUFFER_TRACE(bh, "get_write_access");
1707 err = ext4_journal_get_write_access(handle, bh);
1711 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1712 inline_start, inline_size, 0);
1716 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1717 err = ext4_mark_inode_dirty(handle, dir);
1721 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1723 up_write(&EXT4_I(dir)->xattr_sem);
1726 ext4_std_error(dir->i_sb, err);
1731 * Get the inline dentry at offset.
1733 static inline struct ext4_dir_entry_2 *
1734 ext4_get_inline_entry(struct inode *inode,
1735 struct ext4_iloc *iloc,
1736 unsigned int offset,
1737 void **inline_start,
1742 BUG_ON(offset > ext4_get_inline_size(inode));
1744 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1745 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1746 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1748 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1749 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1750 *inline_size = ext4_get_inline_size(inode) -
1751 EXT4_MIN_INLINE_DATA_SIZE;
1755 *inline_start = inline_pos;
1756 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1759 int empty_inline_dir(struct inode *dir, int *has_inline_data)
1761 int err, inline_size;
1762 struct ext4_iloc iloc;
1764 unsigned int offset;
1765 struct ext4_dir_entry_2 *de;
1768 err = ext4_get_inode_loc(dir, &iloc);
1770 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1775 down_read(&EXT4_I(dir)->xattr_sem);
1776 if (!ext4_has_inline_data(dir)) {
1777 *has_inline_data = 0;
1781 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1782 if (!le32_to_cpu(de->inode)) {
1783 ext4_warning(dir->i_sb,
1784 "bad inline directory (dir #%lu) - no `..'",
1790 offset = EXT4_INLINE_DOTDOT_SIZE;
1791 while (offset < dir->i_size) {
1792 de = ext4_get_inline_entry(dir, &iloc, offset,
1793 &inline_pos, &inline_size);
1794 if (ext4_check_dir_entry(dir, NULL, de,
1795 iloc.bh, inline_pos,
1796 inline_size, offset)) {
1797 ext4_warning(dir->i_sb,
1798 "bad inline directory (dir #%lu) - "
1799 "inode %u, rec_len %u, name_len %d"
1801 dir->i_ino, le32_to_cpu(de->inode),
1802 le16_to_cpu(de->rec_len), de->name_len,
1807 if (le32_to_cpu(de->inode)) {
1811 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1815 up_read(&EXT4_I(dir)->xattr_sem);
1820 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1824 down_write(&EXT4_I(inode)->xattr_sem);
1825 ret = ext4_destroy_inline_data_nolock(handle, inode);
1826 up_write(&EXT4_I(inode)->xattr_sem);
1831 int ext4_inline_data_fiemap(struct inode *inode,
1832 struct fiemap_extent_info *fieinfo,
1833 int *has_inline, __u64 start, __u64 len)
1837 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1840 struct ext4_iloc iloc;
1842 down_read(&EXT4_I(inode)->xattr_sem);
1843 if (!ext4_has_inline_data(inode)) {
1847 inline_len = min_t(size_t, ext4_get_inline_size(inode),
1848 i_size_read(inode));
1849 if (start >= inline_len)
1851 if (start + len < inline_len)
1852 inline_len = start + len;
1853 inline_len -= start;
1855 error = ext4_get_inode_loc(inode, &iloc);
1859 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1860 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1861 physical += offsetof(struct ext4_inode, i_block);
1864 error = fiemap_fill_next_extent(fieinfo, start, physical,
1868 up_read(&EXT4_I(inode)->xattr_sem);
1869 return (error < 0 ? error : 0);
1873 * Called during xattr set, and if we can sparse space 'needed',
1874 * just create the extent tree evict the data to the outer block.
1876 * We use jbd2 instead of page cache to move data to the 1st block
1877 * so that the whole transaction can be committed as a whole and
1878 * the data isn't lost because of the delayed page cache write.
1880 int ext4_try_to_evict_inline_data(handle_t *handle,
1881 struct inode *inode,
1885 struct ext4_xattr_entry *entry;
1886 struct ext4_inode *raw_inode;
1887 struct ext4_iloc iloc;
1889 error = ext4_get_inode_loc(inode, &iloc);
1893 raw_inode = ext4_raw_inode(&iloc);
1894 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1895 EXT4_I(inode)->i_inline_off);
1896 if (EXT4_XATTR_LEN(entry->e_name_len) +
1897 EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1902 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1908 void ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1911 int inline_size, value_len, needed_blocks;
1914 struct ext4_xattr_ibody_find is = {
1915 .s = { .not_found = -ENODATA, },
1917 struct ext4_xattr_info i = {
1918 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1919 .name = EXT4_XATTR_SYSTEM_DATA,
1923 needed_blocks = ext4_writepage_trans_blocks(inode);
1924 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1928 down_write(&EXT4_I(inode)->xattr_sem);
1929 if (!ext4_has_inline_data(inode)) {
1931 ext4_journal_stop(handle);
1935 if (ext4_orphan_add(handle, inode))
1938 if (ext4_get_inode_loc(inode, &is.iloc))
1941 down_write(&EXT4_I(inode)->i_data_sem);
1942 i_size = inode->i_size;
1943 inline_size = ext4_get_inline_size(inode);
1944 EXT4_I(inode)->i_disksize = i_size;
1946 if (i_size < inline_size) {
1947 /* Clear the content in the xattr space. */
1948 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1949 if (ext4_xattr_ibody_find(inode, &i, &is))
1952 BUG_ON(is.s.not_found);
1954 value_len = le32_to_cpu(is.s.here->e_value_size);
1955 value = kmalloc(value_len, GFP_NOFS);
1959 if (ext4_xattr_ibody_get(inode, i.name_index, i.name,
1964 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1965 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1966 if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is))
1970 /* Clear the content within i_blocks. */
1971 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1972 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1973 memset(p + i_size, 0,
1974 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1977 EXT4_I(inode)->i_inline_size = i_size <
1978 EXT4_MIN_INLINE_DATA_SIZE ?
1979 EXT4_MIN_INLINE_DATA_SIZE : i_size;
1983 up_write(&EXT4_I(inode)->i_data_sem);
1986 up_write(&EXT4_I(inode)->xattr_sem);
1989 ext4_orphan_del(handle, inode);
1991 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1992 ext4_mark_inode_dirty(handle, inode);
1994 ext4_handle_sync(handle);
1996 ext4_journal_stop(handle);
2000 int ext4_convert_inline_data(struct inode *inode)
2002 int error, needed_blocks;
2004 struct ext4_iloc iloc;
2006 if (!ext4_has_inline_data(inode)) {
2007 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
2011 needed_blocks = ext4_writepage_trans_blocks(inode);
2014 error = ext4_get_inode_loc(inode, &iloc);
2018 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2019 if (IS_ERR(handle)) {
2020 error = PTR_ERR(handle);
2024 down_write(&EXT4_I(inode)->xattr_sem);
2025 if (!ext4_has_inline_data(inode)) {
2026 up_write(&EXT4_I(inode)->xattr_sem);
2030 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2031 up_write(&EXT4_I(inode)->xattr_sem);
2033 ext4_journal_stop(handle);