poolporg / filter-senderscore Goto Github PK

if senderscore is not answering correctly or running into timeout, endless smtp sessions will be opened (at new incoming messages), but no mail processing is happening. when the filter is running on the connection level, then the messages are neither delivered nor queued, because it "hangs" on the filter processing.
Is there a possibility for a timeout? When senderscore request fails (maybe retry 2-3 times) and then pass the message (even if not scanned by the filter).
On large mx this will be a serious problem, when senderscore is not working, because after 1-2min the "SMTP client limit is reached" and the server isn't responding to SMTP anymore, and there will be a "message jam" after the filter is working again (or not).
Maybe senderscore is also rate-limiting, but that's not documented neither at this plugin's readme nor at senderscore itself.

Would you be interested in a simple test suite that runs filter-senderscore through a bunch of scenarios and validates the output? If so, any preferences when it comes to test frameworks? Would you prefer something that's TAP-compatible?

Hey there,
First of all, thanks for the time involved in all thoses projects that benefits to a lot of people.
Was wondering if it is possible to release a 0.12 version containing the allowedIP code ?
Currently running OpenSMTPd on Alpine but the filte-senderscore package is based on 0.11.
Also, it looks like Alpine package maintainer would not build a package on a "master" branch.

Cheers !

To the best of my knowledge, fmt.Printf is not guaranteed to be atomic and there is the (theoretical) possibility of intermixed output originating from the "delayed" goroutines. Two alternative solutions:

• Use a classical locking mechanism, such as a mutex, to make the output atomic.
• Create a separate routine with an infinite loop that reads from a Go channel and outputs the strings it recevies. In all the other routines, write to that channel instead of writing to stdout.

The second option is probably the "Go way" of implementing this. I'd be happy to implement either solution. Let me know what you think.

Description

Probably since this commit openbsd/src@570e619 the current build of filter-senderscore is segfaulting on OpenBSD -current:

www$ /usr/local/libexec/smtpd/filter-senderscore                                                                                                                 
Segmentation fault

Producing these dmesg entries:

www$ dmesg |grep filter\-sender
[filter-sendersco]14316/423618 pc=469206 inside 29c685000-29c692000: bogus syscall
[filter-sendersco]40021/251269 pc=469206 inside 225ed2000-225edf000: bogus syscall
[...]

Update: Forgot to attach versions:

www# pkg_info  opensmtpd-filter-senderscore 
Information for inst:opensmtpd-filter-senderscore-0.1.1

Comment:
SenderScore integration to the OpenSMTPD daemon

Description:
filter-senderscore is an opensmtpd filter that performs a DNS
lookup in the SenderScore registry to determine reputation of
an IP address and allow reputation-based blocking, junking or
delaying of sessions.

Maintainer: Gilles Chehade <[email protected]>

WWW: https://github.com/poolpOrg/filter-senderscore/


www# uname -a                               
OpenBSD  6.9 GENERIC#374 amd64

Solution

go1.16 does it's syscalls through libc[1], so just rebuilding with go1.16 solved this issue for me. If I'm not misunderstanding the ports Makefile, the binary is directly downloaded from here, so I hope this is the correct place for this issue.

[1] https://utcc.utoronto.ca/~cks/space/blog/programming/Go116OpenBSDUsesLibc