Re: CGI-Output truncated - what does POLLHUP mean?

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Felix von Leitner wrote:
>>    if (pfd[0].revents&POLLHUP) break;
>> which aborts reading into the buffer.
[...]
> I have never experienced CGI truncation with fnord.  Can you please try
> it on a regular Linux and see if you can reproduce it there?
> 
> In general, we should also be able to treat POLLHUP like POLLIN, and
> then read returns an error.

Sorry for not replying earlier, I got distracted by tons of other weird
stuff...

I unfortunately have not been able to try to reproduce this with a
regular Linux, since the CGI in question is pretty specific to the
embedded hardware, I guess that there is a patch in Debian at least
hints that it has happened to someone else. I have now applied a
slightly modified version of the Debian Patch (which basically seems to
do what you suggested) and it seems to work OK.

There is another slight problem: there is no error checking on the
buffer_put() (which returns the error code of the underlying syscall
from the buffer implementation). Which is a problem when you're
streaming content (potentially forever) and the client stops reading the
stuff - there is a useless CGI process hanging around forever and
writing stuff to fnord, which happily wastes cycles writing it to an
invalid file descriptor...

We now do this at line 564:

  if (buffer_put(buffer_1,ibuf,n) < 0) {
    break;
  }

and it helps.

And in case someone else wants to implement CGI's streaming multimedia
content: do an "alarm(0)" as the first thing in your CGI if you wonder
why your CGI stops streaming after five minutes  :)

Bye,
       Simon


-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFH/UBQO2O/RXesiHARAjBBAJ0XakFdXWM399UzLRVL2CeHOHGgqgCfdBI2
kmTO/wU/zCcIAveoh9OMzQI=
=fULT
-----END PGP SIGNATURE-----