Where does ext4 store directory sizes? Are they stored in the directory inode?
For example, when I run du -h, it returns directories' size instantly, so I don't believe it calculates it at that time.
I'm using ext4 on Linux.
1 Answer
Using strace would seem to indicate that the file sizes are indeed calculated by querying the files within the directory.
Example
Say I fill a directory with 3 1MB files.
$ mkdir adir
$ fallocate -l 1M adir/afile1.txt
$ fallocate -l 1M adir/afile2.txt
$ fallocate -l 1M adir/afile3.txt
Now when we trace the du -h command:
$ strace -s 2000 -o du.log du -h adir/
3.1M adir/
Looking at the resulting strace log file du.log:
...
newfstatat(AT_FDCWD, "adir/", {st_mode=S_IFDIR|0775, st_size=4096, ...}, AT_SYMLINK_NOFOLLOW) = 0
fcntl(3, F_DUPFD, 3) = 4
fcntl(4, F_GETFD) = 0
fcntl(4, F_SETFD, FD_CLOEXEC) = 0
getdents(3, /* 5 entries */, 32768) = 144
getdents(3, /* 0 entries */, 32768) = 0
close(3) = 0
newfstatat(4, "afile2.txt", {st_mode=S_IFREG|0644, st_size=1048576, ...}, AT_SYMLINK_NOFOLLOW) = 0
newfstatat(4, "afile3.txt", {st_mode=S_IFREG|0644, st_size=1048576, ...}, AT_SYMLINK_NOFOLLOW) = 0
newfstatat(4, "afile1.txt", {st_mode=S_IFREG|0644, st_size=1048576, ...}, AT_SYMLINK_NOFOLLOW) = 0
brk(0) = 0x231a000
...
Notice the newfstatat system calls? These are getting the size of each file in turn.
Additional Background
If you're interested here's a bit more on the subject.
- This behavior has nothing to do with EXT4. This is just how filesystems work in Unix.
The
statcommand provides no facility for querying anything other then the size of a filesystem object (directory or file).$ stat adir/ File: ‘adir/’ Size: 4096 Blocks: 8 IO Block: 4096 directory Device: fd02h/64770d Inode: 11539929 Links: 2 Access: (0775/drwxrwxr-x) Uid: ( 1000/ saml) Gid: ( 1000/ saml) Context: unconfined_u:object_r:user_home_t:s0 Access: 2014-04-15 22:29:25.289639888 -0400 Modify: 2014-04-15 22:29:44.977638542 -0400 Change: 2014-04-15 22:29:44.977638542 -0400 Birth: -Notice it's 4096 bytes. That's the actual size of the directory itself, not what it contains.
References
-
1I'd add that this isn't just ext4;
statsimply doesn't have a way to return multiple sizes—so it can only return the size of the directory itself, not of its contents. And also "directory size including contents" becomes less clear when you have hardlinked files.derobert– derobert2014-04-16 02:27:28 +00:00Commented Apr 16, 2014 at 2:27
dufeels instantaneous it's because the files are cached by the kernel. You can see this if you try to run it multiple times on a recently mounted file system - it'll be much faster after the first run.