Re: df displays wrong figures for network mounts

[Bruno Haible]

Paul Eggert wrote:
> I installed this
> patch into CVS coreutils.  I don't think it'll fix the problem in
> general, but does it fix your particular problem?

Thanks. The patch does have the effect that, in my case,

  address@hidden:/smb/ibook/data/temp $ df .

calls

  statfs(".", {f_type="SMB_SUPER_MAGIC", f_bsize=4096, f_blocks=3456896, 
f_bfree=253440, f_files=0, f_ffree=0, f_namelen=1024}) = 0

where it earlier called

  statfs("/smb/ibook", {f_type="SMB_SUPER_MAGIC", f_bsize=4096, 
f_blocks=3456896, f_bfree=253440, f_files=0, f_ffree=0, f_namelen=1024}) = 0

But the result in terms of blocks is still, unfortunately, the same.

From the Linux-2.4.21 and Linux-2.6.12.5 sources, it appears to be a
limitation in the Linux kernel. fs/open.c:

asmlinkage long sys_statfs(const char * path, struct statfs * buf)
{
        struct nameidata nd;
        int error;

        error = user_path_walk(path, &nd);
        if (!error) {
                struct statfs tmp;
                error = vfs_statfs(nd.dentry->d_inode->i_sb, &tmp);
                if (!error && copy_to_user(buf, &tmp, sizeof(struct statfs)))
                        error = -EFAULT;
                path_release(&nd);
        }
        return error;
}

As you can see, the original path is not passed to vfs_statfs(),
only the inode's superblock on the client (representing the mount or
mountpoint on the client) is.

Bruno