Re: Patch to make can_network stronger and remove nscd tunable.

[Daniel J Walsh <dwalsh@redhat.com>]

Russell Coker wrote:

>On Tue, 2 Nov 2004 03:18, Daniel J Walsh <dwalsh@redhat.com> wrote:
>
>+allow crond_t self:{ tcp_socket udp_socket } connect;
>
>crond.te has no can_network() invocation.  Maybe we should have the following 
>in the definition of uncond_can_ypbind():
>allow $1 self:{ tcp_socket udp_socket } connect;
>
>It seems that cnan_ypbind() is the only network use in crond.te.
>
>  
>
Ok I will change.

>-allow dictd_t self:capability { setuid setgid };
>+allow dictd_t self:capability { setuid setgid net_bind_service };
>
>dictd_t is not permitted to bind to any low ports.  How does it need 
>net_bind_service capability?
>
>  
>
Maybe ypbind also.

>+allow hald_t { device_t }:{ chr_file } { create_file_perms };
>
>Three sets of redundant braces.  Why does it need to create character device 
>nodes anyway?  We have udev to do that!
>
>  
>
Hal creates a device when using cardmgr. pcmcia currently does not work 
with udev.

>+file_type_auto_trans(kudzu_t, tmp_t, kudzu_tmp_t, chr_file)
>
>Why is kudzu creating device nodes under /tmp?  This sounds like a bug in 
>kudzu to me.
>  
>
I think cardmgr again.

>+dontaudit mailman_queue_t src_t:dir { search };
>
>I've filed a bugzilla about that one:
>https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=137863
>
>We should have ifdef(`hide_broken_symptoms', around it too.
>
>-allow mysqld_t self:capability { dac_override setgid setuid };
>+allow mysqld_t self:capability { dac_override setgid setuid 
>net_bind_service };
>
>Why does mysqld_t need name_bind_service?  It doesn't seem to be allowed to 
>bind to any low ports anyway.
>
>  
>
ypbind.

>-allow postfix_$1_t self:capability { setuid setgid dac_override };
>+allow postfix_$1_t self:capability { setuid setgid dac_override 
>net_bind_service };
>
>What is this for?  Which Postfix programs need such access?  Maybe you should 
>have net_bind_service inside the can_ypbind() macro or something.  Normal 
>Postfix operation does not need such a change.
>allow postfi
>  
>
ypbind.

I will add

allow $1_t self:capability net_bind_service; to ypbind.

>-allow radiusd_t self:capability { chown dac_override fsetid kill setgid 
>setuid sys_resource sys_tty_config };
>+allow radiusd_t self:capability { chown dac_override fsetid kill setgid 
>setuid sys_resource sys_tty_config net_bind_service };
>
>Once again, this should not be needed.
>
>
>If every instance of daemon_domain() is going to get nscd_client_domain added, 
>then perhaps we should just change the definition of daemon_domain() 
>accordingly?
>
>
>Why isn't allow $1 self:{ tcp_socket udp_socket } connect; in can_network()?
>
>
>  
>
Because we don't want all network daemons to be able to connect out.

>I think that some structural changes need to be made before any of the changes 
>in this can go in the CVS.
>
>  
>


--
This message was distributed to subscribers of the selinux mailing list.
If you no longer wish to subscribe, send mail to majordomo@tycho.nsa.gov with
the words "unsubscribe selinux" without quotes as the message.