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,
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 down_write(&EXT4_I(inode)->xattr_sem);
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 up_write(&EXT4_I(inode)->xattr_sem);
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);
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;
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 down_write(&EXT4_I(inode)->xattr_sem);
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 up_write(&EXT4_I(inode)->xattr_sem);
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 up_write(&EXT4_I(inode)->xattr_sem);
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 flags |= AOP_FLAG_NOFS;
683 page = grab_cache_page_write_begin(mapping, 0, flags);
690 down_read(&EXT4_I(inode)->xattr_sem);
691 if (!ext4_has_inline_data(inode)) {
694 page_cache_release(page);
698 if (!PageUptodate(page)) {
699 ret = ext4_read_inline_page(inode, page);
707 up_read(&EXT4_I(inode)->xattr_sem);
710 ext4_journal_stop(handle);
714 return ext4_convert_inline_data_to_extent(mapping,
718 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
719 unsigned copied, struct page *page)
723 struct ext4_iloc iloc;
725 if (unlikely(copied < len)) {
726 if (!PageUptodate(page)) {
732 ret = ext4_get_inode_loc(inode, &iloc);
734 ext4_std_error(inode->i_sb, ret);
739 down_write(&EXT4_I(inode)->xattr_sem);
740 BUG_ON(!ext4_has_inline_data(inode));
742 kaddr = kmap_atomic(page);
743 ext4_write_inline_data(inode, &iloc, kaddr, pos, len);
744 kunmap_atomic(kaddr);
745 SetPageUptodate(page);
746 /* clear page dirty so that writepages wouldn't work for us. */
747 ClearPageDirty(page);
749 up_write(&EXT4_I(inode)->xattr_sem);
756 ext4_journalled_write_inline_data(struct inode *inode,
762 struct ext4_iloc iloc;
764 ret = ext4_get_inode_loc(inode, &iloc);
766 ext4_std_error(inode->i_sb, ret);
770 down_write(&EXT4_I(inode)->xattr_sem);
771 kaddr = kmap_atomic(page);
772 ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
773 kunmap_atomic(kaddr);
774 up_write(&EXT4_I(inode)->xattr_sem);
780 * Try to make the page cache and handle ready for the inline data case.
781 * We can call this function in 2 cases:
782 * 1. The inode is created and the first write exceeds inline size. We can
783 * clear the inode state safely.
784 * 2. The inode has inline data, then we need to read the data, make it
785 * update and dirty so that ext4_da_writepages can handle it. We don't
786 * need to start the journal since the file's metatdata isn't changed now.
788 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
793 int ret = 0, inline_size;
796 page = grab_cache_page_write_begin(mapping, 0, flags);
800 down_read(&EXT4_I(inode)->xattr_sem);
801 if (!ext4_has_inline_data(inode)) {
802 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
806 inline_size = ext4_get_inline_size(inode);
808 if (!PageUptodate(page)) {
809 ret = ext4_read_inline_page(inode, page);
814 ret = __block_write_begin(page, 0, inline_size,
815 ext4_da_get_block_prep);
817 up_read(&EXT4_I(inode)->xattr_sem);
819 page_cache_release(page);
820 ext4_truncate_failed_write(inode);
825 SetPageUptodate(page);
826 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
827 *fsdata = (void *)CONVERT_INLINE_DATA;
830 up_read(&EXT4_I(inode)->xattr_sem);
833 page_cache_release(page);
839 * Prepare the write for the inline data.
840 * If the the data can be written into the inode, we just read
841 * the page and make it uptodate, and start the journal.
842 * Otherwise read the page, makes it dirty so that it can be
843 * handle in writepages(the i_disksize update is left to the
844 * normal ext4_da_write_end).
846 int ext4_da_write_inline_data_begin(struct address_space *mapping,
848 loff_t pos, unsigned len,
853 int ret, inline_size;
856 struct ext4_iloc iloc;
859 ret = ext4_get_inode_loc(inode, &iloc);
864 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
865 if (IS_ERR(handle)) {
866 ret = PTR_ERR(handle);
870 inline_size = ext4_get_max_inline_size(inode);
873 if (inline_size >= pos + len) {
874 ret = ext4_prepare_inline_data(handle, inode, pos + len);
875 if (ret && ret != -ENOSPC)
880 * We cannot recurse into the filesystem as the transaction
881 * is already started.
883 flags |= AOP_FLAG_NOFS;
885 if (ret == -ENOSPC) {
886 ret = ext4_da_convert_inline_data_to_extent(mapping,
890 ext4_journal_stop(handle);
891 if (ret == -ENOSPC &&
892 ext4_should_retry_alloc(inode->i_sb, &retries))
898 page = grab_cache_page_write_begin(mapping, 0, flags);
904 down_read(&EXT4_I(inode)->xattr_sem);
905 if (!ext4_has_inline_data(inode)) {
907 goto out_release_page;
910 if (!PageUptodate(page)) {
911 ret = ext4_read_inline_page(inode, page);
913 goto out_release_page;
916 up_read(&EXT4_I(inode)->xattr_sem);
921 up_read(&EXT4_I(inode)->xattr_sem);
923 page_cache_release(page);
925 ext4_journal_stop(handle);
931 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos,
932 unsigned len, unsigned copied,
935 int i_size_changed = 0;
937 copied = ext4_write_inline_data_end(inode, pos, len, copied, page);
940 * No need to use i_size_read() here, the i_size
941 * cannot change under us because we hold i_mutex.
943 * But it's important to update i_size while still holding page lock:
944 * page writeout could otherwise come in and zero beyond i_size.
946 if (pos+copied > inode->i_size) {
947 i_size_write(inode, pos+copied);
951 page_cache_release(page);
954 * Don't mark the inode dirty under page lock. First, it unnecessarily
955 * makes the holding time of page lock longer. Second, it forces lock
956 * ordering of page lock and transaction start for journaling
960 mark_inode_dirty(inode);
965 #ifdef INLINE_DIR_DEBUG
966 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
967 void *inline_start, int inline_size)
970 unsigned short de_len;
971 struct ext4_dir_entry_2 *de = inline_start;
972 void *dlimit = inline_start + inline_size;
974 trace_printk("inode %lu\n", dir->i_ino);
976 while ((void *)de < dlimit) {
977 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
978 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
979 offset, de_len, de->name_len, de->name,
980 de->name_len, le32_to_cpu(de->inode));
981 if (ext4_check_dir_entry(dir, NULL, de, bh,
982 inline_start, inline_size, offset))
986 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
990 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
994 * Add a new entry into a inline dir.
995 * It will return -ENOSPC if no space is available, and -EIO
996 * and -EEXIST if directory entry already exists.
998 static int ext4_add_dirent_to_inline(handle_t *handle,
999 struct ext4_filename *fname,
1000 struct dentry *dentry,
1001 struct inode *inode,
1002 struct ext4_iloc *iloc,
1003 void *inline_start, int inline_size)
1005 struct inode *dir = d_inode(dentry->d_parent);
1007 struct ext4_dir_entry_2 *de;
1009 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1010 inline_size, fname, &de);
1014 BUFFER_TRACE(iloc->bh, "get_write_access");
1015 err = ext4_journal_get_write_access(handle, iloc->bh);
1018 ext4_insert_dentry(dir, inode, de, inline_size, fname);
1020 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1023 * XXX shouldn't update any times until successful
1024 * completion of syscall, but too many callers depend
1027 * XXX similarly, too many callers depend on
1028 * ext4_new_inode() setting the times, but error
1029 * recovery deletes the inode, so the worst that can
1030 * happen is that the times are slightly out of date
1031 * and/or different from the directory change time.
1033 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1034 ext4_update_dx_flag(dir);
1036 ext4_mark_inode_dirty(handle, dir);
1040 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1041 struct ext4_iloc *iloc)
1043 struct ext4_xattr_entry *entry;
1044 struct ext4_xattr_ibody_header *header;
1046 BUG_ON(!EXT4_I(inode)->i_inline_off);
1048 header = IHDR(inode, ext4_raw_inode(iloc));
1049 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1050 EXT4_I(inode)->i_inline_off);
1052 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1055 /* Set the final de to cover the whole block. */
1056 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1058 struct ext4_dir_entry_2 *de, *prev_de;
1062 de = (struct ext4_dir_entry_2 *)de_buf;
1064 limit = de_buf + old_size;
1067 de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1069 de = (struct ext4_dir_entry_2 *)de_buf;
1070 } while (de_buf < limit);
1072 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1073 old_size, new_size);
1075 /* this is just created, so create an empty entry. */
1077 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1081 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1082 struct ext4_iloc *iloc)
1085 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1086 int new_size = get_max_inline_xattr_value_size(dir, iloc);
1088 if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
1091 ret = ext4_update_inline_data(handle, dir,
1092 new_size + EXT4_MIN_INLINE_DATA_SIZE);
1096 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1097 EXT4_I(dir)->i_inline_size -
1098 EXT4_MIN_INLINE_DATA_SIZE);
1099 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1103 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1104 struct ext4_iloc *iloc,
1105 void *buf, int inline_size)
1107 ext4_create_inline_data(handle, inode, inline_size);
1108 ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1109 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1112 static int ext4_finish_convert_inline_dir(handle_t *handle,
1113 struct inode *inode,
1114 struct buffer_head *dir_block,
1118 int err, csum_size = 0, header_size = 0;
1119 struct ext4_dir_entry_2 *de;
1120 struct ext4_dir_entry_tail *t;
1121 void *target = dir_block->b_data;
1124 * First create "." and ".." and then copy the dir information
1125 * back to the block.
1127 de = (struct ext4_dir_entry_2 *)target;
1128 de = ext4_init_dot_dotdot(inode, de,
1129 inode->i_sb->s_blocksize, csum_size,
1130 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1131 header_size = (void *)de - target;
1133 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1134 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1136 if (ext4_has_metadata_csum(inode->i_sb))
1137 csum_size = sizeof(struct ext4_dir_entry_tail);
1139 inode->i_size = inode->i_sb->s_blocksize;
1140 i_size_write(inode, inode->i_sb->s_blocksize);
1141 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1142 ext4_update_final_de(dir_block->b_data,
1143 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1144 inode->i_sb->s_blocksize - csum_size);
1147 t = EXT4_DIRENT_TAIL(dir_block->b_data,
1148 inode->i_sb->s_blocksize);
1149 initialize_dirent_tail(t, inode->i_sb->s_blocksize);
1151 set_buffer_uptodate(dir_block);
1152 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
1155 set_buffer_verified(dir_block);
1160 static int ext4_convert_inline_data_nolock(handle_t *handle,
1161 struct inode *inode,
1162 struct ext4_iloc *iloc)
1166 struct buffer_head *data_bh = NULL;
1167 struct ext4_map_blocks map;
1170 inline_size = ext4_get_inline_size(inode);
1171 buf = kmalloc(inline_size, GFP_NOFS);
1177 error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1182 * Make sure the inline directory entries pass checks before we try to
1183 * convert them, so that we avoid touching stuff that needs fsck.
1185 if (S_ISDIR(inode->i_mode)) {
1186 error = ext4_check_all_de(inode, iloc->bh,
1187 buf + EXT4_INLINE_DOTDOT_SIZE,
1188 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1193 error = ext4_destroy_inline_data_nolock(handle, inode);
1200 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1203 if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1208 data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1214 lock_buffer(data_bh);
1215 error = ext4_journal_get_create_access(handle, data_bh);
1217 unlock_buffer(data_bh);
1221 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1223 if (!S_ISDIR(inode->i_mode)) {
1224 memcpy(data_bh->b_data, buf, inline_size);
1225 set_buffer_uptodate(data_bh);
1226 error = ext4_handle_dirty_metadata(handle,
1229 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1233 unlock_buffer(data_bh);
1236 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1245 * Try to add the new entry to the inline data.
1246 * If succeeds, return 0. If not, extended the inline dir and copied data to
1247 * the new created block.
1249 int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1250 struct dentry *dentry, struct inode *inode)
1252 int ret, inline_size;
1254 struct ext4_iloc iloc;
1255 struct inode *dir = d_inode(dentry->d_parent);
1257 ret = ext4_get_inode_loc(dir, &iloc);
1261 down_write(&EXT4_I(dir)->xattr_sem);
1262 if (!ext4_has_inline_data(dir))
1265 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1266 EXT4_INLINE_DOTDOT_SIZE;
1267 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1269 ret = ext4_add_dirent_to_inline(handle, fname, dentry, inode, &iloc,
1270 inline_start, inline_size);
1274 /* check whether it can be inserted to inline xattr space. */
1275 inline_size = EXT4_I(dir)->i_inline_size -
1276 EXT4_MIN_INLINE_DATA_SIZE;
1278 /* Try to use the xattr space.*/
1279 ret = ext4_update_inline_dir(handle, dir, &iloc);
1280 if (ret && ret != -ENOSPC)
1283 inline_size = EXT4_I(dir)->i_inline_size -
1284 EXT4_MIN_INLINE_DATA_SIZE;
1288 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1290 ret = ext4_add_dirent_to_inline(handle, fname, dentry,
1291 inode, &iloc, inline_start,
1299 * The inline space is filled up, so create a new block for it.
1300 * As the extent tree will be created, we have to save the inline
1303 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1306 ext4_mark_inode_dirty(handle, dir);
1307 up_write(&EXT4_I(dir)->xattr_sem);
1313 * This function fills a red-black tree with information from an
1314 * inlined dir. It returns the number directory entries loaded
1315 * into the tree. If there is an error it is returned in err.
1317 int htree_inlinedir_to_tree(struct file *dir_file,
1318 struct inode *dir, ext4_lblk_t block,
1319 struct dx_hash_info *hinfo,
1320 __u32 start_hash, __u32 start_minor_hash,
1321 int *has_inline_data)
1323 int err = 0, count = 0;
1324 unsigned int parent_ino;
1326 struct ext4_dir_entry_2 *de;
1327 struct inode *inode = file_inode(dir_file);
1328 int ret, inline_size = 0;
1329 struct ext4_iloc iloc;
1330 void *dir_buf = NULL;
1331 struct ext4_dir_entry_2 fake;
1332 struct ext4_str tmp_str;
1334 ret = ext4_get_inode_loc(inode, &iloc);
1338 down_read(&EXT4_I(inode)->xattr_sem);
1339 if (!ext4_has_inline_data(inode)) {
1340 up_read(&EXT4_I(inode)->xattr_sem);
1341 *has_inline_data = 0;
1345 inline_size = ext4_get_inline_size(inode);
1346 dir_buf = kmalloc(inline_size, GFP_NOFS);
1349 up_read(&EXT4_I(inode)->xattr_sem);
1353 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1354 up_read(&EXT4_I(inode)->xattr_sem);
1359 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1360 while (pos < inline_size) {
1362 * As inlined dir doesn't store any information about '.' and
1363 * only the inode number of '..' is stored, we have to handle
1367 fake.inode = cpu_to_le32(inode->i_ino);
1369 strcpy(fake.name, ".");
1370 fake.rec_len = ext4_rec_len_to_disk(
1371 EXT4_DIR_REC_LEN(fake.name_len),
1373 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1375 pos = EXT4_INLINE_DOTDOT_OFFSET;
1376 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1377 fake.inode = cpu_to_le32(parent_ino);
1379 strcpy(fake.name, "..");
1380 fake.rec_len = ext4_rec_len_to_disk(
1381 EXT4_DIR_REC_LEN(fake.name_len),
1383 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1385 pos = EXT4_INLINE_DOTDOT_SIZE;
1387 de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1388 pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1389 if (ext4_check_dir_entry(inode, dir_file, de,
1391 inline_size, pos)) {
1397 ext4fs_dirhash(de->name, de->name_len, hinfo);
1398 if ((hinfo->hash < start_hash) ||
1399 ((hinfo->hash == start_hash) &&
1400 (hinfo->minor_hash < start_minor_hash)))
1404 tmp_str.name = de->name;
1405 tmp_str.len = de->name_len;
1406 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1407 hinfo->minor_hash, de, &tmp_str);
1422 * So this function is called when the volume is mkfsed with
1423 * dir_index disabled. In order to keep f_pos persistent
1424 * after we convert from an inlined dir to a blocked based,
1425 * we just pretend that we are a normal dir and return the
1426 * offset as if '.' and '..' really take place.
1429 int ext4_read_inline_dir(struct file *file,
1430 struct dir_context *ctx,
1431 int *has_inline_data)
1433 unsigned int offset, parent_ino;
1435 struct ext4_dir_entry_2 *de;
1436 struct super_block *sb;
1437 struct inode *inode = file_inode(file);
1438 int ret, inline_size = 0;
1439 struct ext4_iloc iloc;
1440 void *dir_buf = NULL;
1441 int dotdot_offset, dotdot_size, extra_offset, extra_size;
1443 ret = ext4_get_inode_loc(inode, &iloc);
1447 down_read(&EXT4_I(inode)->xattr_sem);
1448 if (!ext4_has_inline_data(inode)) {
1449 up_read(&EXT4_I(inode)->xattr_sem);
1450 *has_inline_data = 0;
1454 inline_size = ext4_get_inline_size(inode);
1455 dir_buf = kmalloc(inline_size, GFP_NOFS);
1458 up_read(&EXT4_I(inode)->xattr_sem);
1462 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1463 up_read(&EXT4_I(inode)->xattr_sem);
1469 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1473 * dotdot_offset and dotdot_size is the real offset and
1474 * size for ".." and "." if the dir is block based while
1475 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1476 * So we will use extra_offset and extra_size to indicate them
1477 * during the inline dir iteration.
1479 dotdot_offset = EXT4_DIR_REC_LEN(1);
1480 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
1481 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1482 extra_size = extra_offset + inline_size;
1485 * If the version has changed since the last call to
1486 * readdir(2), then we might be pointing to an invalid
1487 * dirent right now. Scan from the start of the inline
1490 if (file->f_version != inode->i_version) {
1491 for (i = 0; i < extra_size && i < offset;) {
1493 * "." is with offset 0 and
1494 * ".." is dotdot_offset.
1499 } else if (i == dotdot_offset) {
1503 /* for other entry, the real offset in
1504 * the buf has to be tuned accordingly.
1506 de = (struct ext4_dir_entry_2 *)
1507 (dir_buf + i - extra_offset);
1508 /* It's too expensive to do a full
1509 * dirent test each time round this
1510 * loop, but we do have to test at
1511 * least that it is non-zero. A
1512 * failure will be detected in the
1513 * dirent test below. */
1514 if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1515 < EXT4_DIR_REC_LEN(1))
1517 i += ext4_rec_len_from_disk(de->rec_len,
1522 file->f_version = inode->i_version;
1525 while (ctx->pos < extra_size) {
1526 if (ctx->pos == 0) {
1527 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1529 ctx->pos = dotdot_offset;
1533 if (ctx->pos == dotdot_offset) {
1534 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1536 ctx->pos = dotdot_size;
1540 de = (struct ext4_dir_entry_2 *)
1541 (dir_buf + ctx->pos - extra_offset);
1542 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1543 extra_size, ctx->pos))
1545 if (le32_to_cpu(de->inode)) {
1546 if (!dir_emit(ctx, de->name, de->name_len,
1547 le32_to_cpu(de->inode),
1548 get_dtype(sb, de->file_type)))
1551 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1559 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1560 struct ext4_dir_entry_2 **parent_de,
1563 struct ext4_iloc iloc;
1565 *retval = ext4_get_inode_loc(inode, &iloc);
1569 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1575 * Try to create the inline data for the new dir.
1576 * If it succeeds, return 0, otherwise return the error.
1577 * In case of ENOSPC, the caller should create the normal disk layout dir.
1579 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1580 struct inode *inode)
1582 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1583 struct ext4_iloc iloc;
1584 struct ext4_dir_entry_2 *de;
1586 ret = ext4_get_inode_loc(inode, &iloc);
1590 ret = ext4_prepare_inline_data(handle, inode, inline_size);
1595 * For inline dir, we only save the inode information for the ".."
1596 * and create a fake dentry to cover the left space.
1598 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1599 de->inode = cpu_to_le32(parent->i_ino);
1600 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1602 de->rec_len = ext4_rec_len_to_disk(
1603 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1605 set_nlink(inode, 2);
1606 inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1612 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1613 struct ext4_filename *fname,
1614 const struct qstr *d_name,
1615 struct ext4_dir_entry_2 **res_dir,
1616 int *has_inline_data)
1619 struct ext4_iloc iloc;
1623 if (ext4_get_inode_loc(dir, &iloc))
1626 down_read(&EXT4_I(dir)->xattr_sem);
1627 if (!ext4_has_inline_data(dir)) {
1628 *has_inline_data = 0;
1632 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1633 EXT4_INLINE_DOTDOT_SIZE;
1634 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1635 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1636 dir, fname, d_name, 0, res_dir);
1642 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1645 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1646 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1648 ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1649 dir, fname, d_name, 0, res_dir);
1657 up_read(&EXT4_I(dir)->xattr_sem);
1661 int ext4_delete_inline_entry(handle_t *handle,
1663 struct ext4_dir_entry_2 *de_del,
1664 struct buffer_head *bh,
1665 int *has_inline_data)
1667 int err, inline_size;
1668 struct ext4_iloc iloc;
1671 err = ext4_get_inode_loc(dir, &iloc);
1675 down_write(&EXT4_I(dir)->xattr_sem);
1676 if (!ext4_has_inline_data(dir)) {
1677 *has_inline_data = 0;
1681 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1682 EXT4_MIN_INLINE_DATA_SIZE) {
1683 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1684 EXT4_INLINE_DOTDOT_SIZE;
1685 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1686 EXT4_INLINE_DOTDOT_SIZE;
1688 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1689 inline_size = ext4_get_inline_size(dir) -
1690 EXT4_MIN_INLINE_DATA_SIZE;
1693 BUFFER_TRACE(bh, "get_write_access");
1694 err = ext4_journal_get_write_access(handle, bh);
1698 err = ext4_generic_delete_entry(handle, dir, de_del, bh,
1699 inline_start, inline_size, 0);
1703 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1704 err = ext4_mark_inode_dirty(handle, dir);
1708 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1710 up_write(&EXT4_I(dir)->xattr_sem);
1713 ext4_std_error(dir->i_sb, err);
1718 * Get the inline dentry at offset.
1720 static inline struct ext4_dir_entry_2 *
1721 ext4_get_inline_entry(struct inode *inode,
1722 struct ext4_iloc *iloc,
1723 unsigned int offset,
1724 void **inline_start,
1729 BUG_ON(offset > ext4_get_inline_size(inode));
1731 if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1732 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1733 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1735 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1736 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1737 *inline_size = ext4_get_inline_size(inode) -
1738 EXT4_MIN_INLINE_DATA_SIZE;
1742 *inline_start = inline_pos;
1743 return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1746 int empty_inline_dir(struct inode *dir, int *has_inline_data)
1748 int err, inline_size;
1749 struct ext4_iloc iloc;
1751 unsigned int offset;
1752 struct ext4_dir_entry_2 *de;
1755 err = ext4_get_inode_loc(dir, &iloc);
1757 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block",
1762 down_read(&EXT4_I(dir)->xattr_sem);
1763 if (!ext4_has_inline_data(dir)) {
1764 *has_inline_data = 0;
1768 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1769 if (!le32_to_cpu(de->inode)) {
1770 ext4_warning(dir->i_sb,
1771 "bad inline directory (dir #%lu) - no `..'",
1777 offset = EXT4_INLINE_DOTDOT_SIZE;
1778 while (offset < dir->i_size) {
1779 de = ext4_get_inline_entry(dir, &iloc, offset,
1780 &inline_pos, &inline_size);
1781 if (ext4_check_dir_entry(dir, NULL, de,
1782 iloc.bh, inline_pos,
1783 inline_size, offset)) {
1784 ext4_warning(dir->i_sb,
1785 "bad inline directory (dir #%lu) - "
1786 "inode %u, rec_len %u, name_len %d"
1788 dir->i_ino, le32_to_cpu(de->inode),
1789 le16_to_cpu(de->rec_len), de->name_len,
1794 if (le32_to_cpu(de->inode)) {
1798 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1802 up_read(&EXT4_I(dir)->xattr_sem);
1807 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1811 down_write(&EXT4_I(inode)->xattr_sem);
1812 ret = ext4_destroy_inline_data_nolock(handle, inode);
1813 up_write(&EXT4_I(inode)->xattr_sem);
1818 int ext4_inline_data_fiemap(struct inode *inode,
1819 struct fiemap_extent_info *fieinfo,
1820 int *has_inline, __u64 start, __u64 len)
1824 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
1827 struct ext4_iloc iloc;
1829 down_read(&EXT4_I(inode)->xattr_sem);
1830 if (!ext4_has_inline_data(inode)) {
1834 inline_len = min_t(size_t, ext4_get_inline_size(inode),
1835 i_size_read(inode));
1836 if (start >= inline_len)
1838 if (start + len < inline_len)
1839 inline_len = start + len;
1840 inline_len -= start;
1842 error = ext4_get_inode_loc(inode, &iloc);
1846 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1847 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1848 physical += offsetof(struct ext4_inode, i_block);
1851 error = fiemap_fill_next_extent(fieinfo, start, physical,
1855 up_read(&EXT4_I(inode)->xattr_sem);
1856 return (error < 0 ? error : 0);
1860 * Called during xattr set, and if we can sparse space 'needed',
1861 * just create the extent tree evict the data to the outer block.
1863 * We use jbd2 instead of page cache to move data to the 1st block
1864 * so that the whole transaction can be committed as a whole and
1865 * the data isn't lost because of the delayed page cache write.
1867 int ext4_try_to_evict_inline_data(handle_t *handle,
1868 struct inode *inode,
1872 struct ext4_xattr_entry *entry;
1873 struct ext4_inode *raw_inode;
1874 struct ext4_iloc iloc;
1876 error = ext4_get_inode_loc(inode, &iloc);
1880 raw_inode = ext4_raw_inode(&iloc);
1881 entry = (struct ext4_xattr_entry *)((void *)raw_inode +
1882 EXT4_I(inode)->i_inline_off);
1883 if (EXT4_XATTR_LEN(entry->e_name_len) +
1884 EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) {
1889 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
1895 void ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1898 int inline_size, value_len, needed_blocks;
1901 struct ext4_xattr_ibody_find is = {
1902 .s = { .not_found = -ENODATA, },
1904 struct ext4_xattr_info i = {
1905 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1906 .name = EXT4_XATTR_SYSTEM_DATA,
1910 needed_blocks = ext4_writepage_trans_blocks(inode);
1911 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1915 down_write(&EXT4_I(inode)->xattr_sem);
1916 if (!ext4_has_inline_data(inode)) {
1918 ext4_journal_stop(handle);
1922 if (ext4_orphan_add(handle, inode))
1925 if (ext4_get_inode_loc(inode, &is.iloc))
1928 down_write(&EXT4_I(inode)->i_data_sem);
1929 i_size = inode->i_size;
1930 inline_size = ext4_get_inline_size(inode);
1931 EXT4_I(inode)->i_disksize = i_size;
1933 if (i_size < inline_size) {
1934 /* Clear the content in the xattr space. */
1935 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1936 if (ext4_xattr_ibody_find(inode, &i, &is))
1939 BUG_ON(is.s.not_found);
1941 value_len = le32_to_cpu(is.s.here->e_value_size);
1942 value = kmalloc(value_len, GFP_NOFS);
1946 if (ext4_xattr_ibody_get(inode, i.name_index, i.name,
1951 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1952 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1953 if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is))
1957 /* Clear the content within i_blocks. */
1958 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1959 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1960 memset(p + i_size, 0,
1961 EXT4_MIN_INLINE_DATA_SIZE - i_size);
1964 EXT4_I(inode)->i_inline_size = i_size <
1965 EXT4_MIN_INLINE_DATA_SIZE ?
1966 EXT4_MIN_INLINE_DATA_SIZE : i_size;
1970 up_write(&EXT4_I(inode)->i_data_sem);
1973 up_write(&EXT4_I(inode)->xattr_sem);
1976 ext4_orphan_del(handle, inode);
1978 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1979 ext4_mark_inode_dirty(handle, inode);
1981 ext4_handle_sync(handle);
1983 ext4_journal_stop(handle);
1987 int ext4_convert_inline_data(struct inode *inode)
1989 int error, needed_blocks;
1991 struct ext4_iloc iloc;
1993 if (!ext4_has_inline_data(inode)) {
1994 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1998 needed_blocks = ext4_writepage_trans_blocks(inode);
2001 error = ext4_get_inode_loc(inode, &iloc);
2005 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2006 if (IS_ERR(handle)) {
2007 error = PTR_ERR(handle);
2011 down_write(&EXT4_I(inode)->xattr_sem);
2012 if (!ext4_has_inline_data(inode)) {
2013 up_write(&EXT4_I(inode)->xattr_sem);
2017 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2018 up_write(&EXT4_I(inode)->xattr_sem);
2020 ext4_journal_stop(handle);