[Popcon-developers] Bug#310327: popularity-contest: broken pipeo

[Wouter Van Hemel]

reassign 310327 ssmtp

> On Mon, May 23, 2005 at 06:51:07AM +0200, Wouter Van Hemel wrote:
>> Package: popularity-contest
>> Version: 1.28
>> Severity: normal
>> Tags: sid
>>
>> I've just received an email from cron and popularity-contest seems to
>> break with:
>>
>> /etc/cron.weekly/popularity-contest:
>> tee: standard output: Broken pipe
>> tee: write error
>

I think I traced this back to ssmtp weirdness. For instance, this command
fails:

(echo "From: bob"; echo; HOME=/tmp su nobody -pc "sh -c
/usr/sbin/popularity-contest") | tee /tmp/delme | strace
/usr/sbin/sendmail -oi "my@email.address"; tail -1 /var/log/mail.log

... but this one works (notice one missing echo):

(echo; HOME=/tmp su nobody -pc "sh -c /usr/sbin/popularity-contest") | tee
/tmp/delme | strace /usr/sbin/sendmail -oi "my@email.address"; tail -1
/var/log/mail.log

The some problem occurs with a 'cat *.txt' as "a long, interleaved
command", so it is not related to popcon. In some circumstances, ssmtp
sends the email before the pipe has ended.

Without looking at the code, I dare to make a wild guess -- I think that
when ssmtp receives a timeout on stdin, it just thinks that the pipe ended
and sends the email without the body -- in this case the popcon output.

Note the last write which ends the email prematurely (command above):

[...]
write(3, "Received: by mysystem.mydomain.org (s"..., 91) = 91
write(3, "Date: Tue, 24 May 2005 05:34:53 "..., 39) = 39
write(3, "From: jim@example.com\r\n", 23) = 23
alarm(300)                              = 300
write(3, "\r\n", 2)                     = 2
fcntl64(0, F_SETFL, O_RDONLY|O_NONBLOCK) = 0
read(0, 0xb7fe8000, 4096)               = -1 EAGAIN (Resource temporarily
unavailable)
write(3, ".\r\n", 3)                    = 3
alarm(600)                              = 300
read(3, "2", 1)                         = 1
read(3, "5", 1)                         = 1
read(3, "0", 1)                         = 1
read(3, " ", 1)                         = 1
read(3, "o", 1)                         = 1
read(3, "k", 1)                         = 1
[...]

another one (cat *.txt):

[...]
write(3, "onze noord-westelijke samenlevin"..., 73) = 73
alarm(300)                              = 300
write(3, "wat normaal is in het echte leve"..., 72) = 72
alarm(300)                              = 300
write(3, "de natuur in gaat, maakt ze heil"..., 65) = 65
alarm(300)                              = 300
read(0, 0xb7fe8000, 4096)               = -1 EAGAIN (Resource temporarily
unavailable)
write(3, "Nietzsche, de haat van het leve\r"..., 33) = 33
alarm(300)                              = 300
read(0, 0xb7fe8000, 4096)               = -1 EAGAIN (Resource temporarily
unavailable)
write(3, ".\r\n", 3)                    = 3
alarm(600)                              = 300
read(3, "2", 1)                         = 1
read(3, "5", 1)                         = 1
read(3, "0", 1)                         = 1
read(3, " ", 1)                         = 1
read(3, "o", 1)                         = 1
read(3, "k", 1)                         = 1
[...]

If this indeed is the problem, upstream or ssmtp maintainer (perhaps they
are one and the same for ssmtp?) could add a check on the type of error
when reading from stdin, so EAGAIN just keeps on looping instead of ending
the pipe prematurely.

Thanks for your time,

  Wouter Van Hemel