Bug#929860: apt-get hangs forever when invoked via timeout

[Martin Olsson <martin.olsson+debian@sentorsecurity.com>]

Package: apt
Version: 1.4.9

Hi!
I've found a reproduceable bug when using 'apt-get' together with
'timeout' within a shell.

How to reproduce:

I install a system using the debian-9.9.0-amd64-xfce-CD-1.iso.
I don't install anything at all except a ssh-server (not even any
sys-utils, just the base OS and sshd).
I login and create this script /root/foo.sh:   ('bc' can be replaced
with any package)

  #!/bin/sh
  timeout 1m apt-get -y -qq install bc

I run 'chmod 755 /root/foo.sh' and then execute it:
# /root/foo.sh
Selecting previously unselected package bc.
(Reading database ... 18047 files and directories currently installed.)
Preparing to unpack .../bc_1.06.95-9+b3_amd64.deb ...
Unpacking bc (1.06.95-9+b3) ...
Processing triggers for man-db (2.7.6.1-2) ...
Setting up bc (1.06.95-9+b3) ...

Here it freezes!
Nothing more happens, and it would hang forever if not my timeout
expired after a minute, making foo.sh continue.
foo.sh has nothing more to do so it terminates and I get a new prompt.

But for some reason, the tty is no longer echoing my keystrokes!
I blindly run 'echo $?' and get the response "124".
This is the return code from 'timeout' after it killed the apt-get job.


1)
What is happening here? The package is successfully installed but
apt-get never terminates.

2)
Running 'timeout 1m apt-get -y -qq install bc' directly on the
commandline works fine.
Running 'apt-get -y -qq install bc' directly in the foo.sh shell,
without any timeout, works fine.
So why doesn't apt-get terminate when started via 'timeout', when run
inside a shell?

3)
Why is the tty no longer echoing my keystrokes?

4)
I'm pretty sure my install-script in Debian 9.6 worked just fine with
'apt-get' together with 'timeout' within a shell.
That is, 'apt-get' did its job and terminated immediately and sucessfully.
Now when I install a new machine using Debian 9.9 it didn't work as expected.
So I guess the problem was introduced somewhere between Debian 9.6 and 9.9.



Anyhow... I can now reproduce the bug over and over:
I run 'reset' to get a normal tty again, purge the installed 'bc' tool
and then install it again via foo.sh:

# reset
# apt-get -y purge bc
# /root/foo.sh

Again, apt-get hangs after installing 'bc'.



(
Why do I call this a bug and not just an annoyance?
Here's my use-case (that used to work fine):

I have a script that declare a function to add a new APT source,
update and then install a package using that source:

  install_foobar() {
    cd /tmp
    timeout 1m wget "https://apt.foobar.com/foobar.deb"; || return $?
    dpkg -i foobar.deb || return $?
    timeout 1m apt-get update || return $?
    timeout 4m apt-get -y install foobar-agent || return $?
    return 0
  }

The function is called by the script, and depending on its exit status
it does things differently:

  install_foobar || touch /root/foobar_install_failed
  if [ ! -f /root/foobar_install_failed ]; then
    ...perform cleanup...
    ...disable unnecessary services...
    ...create a foobar user...
    ...set permissions...
  fi

Now when the apt-get command timeout instead of exiting normally, code
124 is returned.
This in turn creates the 'foobar_install_failed' file, and the user
and settings, etc are not processed even though the foobar-agent
package was successfully installed.
)