[PATCH] ubifs: Fix inconsistent inode size when powercut happens during appendant writing
Zhihao Cheng
chengzhihao1 at huawei.com
Tue Jun 4 00:52:16 PDT 2024
UBIFS always make sure that the data length won't beyond the inode size
by writing inode before writing page(See ubifs_writepage.). After commit
c35acef383f4a2f2cfc30("ubifs: Convert ubifs_writepage to use a folio"),
the rule is broken in one case: Given a file with size 3, then write 4096
from the offset 0, following process will make inode size be smaller than
file data length after powercut & recovery:
P1 P2
ubifs_writepage
len = folio_size(folio) // 4096
if (folio_pos(folio) + len <= i_size) // condition 1: 0 + 4096 <= 4096
//(i_size is updated as 4096 in ubifs_write_end)
if (folio_pos(folio) >= synced_i_size) // condition 2: 0 >= 3, false
write_inode // Skipped, because condition 2 is false
do_writepage(folio, len) // write one page
do_commit // data node won't be replayed in next mounting
>> Powercut <<
So, inode size(4096) is not updated into disk, we will get following
error messages in next mounting(chk_fs = 1):
check_leaf [ubifs]: data node at LEB 14:2048 is not within inode size 3
dbg_walk_index [ubifs]: leaf checking function returned error -22, for
leaf at LEB 14:2048
Fix it by modifying condition 2 as original comparison(Compare the page
index of synced_i_size with current page index).
Fixes: c35acef383f4 ("ubifs: Convert ubifs_writepage to use a folio")
Link: https://bugzilla.kernel.org/show_bug.cgi?id=218934
Signed-off-by: Zhihao Cheng <chengzhihao1 at huawei.com>
---
fs/ubifs/file.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/fs/ubifs/file.c b/fs/ubifs/file.c
index a1f46919934c..b32578783382 100644
--- a/fs/ubifs/file.c
+++ b/fs/ubifs/file.c
@@ -1027,7 +1027,7 @@ static int ubifs_writepage(struct folio *folio, struct writeback_control *wbc,
/* Is the folio fully inside i_size? */
if (folio_pos(folio) + len <= i_size) {
- if (folio_pos(folio) >= synced_i_size) {
+ if (folio->index >= synced_i_size >> PAGE_SHIFT) {
err = inode->i_sb->s_op->write_inode(inode, NULL);
if (err)
goto out_redirty;
--
2.39.2
More information about the linux-mtd
mailing list