[Info-vax] disabling a CTRL/Y ast

[VAXman- at SendSpamHere.ORG]

In article <878f843d-0a3e-4717-a02c-a0bf4f448be1 at googlegroups.com>, Tom Adams <tadamsmar at gmail.com> writes:
>On Thursday, January 7, 2016 at 1:37:55 PM UTC-5, VAXman- wrote:
>> In article <fd6d6c6a-1c7d-432d-ac2b-1244dfa0fbb6 at googlegroups.com>, Tom Adams <tadamsmar at gmail.com> writes:
>> >On Thursday, January 7, 2016 at 11:21:27 AM UTC-5, VAXman- wrote:
>> >> In article <3a583568-e33e-446e-9805-6ea94f9ebf71 at googlegroups.com>, Tom Adams <tadamsmar at gmail.com> writes:
>> >> >On Wednesday, January 6, 2016 at 12:22:06 PM UTC-5, Stephen Hoffman wrote:
>> >> >> On 2016-01-06 16:46:32 +0000, Tom Adams said:
>> >> >> 
>> >> >> > I want to turn ctrl/y into a noop in certain regions of code and turn 
>> >> >> > ctrl/y into an exit in other regions of code.
>> >> >> 
>> >> >> Some background: I've posted more than a few replies here in the 
>> >> >> comp.os.vms newsgroup, specifically discussing the particular 
>> >> >> application in question and its general design, and an unfortunate 
>> >> >> aggregation of the UI and the more critical code paths, and your 
>> >> >> understandable preference to avoid restructuring the application code 
>> >> >> through reworking the existing design to block ^Y from arriving within 
>> >> >> the most critical sections of the code.
>> >> >> 
>> >> >> > I do find that it is necessary to run lib$disable_ctrl at the beginning 
>> >> >> > of the program and lib$enable_ctrl at the end of the program to restore 
>> >> >> > the previous CLI state, but I need to run asts to do the rest.
>> >> >> 
>> >> >> If you want ASTs in some parts of the code and not in others, then 
>> >> >> you're going to have to wrap those sections appropriately.   That's if 
>> >> >> you don't decide to migrate those critical code sections into another 
>> >> >> process context, or related steps.  Because beyond the potential for ^Y 
>> >> >> in these sections and the associated issues you're having with the 
>> >> >> image rundown processing in the absence of an EXIT command, other sorts 
>> >> >> of generic application failures ("bugs") or any arriving $forcex calls 
>> >> >> or such activities can potentially still leave the application in an 
>> >> >> indeterminate state.
>> >> >> 
>> >> >> 
>> >> >> 
>> >> >> -- 
>> >> >> Pure Personal Opinion | HoffmanLabs LLC
>> >> >
>> >> >Here's an example of what I am trying to do that seems to work OK.
>> >> >I guess a forced exit from a bug could leave the CTRL\Y in
>> >> >the disabled state at the CLI instead of the original state.
>> >> >
>> >> >Also, it's possible that someone could define TT to be some terminal
>> >> >other than the login terminal.  VMS leaves all the built-in process
>> >> >logicals that refer to the login terminal vulnerable to redefinition.
>> >> >It may be possible to roll your own safe (executive, no_alias) logical
>> >> >name via a call in syslogin.
>> >> >
>> >> >Also, I still need to cover the cases where this code runs in a context
>> >> >where TT is not defined.
>> >> 
>> >> 	PROGRAM MY_TERMINAL
>> >> 	IMPLICIT NONE
>> >> 
>> >> 	INCLUDE '($DSCDEF)'
>> >> 	INCLUDE '($JPIDEF)'
>> >> 	INCLUDE '(LIB$ROUTINES)'
>> >> 
>> >> 	INTEGER*4 STATUS
>> >> 
>> >> 	RECORD	/DSCDEF1/ TERMINAL
>> >> 
>> >> 	TERMINAL.DSC$W_MAXSTRLEN = 0
>> >> 	TERMINAL.DSC$B_DTYPE = DSC$K_DTYPE_T
>> >> 	TERMINAL.DSC$B_CLASS = DSC$K_CLASS_D
>> >> 	TERMINAL.DSC$A_POINTER = 0
>> >> 
>> >> 	STATUS = LIB$GETJPI(JPI$_TERMINAL,,,,TERMINAL,)
>> >> 	IF (STATUS.AND.1) THEN
>> >> 	  STATUS = LIB$PUT_OUTPUT(TERMINAL)
>> >> 	ENDIF
>> >> 
>> >> 	CALL SYS$EXIT(%VAL(STATUS))
>> >> 	END
>> >> 
>> >> -- 
>> >> VAXman- A Bored Certified VMS Kernel Mode Hacker    VAXman(at)TMESIS(dot)ORG
>> >> 
>> >> I speak to machines with the voice of humanity.
>> >
>> >That's better than TT.
>> >
>> >But that is also vulnerable to a non-priv user. For instance
>> >if the login terminal is RTA1:  the a non-priv user can define RTA1
>> >thusly:
>> >
>> >DEFINE RTA1 LTA180:
>> >
>> >Where lta180: is a valid device name.  Calls like SYS$ASSIGN will dutifully
>> >translate RTA1: to LTA180: assign it a channel, and return a
>> >normal status.
>> >
>> >Admittedly, it's not very likely that a non-priv user would do that.
>> 
>> Unless your nonprivied user as discovered a way to write PCB[PCB$T_TERMINAL],
>> I doubt you'll have much to worry about. ;)
>> 
>Not sure what you mean?
>
>Do you mean that a non-priv user probably does not have alternative valid name
>to use.  You are right about that.

I mean that your nonprivied user will not be able to alter the system memory
where the terminal name is obtained by the SYS$GETJPI request.  I've used the
LIB$ wrapper around SYS$GETJPI because I can pass a dynamic string descriptor
to it and VMS will allocate some process memory into which it will write the 
terminal name.

-- 
VAXman- A Bored Certified VMS Kernel Mode Hacker    VAXman(at)TMESIS(dot)ORG

I speak to machines with the voice of humanity.