On Sat, Oct 08, 2011 at 02:45:27PM +0200, Axel Beckert wrote:
> Hi,
>
> I'm currently preparing an upload of the current screen HEAD to Debian
> (either Unstable or Experimental, not yet decided) and I noticed that,
> if I have a screen 4.0.3 (as currently in Debian Stable/Unstable) is
> running, I can't reattach to that with the new screen 4.1.0 snapshot,
> it just hangs until I kill it with "kill -TERM" as Ctrl-C does not
> help.
>
> As for me the common way to dist-upgrade Debian or Ubuntu boxes
> (especially remote servers) is to run the whole process inside a
> screen, this is a quite critical issue. So I wonder:
>
> Is this issue known? Not circumventable? An unexpected bug? Anyone has
> an idea where this comes from? Or does it not happen at all with vanilla
> screen versions?
I noticed the message while upgrading screen version sid → experimental,
thanks for making it loud.
> I tried to strace "screen -r" to find any differences, but
> interestingly it behaves differently when being traced (bails out with
> claiming /var/run/screen has wrong permissions or so) than when not.
This is because screen is setgid utmp:
% ls -l =screen
-rwxr-sr-x 1 root utmp 414432 Oct 9 05:02 /usr/bin/screen*
strace() will defuse setuid/setgid so it won't work as expected, but we can
always try as root:
# aptitude install screen=4.0.3-14
# screen -dmS test
# strace -o /tmp/log_4.0.3 screen -r test
[... works, I detach ...]
# aptitude install screen=4.1.0~20110819git450e8f3-1
# strace -o /tmp/log_4.1.0 screen -r test
[... hangs, gets killed ...]
The difference I see at the end is:
4.0.3 → write(4, "\0gsm\2\0\0\0/dev/pts/46\0\0\0\0\0\0\0\0\0\0\0\0\0"...,
12336) = 12336
4.1.0 → write(4, "\2gsm\2\0\0\0/dev/pts/46\0\0\0\0\0\0\0\0\0\0\0\0\0"...,
12376) = 12376
The first byte was \0, now it is \2.
Is tracing the background process perhaps more interesting?
# aptitude install screen=4.0.3-14
# strace -t -o /tmp/server_log_4.0.3 -p 26322
[... works, I detach, then ^C this strace ...]
# aptitude install screen=4.1.0~20110819git450e8f3-1
# strace -t -o /tmp/server_log_4.1.0 -p 26322
[... hangs, I kill client, then ^C this strace ...]
Differences here:
4.0.3:
0.000083 read(3, "\0gsm\2\0\0\0/dev/pts/46\0\0\0\0\0\0\0\0\0\0\0\0\0"...,
12336) = 12336
0.000125 close(3) = 0
0.000075 geteuid() = 0
0.000067 getegid() = 43
0.000067 getgid() = 0
0.000068 setregid(43, 0) = 0
0.000077 open("/var/run/screen/S-root/26322.test", O_RDONLY|O_NONBLOCK) = 3
0.000106 geteuid() = 0
0.000068 getegid() = 0
0.000067 getgid() = 43
0.000069 setregid(0, 43) = 0
0.000074 kill(26948, SIG_0) = 0
0.000074 geteuid() = 0
0.000067 getegid() = 43
0.000067 getgid() = 0
0.000068 setregid(43, 0) = 0
0.000072 open("/dev/pts/46", O_RDWR|O_NONBLOCK) = 5
[... goes on and on ...]
4.1.0:
0.000087 read(3, "\2gsm\2\0\0\0/dev/pts/46\0\0\0\0\0\0\0\0\0\0\0\0\0"...,
12336) = 12336
0.000123 close(3) = 0
0.000079 geteuid() = 0
0.000073 getegid() = 43
0.000072 getgid() = 0
0.000074 setregid(43, 0) = 0
0.000081 open("/var/run/screen/S-root/26322.test", O_RDONLY|O_NONBLOCK) = 3
0.000112 geteuid() = 0
0.000073 getegid() = 0
0.000072 getgid() = 43
0.000074 setregid(0, 43) = 0
0.000089 select(1024, [3 4], [], NULL, NULL) = ? ERESTARTNOHAND (To be
restarted)
5.869203 --- SIGINT (Interrupt) @ 0 (0) ---
So yeah, the background screen is waiting for more traffic from the new
version, which it doesn't get.
The screen.git repo has no tags. The experimental package has a commit SHA1
in its version string, which helps. How do I find out exactly which commit
corresponds to "Screen version 4.00.03jw4 (FAU) 2-May-06" ? I'd like to
bisect the issue, and knowing where to start would be a good start :)
--
Fernando Vezzosi perl -E'$_="Pop Corn",s/\b\S(?{$0=$&^_^L})/$0/g,say'
qw(MDAx MTAw MDEw MDEx MDAw _5 MTEw _6 _1 _5 _5 _4 _2 _2 _6
MTEx _1 _5 _5 _4 _5 _2 _6 _1 _1 _2 _2 _3 _5 _5 _6 _1 _1 _2
_5 _4 _3 _5 _2 _5 _1 _2 _3 _4 _5 MA==)
