File system stats are being queried with path reduced to mount dir

[zaltysz]

Is there an existing issue for this?

• I have searched the existing issues

Midnight Commander version and build configuration

4.8.33

Operating system

OpenSuse Tumbleweed with Kernel 6.18.7-1-default

Is this issue reproducible using the latest version of Midnight Commander?

• I confirm the issue is still reproducible with the latest version of Midnight Commander

How to reproduce

cifs or sshfs mount to remote system, where exported tree is composed of multiple file systems (via mount points). I.e.:

/ (/dev/sda1 mounted)
/data
/data/d1 (/dev/sdb1 mounted)
/data/d2 (/dev/sdb2 mounted)

Use mc to traverse such mount and observe reported space usage info.

Note: nfs is not suitable for reproduction, because it allows only single fs per export, while reproduction needs multiple fs per export.

Expected behavior

Path dependent space usage reported, or in other words: space usage of fs where current dir resides on remote system is reported. Should coincide with what "df -h ." returns.

I.e. entering /data/d1 and /data/d2 should show space usage for /dev/sdb1 and /dev/sdb2 respectively.

Actual behavior

Space usage of fs where top of exported tree resides is always reported even when entering dir residing on different fs on remote system.

I.e. whenever /data/d1 or /data/d2 is entered, space usage for / (/dev/sda1) is always shown.

Additional context

In source: my_statfs(stats,path) loops over mount list and determines mount_dir for given path, then calls get_fs_usage(mount_dir,null,fs_use), which effectively feeds that mount_dir to statvfs() on POSIX system. This is why mc always queries fs on which top of exported tree resides.

At least on Linux, statvfs() accepts full path and then magic follows for such cases. So, trivial fix would be to pass path instead of mount_dir to get_fs_usage(). No, idea about other OSes though.

[egmontkob]

Thanks for the bugreport!

POSIX says:

[...] obtain information about the file system containing the file [...]

and later in the examples

obtain file system information for the file system upon which the file named /home/cnd/mod1 resides

FreeBSD and macOS say:

Applications must pass a pathname or file descriptor which refers to a file on the file system in which they are interested.

Solaris says:

The path argument should name a file that resides on that file system.

I think it's pretty safe to say that the suggested change won't cause any harm. It's pretty reasonable that df and friends (incl. mc) shouldn't have to go into the business of locating the mount point.