In current 2.5, the sys_nfsservctl has been fundamentally changed. The functionality has been moved to a filesystem called nfsd. I see three options for supporting nfsservctl hooks as we move forward: 1) Simple hook that only checks the command, not the args (standard copy_from_user TOCTTOU race avoided this way). As the comments in the fs/nfsctl.c source suggest, this hook would disappear once this functionality treated as a standard filesystem. The first attachment is a rough idea of this approach. 2) Add code to the nfs server module that specifically calls out for each command with the args (after copy_from_user). This directly preserves current functionality, but is more invasive in the nfs server. The second attachment is a rough idea of this approach. 3) This is a filesystem, remove the nfsservctl hook since we have the standard filesystem hooks. This conceptually preserves our current stance which tries to avoid placing filesystem specific hooks. And reading the data passed in the buffers during a file write seems a bit strange. Labelling via standard post_lookup is a little tough, since currently this is an in kernel fs and it manually populates the dcache. I bring this up because I think this is an example of how filesystems will be more prolific in 2.5+ Thoughts/flames? -chris
This archive was generated by hypermail 2b30 : Thu Apr 11 2002 - 23:57:55 PDT