R devel - Apr 2005 - Ctrl-c crashes R when run as sudo (PR#7819)

I tried to submit this in R, but not sure if it worked.

When running R as sudo, using ctrl-c dumps me to the command line.
Hitting exit to exit the terminal window results in R taking 100% of
resources.

I am using R-2.1.0 on Fedora Core 3.

Thanks.

Manuel

On Tue, 2005-04-26 at 16:47 +0200, Manuel.A.Morales@williams.edu
wrote:
> I tried to submit this in R, but not sure if it worked.
> 
> When running R as sudo, using ctrl-c dumps me to the command line.
> Hitting exit to exit the terminal window results in R taking 100% of
> resources.
> 
> I am using R-2.1.0 on Fedora Core 3.
> 
> Thanks.
> 
> Manuel
I suspect that we are going to need more information.

Running on a fully updated FC3 box using Xfce's Terminal, gnome-terminal
and kconsole, I cannot duplicate this issue either as a user or as root
using sudo.

Attach the output of:
> R.version 
Was there a program running in R at the time, or were you at the R
prompt when you hit ctrl-c?

Which desktop environment and terminal emulator are you using? 

Have you perhaps re-mapped ctrl-c using modified keybindings?

Marc

On Tue, 2005-04-26 at 11:41 -0400, Manuel Morales wrote:
> On Tue, 2005-04-26 at 10:34 -0500, Marc Schwartz wrote:
> > On Tue, 2005-04-26 at 16:47 +0200, Manuel.A.Morales@williams.edu
wrote:
> > > I tried to submit this in R, but not sure if it worked.
> > > 
> > > When running R as sudo, using ctrl-c dumps me to the command
line.
> > > Hitting exit to exit the terminal window results in R taking 100%
of
> > > resources.
> > > 
> > > I am using R-2.1.0 on Fedora Core 3.
> > > 
> > > Thanks.
> > > 
> > > Manuel
> > 
> > I suspect that we are going to need more information.
> > 
> > Running on a fully updated FC3 box using Xfce's Terminal,
gnome-terminal
> > and kconsole, I cannot duplicate this issue either as a user or as
root
> > using sudo.
> > 
> Note that this doesn't happen if I run R as a normal user or as root.
> I.e.,
> 
> R
> <ctrl-c>
> or
> 
> su
> R
> <ctrl-c>
> works fine.
> 
> sudo R
> <ctrl-c>
> does not.
> 
> > Attach the output of:
> > 
> > > R.version 
> > 
> 
> > R.version
>          _
> platform i686-pc-linux-gnu
> arch     i686
> os       linux-gnu
> system   i686, linux-gnu
> status
> major    2
> minor    1.0
> year     2005
> month    04
> day      18
> language R
> 
> 
> > Was there a program running in R at the time, or were you at the R
> > prompt when you hit ctrl-c?
> > 
> 
> R prompt
> 
> > Which desktop environment and terminal emulator are you using? 
> > 
>  
> Gnome and gnome-terminal, although the problem also happens when using
> xterm.
> 
> > Have you perhaps re-mapped ctrl-c using modified keybindings?
> 
> No. And ctrl-c works fine as a normal user or as root.
Manuel,

Please be sure to copy R-bugs@biostat.ku.dk when you reply so that the
thread is properly archived in the bug tracking system.

I tried this under GNOME using gnome-terminal and the unpatched version
of R 2.1.0 and still could not duplicate the problem.

Can you post back with the contents of /etc/sudoers?

The other possibility would be to run:

sudo R -d gdb

in gnome-terminal. When you get to the (gdb) prompt, type 'r' (without
the quotes) for run, which will start up R.

If you can replicate the problem under gdb and R exits after a ctrl-c,
you should be left at a (gdb) prompt. If that occurs, type 'bt' (without
the quotes), which will display a backtrace. If that all works, post the
output of gdb from the crtl-c through and including the backtrace here.

Marc

On Tue, 2005-04-26 at 11:41 -0400, Manuel Morales wrote:
> On Tue, 2005-04-26 at 10:34 -0500, Marc Schwartz wrote:
> > On Tue, 2005-04-26 at 16:47 +0200, Manuel.A.Morales@williams.edu
wrote:
> > > I tried to submit this in R, but not sure if it worked.
> > > 
> > > When running R as sudo, using ctrl-c dumps me to the command
line.
> > > Hitting exit to exit the terminal window results in R taking 100%
of
> > > resources.
> > > 
> > > I am using R-2.1.0 on Fedora Core 3.
> > > 
> > > Thanks.
> > > 
> > > Manuel
> > 
> > I suspect that we are going to need more information.
> > 
> > Running on a fully updated FC3 box using Xfce's Terminal,
gnome-terminal
> > and kconsole, I cannot duplicate this issue either as a user or as
root
> > using sudo.
> > 
> Note that this doesn't happen if I run R as a normal user or as root.
> I.e.,
> 
> R
> <ctrl-c>
> or
> 
> su
> R
> <ctrl-c>
> works fine.
> 
> sudo R
> <ctrl-c>
> does not.
> 
> > Attach the output of:
> > 
> > > R.version 
> > 
> 
> > R.version
>          _
> platform i686-pc-linux-gnu
> arch     i686
> os       linux-gnu
> system   i686, linux-gnu
> status
> major    2
> minor    1.0
> year     2005
> month    04
> day      18
> language R
> 
> 
> > Was there a program running in R at the time, or were you at the R
> > prompt when you hit ctrl-c?
> > 
> 
> R prompt
> 
> > Which desktop environment and terminal emulator are you using? 
> > 
>  
> Gnome and gnome-terminal, although the problem also happens when using
> xterm.
> 
> > Have you perhaps re-mapped ctrl-c using modified keybindings?
> 
> No. And ctrl-c works fine as a normal user or as root.
Manuel,

Please be sure to copy R-bugs@biostat.ku.dk when you reply so that the
thread is properly archived in the bug tracking system.

I tried this under GNOME using gnome-terminal and the unpatched version
of R 2.1.0 and still could not duplicate the problem.

Can you post back with the contents of /etc/sudoers?

The other possibility would be to run:

sudo R -d gdb

in gnome-terminal. When you get to the (gdb) prompt, type 'r' (without
the quotes) for run, which will start up R.

If you can replicate the problem under gdb and R exits after a ctrl-c,
you should be left at a (gdb) prompt. If that occurs, type 'bt' (without
the quotes), which will display a backtrace. If that all works, post the
output of gdb from the crtl-c through and including the backtrace here.

Marc

On Tue, 2005-04-26 at 11:30 -0500, Marc Schwartz wrote:
> On Tue, 2005-04-26 at 11:41 -0400, Manuel Morales wrote:
> > On Tue, 2005-04-26 at 10:34 -0500, Marc Schwartz wrote:
> > > On Tue, 2005-04-26 at 16:47 +0200, Manuel.A.Morales@williams.edu
wrote:
> > > > I tried to submit this in R, but not sure if it worked.
> > > > 
> > > > When running R as sudo, using ctrl-c dumps me to the command
line.
> > > > Hitting exit to exit the terminal window results in R taking
100% of
> > > > resources.
> > > > 
> > > > I am using R-2.1.0 on Fedora Core 3.
> > > > 
> > > > Thanks.
> > > > 
> > > > Manuel
> > > 
> > > I suspect that we are going to need more information.
> > > 
> > > Running on a fully updated FC3 box using Xfce's Terminal,
gnome-terminal
> > > and kconsole, I cannot duplicate this issue either as a user or
as root
> > > using sudo.
> > > 
> > Note that this doesn't happen if I run R as a normal user or as
root.
> > I.e.,
> > 
> > R
> > <ctrl-c>
> > or
> > 
> > su
> > R
> > <ctrl-c>
> > works fine.
> > 
> > sudo R
> > <ctrl-c>
> > does not.
> > 
> > > Attach the output of:
> > > 
> > > > R.version 
> > > 
> > 
> > > R.version
> >          _
> > platform i686-pc-linux-gnu
> > arch     i686
> > os       linux-gnu
> > system   i686, linux-gnu
> > status
> > major    2
> > minor    1.0
> > year     2005
> > month    04
> > day      18
> > language R
> > 
> > 
> > > Was there a program running in R at the time, or were you at the
R
> > > prompt when you hit ctrl-c?
> > > 
> > 
> > R prompt
> > 
> > > Which desktop environment and terminal emulator are you using? 
> > > 
> >  
> > Gnome and gnome-terminal, although the problem also happens when using
> > xterm.
> > 
> > > Have you perhaps re-mapped ctrl-c using modified keybindings?
> > 
> > No. And ctrl-c works fine as a normal user or as root.
> 
> Manuel,
> 
> Please be sure to copy R-bugs@biostat.ku.dk when you reply so that the
> thread is properly archived in the bug tracking system.
> 
> I tried this under GNOME using gnome-terminal and the unpatched version
> of R 2.1.0 and still could not duplicate the problem.
> 
> Can you post back with the contents of /etc/sudoers?
> 
> The other possibility would be to run:
> 
> sudo R -d gdb
> 
> in gnome-terminal. When you get to the (gdb) prompt, type 'r'
(without
> the quotes) for run, which will start up R.
> 
> If you can replicate the problem under gdb and R exits after a ctrl-c,
> you should be left at a (gdb) prompt. If that occurs, type 'bt'
(without
> the quotes), which will display a backtrace. If that all works, post the
> output of gdb from the crtl-c through and including the backtrace here.
> 
> Marc
> 
>
Program received signal SIGINT, Interrupt.
0x006547a2 in _dl_sysinfo_int80 () from /lib/ld-linux.so.2
(gdb) bt
#0  0x006547a2 in _dl_sysinfo_int80 () from /lib/ld-linux.so.2
#1  0x0072da1d in ___newselect_nocancel () from /lib/tls/libc.so.6
#2  0x0813e060 in R_SelectEx (n=1, readfds=0x825dde0, writefds=0x0,
    exceptfds=0x0, timeout=0x0, intr=0x813e5a8 <handleInterrupt>)
    at sys-std.c:138
#3  0x0813e281 in R_checkActivityEx (usec=-1, ignore_stdin=0,
    intr=0x813e5a8 <handleInterrupt>) at sys-std.c:302
#4  0x0813e6f0 in Rstd_ReadConsole (prompt=0x0, buf=0xbfffb33c "",
len=1024,
    addtohistory=-514) at sys-std.c:642
#5  0x080cac4a in Rf_ReplIteration (rho=0x9522cb8, savestack=-514,
    browselevel=0, state=0xbfffb330) at main.c:208
#6  0x080cadef in R_ReplConsole (rho=0x9522cb8, savestack=0,
browselevel=0)
    at main.c:306
#7  0x080cb03d in run_Rmainloop () at main.c:685
#8  0x0805d948 in main (ac=1, av=0xbfffb844) at Rmain.c:31

The relevant line in my /etc/sudoers:
mmorales        ALL=(ALL) ALL

On 26.04.2005, at 12:52, Manuel.A.Morales@williams.edu wrote:
> Program received signal SIGINT, Interrupt.
> 0x006547a2 in _dl_sysinfo_int80 () from /lib/ld-linux.so.2
Thanks you, however Marc omitted to mention that you need to type
signal SIGINT
before running the backtrace (bt), because gdb will catch the INT  
signal thus not leading to the desired crash and the backtrace just  
shows when you hit Ctrl-C, not what happens after. Only after the  
signal SIGINT you should get the crash (if at all).

Cheers,
Simon

On Tue, 2005-04-26 at 13:14 -0400, Simon Urbanek wrote:
> On 26.04.2005, at 12:52, Manuel.A.Morales@williams.edu wrote:
> 
> > Program received signal SIGINT, Interrupt.
> > 0x006547a2 in _dl_sysinfo_int80 () from /lib/ld-linux.so.2
> 
> Thanks you, however Marc omitted to mention that you need to type
> signal SIGINT
> before running the backtrace (bt), because gdb will catch the INT  
> signal thus not leading to the desired crash and the backtrace just  
> shows when you hit Ctrl-C, not what happens after. Only after the  
> signal SIGINT you should get the crash (if at all).
> 
> Cheers,
> Simon

Thanks Simon for picking that up.  My error.

Also, apologies for the double post, not sure how that happened.



Manuel, I should have asked earlier, but I presume that you installed
from source as I don't see RPMs for 2.1.0 yet?

Marc

On Tue, 26 Apr 2005, Peter Dalgaard wrote:
> MSchwartz@medanalytics.com writes:
>
>>> Thanks you, however Marc omitted to mention that you need to type
>>> signal SIGINT
>>> before running the backtrace (bt), because gdb will catch the INT
>>> signal thus not leading to the desired crash and the backtrace just
>>> shows when you hit Ctrl-C, not what happens after. Only after the
>>> signal SIGINT you should get the crash (if at all).
> ...
>> Manuel, I should have asked earlier, but I presume that you installed
>> from source as I don't see RPMs for 2.1.0 yet?
>>
>> Marc
>
> The effect seem to have been neatly backported to 2.0.1 though...
>
> Gdb doesn't seem to help. If you run a ps while "sudo R" is
running,
> you'll see something like this:
>
> root     30416  0.0  0.1  2356  252 pts/5    S+   19:48   0:00 sesh
/usr/bin/R
> root     30417 10.0  7.2 18016 13860 pts/5   S+   19:48   0:01
/usr/lib/R/bin/exec/R
>
> What I suspect is happening is that the ^C kills the sesh process, but
> that in turn does not manage to kill R.
Should it?  R traps that signal.

I'll now make the comment I thought when I first saw this:

 	Why is this a bug report against R not sudo?

-- 
Brian D. Ripley,                  ripley@stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

On Tue, 2005-04-26 at 19:59 +0200, Peter Dalgaard wrote:
> MSchwartz@medanalytics.com writes:
> 
> > > Thanks you, however Marc omitted to mention that you need to type
> > > signal SIGINT
> > > before running the backtrace (bt), because gdb will catch the INT
> > > signal thus not leading to the desired crash and the backtrace
just
> > > shows when you hit Ctrl-C, not what happens after. Only after the
> > > signal SIGINT you should get the crash (if at all).
> ...
> > Manuel, I should have asked earlier, but I presume that you installed
> > from source as I don't see RPMs for 2.1.0 yet?
> > 
> > Marc
> 
> The effect seem to have been neatly backported to 2.0.1 though...
> 
> Gdb doesn't seem to help. If you run a ps while "sudo R" is
running,
> you'll see something like this:
> 
> root     30416  0.0  0.1  2356  252 pts/5    S+   19:48   0:00 sesh
/usr/bin/R
> root     30417 10.0  7.2 18016 13860 pts/5   S+   19:48   0:01
/usr/lib/R/bin/exec/R
> 
> What I suspect is happening is that the ^C kills the sesh process, but
> that in turn does not manage to kill R.
So is this perhaps an SELinux issue? I don't get sesh running when I use
sudo R.

That would perhaps explain why I am not seeing it. Given the almost
daily changes in SELinux policies, I have foregone using it for the time
being. I was having all kinds of 'avc' related errors, so I gave up
until things stabilize more.

Marc

I get no crash after typing signal SIGINT - it just returns me to the R
prompt.

On Tue, 2005-04-26 at 13:14 -0400, Simon Urbanek wrote:
> On 26.04.2005, at 12:52, Manuel.A.Morales@williams.edu wrote:
> 
> > Program received signal SIGINT, Interrupt.
> > 0x006547a2 in _dl_sysinfo_int80 () from /lib/ld-linux.so.2
> 
> Thanks you, however Marc omitted to mention that you need to type
> signal SIGINT
> before running the backtrace (bt), because gdb will catch the INT  
> signal thus not leading to the desired crash and the backtrace just  
> shows when you hit Ctrl-C, not what happens after. Only after the  
> signal SIGINT you should get the crash (if at all).
> 
> Cheers,
> Simon

> On Tue, 2005-04-26 at 19:59 +0200, Peter Dalgaard wrote:
> > MSchwartz at medanalytics.com writes:
> > The effect seem to have been neatly backported to 2.0.1 though...
> > 
> > Gdb doesn't seem to help. If you run a ps while "sudo R"
is running,
> > you'll see something like this:
> > 
> > root     30416  0.0  0.1  2356  252 pts/5    S+   19:48   0:00 sesh
/usr/bin/R
> > root     30417 10.0  7.2 18016 13860 pts/5   S+   19:48   0:01
/usr/lib/R/bin/exec/R
> > 
> > What I suspect is happening is that the ^C kills the sesh process, but
> > that in turn does not manage to kill R.
> 
> So is this perhaps an SELinux issue? I don't get sesh running when I
use
> sudo R.
> 
> That would perhaps explain why I am not seeing it. Given the almost
> daily changes in SELinux policies, I have foregone using it for the time
> being. I was having all kinds of 'avc' related errors, so I gave up
> until things stabilize more.
Hi all,

Just a quick note that I seem to have confirmed this as a SELinux/sudo
interaction issue.

I re-booted my system this morning, with SELinux enabled using the
targeted policy, which is the default for FC3.

I was then able to confirm the behavior that Manuel initially reported
here.

As a result, I have filed a bug report with Fedora's Bugzilla against
sudo:

https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=156086

and we'll let the FC folks take it from there.

Needless to say, I am now back with SELinux disabled...

Presumably we can now close out the R bug report.

HTH,

Marc Schwartz