Skip to contentSkip to conversation

ext2: fix fs corruption when trying to remove a non-empty directory with IO error

PATCH In review Ye Bin 2 participants 2 messages Jun 15, 2022 ext2 file system

lore
Share & tools
lore

Ye Bin

author
#0 lore raw

PATCH 1/1 ext2: fix fs corruption when trying to remove a non-empty directory with IO error

We got issue as follows:
[home]# mount /dev/sdd test
[home]# cd test
[test]# ls
dir1 lost+found
[test]# rmdir dir1
ext2_empty_dir: inject fault
[test]# ls
lost+found
[test]# cd ..
[home]# umount test
[home]# fsck.ext2 -fn /dev/sdd
e2fsck 1.42.9 (28-Dec-2013)
Pass 1: Checking inodes, blocks, and sizes
Inode 4065, i_size is 0, should be 1024. Fix? no

Pass 2: Checking directory structure
Pass 3: Checking directory connectivity
Unconnected directory inode 4065 (/???)
Connect to /lost+found? no

'..' in ... (4065) is / (2), should be <The NULL inode> (0). Fix? no

Pass 4: Checking reference counts
Inode 2 ref count is 3, should be 4. Fix? no

Inode 4065 ref count is 2, should be 3. Fix? no

Pass 5: Checking group summary information

/dev/sdd: ********** WARNING: Filesystem still has errors **********

/dev/sdd: 14/128016 files (0.0% non-contiguous), 18477/512000 blocks

Reason is same with commit 7aab5c84a0f6. We can't assume directory is empty when read directory entry failed.

---
 fs/ext2/dir.c | 9 +++------
 1 file changed, 3 insertions(+), 6 deletions(-)
fs/ext2/dir.c +3 -6
index 3bd5772b401b..8f597753ac12 100644
@@ -672,17 +672,14 @@ int ext2_empty_dir (struct inode * inode)
 	void *page_addr = NULL;
 	struct page *page = NULL;
 	unsigned long i, npages = dir_pages(inode);
-	int dir_has_error = 0;
 
 	for (i = 0; i < npages; i++) {
 		char *kaddr;
 		ext2_dirent * de;
-		page = ext2_get_page(inode, i, dir_has_error, &page_addr);
+		page = ext2_get_page(inode, i, 0, &page_addr);
 
-		if (IS_ERR(page)) {
-			dir_has_error = 1;
-			continue;
-		}
+		if (IS_ERR(page))
+			goto not_empty;
 
 		kaddr = page_addr;
 		de = (ext2_dirent *)kaddr;
Signature · 1 line
2.31.1
#1 lore raw
Quoted 31 lines · Ye Bin

We got issue as follows:
[home]# mount /dev/sdd test
[home]# cd test
[test]# ls
dir1 lost+found
[test]# rmdir dir1
ext2_empty_dir: inject fault
[test]# ls
lost+found
[test]# cd ..
[home]# umount test
[home]# fsck.ext2 -fn /dev/sdd
e2fsck 1.42.9 (28-Dec-2013)
Pass 1: Checking inodes, blocks, and sizes
Inode 4065, i_size is 0, should be 1024. Fix? no

Pass 2: Checking directory structure
Pass 3: Checking directory connectivity
Unconnected directory inode 4065 (/???)
Connect to /lost+found? no

'..' in ... (4065) is / (2), should be <The NULL inode> (0). Fix? no

Pass 4: Checking reference counts
Inode 2 ref count is 3, should be 4. Fix? no

Inode 4065 ref count is 2, should be 3. Fix? no

Pass 5: Checking group summary information

Quoted 8 lines

/dev/sdd: ********** WARNING: Filesystem still has errors **********

/dev/sdd: 14/128016 files (0.0% non-contiguous), 18477/512000 blocks

Reason is same with commit 7aab5c84a0f6. We can't assume directory is empty when read directory entry failed.

Honestly, I'm not very much concerned about corrupting the filesystem more when we've got EIO for metadata but OK, probably what you propose is a saner choice. I've added the patch to my tree.

Honza

Quoted 32 lines
---
 fs/ext2/dir.c | 9 +++------
 1 file changed, 3 insertions(+), 6 deletions(-)
fs/ext2/dir.c +3 -6
index 3bd5772b401b..8f597753ac12 100644
@@ -672,17 +672,14 @@ int ext2_empty_dir (struct inode * inode)
 	void *page_addr = NULL;
 	struct page *page = NULL;
 	unsigned long i, npages = dir_pages(inode);
-	int dir_has_error = 0;
 
 	for (i = 0; i < npages; i++) {
 		char *kaddr;
 		ext2_dirent * de;
-		page = ext2_get_page(inode, i, dir_has_error, &page_addr);
+		page = ext2_get_page(inode, i, 0, &page_addr);
 
-		if (IS_ERR(page)) {
-			dir_has_error = 1;
-			continue;
-		}
+		if (IS_ERR(page))
+			goto not_empty;
 
 		kaddr = page_addr;
 		de = (ext2_dirent *)kaddr;
Signature · 1 line
2.31.1
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR