Skip to contentSkip to conversation

inode: For readonly filesystem, func file_update_time should return -EROFS rather than zero.

PATCH In review majianpeng 2 participants 2 messages Oct 15, 2013 VFS

lore
Share & tools
lore

majianpeng

author
#0 lore raw

PATCH 1/1 inode: For readonly filesystem, func file_update_time should return -EROFS rather than zero.

For ext2 mounted with errors=remount-ro, if write a file and because the harddisk error, the fs became ro.But the following test can't return.

int main()

{
	int ret;
	int i = 0;
	char buff[SIZE];
	int fd = open("/opt/test", O_WRONLY|O_TRUNC|O_CREAT);

	if (fd < 0) {
		printf("open error %s\n", strerror(errno));
		return errno;
	}

	ret = write(fd, buff, SIZE);
	lseek(fd, 0, 0);

	while (1) {
		ret = write(fd, buff, SIZE);
		if (ret < 0) {
			printf("write error %s\n", strerror(errno));
			break;
		}
		lseek(fd, 0, 0);
		i++;
	}
	printf("write count=%d\n", i);
	close(fd);

	return 0;
}

For ext3/ext4, because jbd the test can return. But for ext2, because no jbd and not reading bitmap from harddisk, the test continue.

So we should add check readonly-fs on write-path. Func file_update_time already check the readonly flag,but it can't return -EROFS. For readonly-fs, it can't update a/c/m time of file,it should return -EROFS rather than zero.

---
 fs/inode.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
fs/inode.c +3 -2
index b33ba8e..65302c1 100644
@@ -1644,8 +1644,9 @@ int file_update_time(struct file *file)
 		return 0;
 
 	/* Finally allowed to write? Takes lock. */
-	if (__mnt_want_write_file(file))
-		return 0;
+	ret = __mnt_want_write_file(file);
+	if (ret)
+		return ret;
 
 	ret = update_time(inode, &now, sync_it);
 	__mnt_drop_write_file(file);
Signature · 1 line
1.8.4
reviewer Reviewed-by on v1 1/1
#1 lore raw
Quoted 33 lines · majianpeng

For ext2 mounted with errors=remount-ro, if write a file and because the harddisk error, the fs became ro.But the following test can't return.

int main()

{
	int ret;
	int i = 0;
	char buff[SIZE];
	int fd = open("/opt/test", O_WRONLY|O_TRUNC|O_CREAT);

	if (fd < 0) {
		printf("open error %s\n", strerror(errno));
		return errno;
	}

	ret = write(fd, buff, SIZE);
	lseek(fd, 0, 0);

	while (1) {
		ret = write(fd, buff, SIZE);
		if (ret < 0) {
			printf("write error %s\n", strerror(errno));
			break;
		}
		lseek(fd, 0, 0);
		i++;
	}
	printf("write count=%d\n", i);
	close(fd);

	return 0;
}
Quoted 7 lines

For ext3/ext4, because jbd the test can return. But for ext2, because no jbd and not reading bitmap from harddisk, the test continue.

So we should add check readonly-fs on write-path. Func file_update_time already check the readonly flag,but it can't return -EROFS. For readonly-fs, it can't update a/c/m time of file,it should return -EROFS rather than zero.

The patch looks good to me so feel free to add

Just the reasoning in the changelog doesn't look completely correct - we
are free to succeed doing write to a FMODE_WRITE file descriptor even if the filesystem below it decided to declare game over and switched itself to read-only mode. It is actually ext2/3/4 which is problematic - its switching to read-only mode is rather hacky as we cannot really do it properly from the arbitrary context in which we handle errors. It's not VFS' fault it doesn't notice the switching...

Honza

Quoted 23 lines
---
 fs/inode.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
fs/inode.c +3 -2
index b33ba8e..65302c1 100644
@@ -1644,8 +1644,9 @@ int file_update_time(struct file *file)
 		return 0;
 
 	/* Finally allowed to write? Takes lock. */
-	if (__mnt_want_write_file(file))
-		return 0;
+	ret = __mnt_want_write_file(file);
+	if (ret)
+		return ret;
 
 	ret = update_time(inode, &now, sync_it);
 	__mnt_drop_write_file(file);
Signature · 1 line
1.8.4
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR