dovecot - Mar 2018 - lmtp service timeouting even after receiving full message

On 23.03.2018 10:04, Arkadiusz Mi?kiewicz wrote:
> On Thursday 22 of March 2018, Arkadiusz Mi?kiewicz wrote:
>> I have a problem with some messages passed from exim to dovecot lmtp
>> service:
>>
>> From exim debug:
>>
>> using socket /var/run/dovecot/lmtp
>>   LMTP<< 220 mbox8 ready
>>   LMTP>> LHLO mbox8...
>>   LMTP<< 250-mbox8
>>   LMTP<< 250-STARTTLS
>>   LMTP<< 250-8BITMIME
>>   LMTP<< 250-ENHANCEDSTATUSCODES
>>   LMTP<< 250 PIPELINING
>>   LMTP>> MAIL FROM:...
>>   LMTP<< 250 2.1.0 OK
>>   LMTP>> RCPT TO:...
>>   LMTP<< 250 2.1.5 OK
>>   LMTP>> DATA
>>   LMTP<< 354 OK
>>   LMTP>> writing message and terminating "."
>> cannot use sendfile for body: terminating dot wanted
>> writing data block fd=6 size=6585 timeout=300
>>   LMTP>> QUIT
>>   LMTP<< 421 4.4.2 mbox8 Disconnected client for inactivity
>>
>>
>> now above "writting message and terminating" according to
exim strace looks
>> like this:
>>
>> 13413 22:07:24 write(2, "  LMTP>> writing message and
terminating \".\"\n",
>> 45) = 45 <0.000018> 13413 22:07:24
stat("/etc/localtime",
>> {st_mode=S_IFREG|0644, st_size=2705, ...}) = 0 <0.000025> 13413
22:07:24
>> write(2, "cannot use sendfile for body: terminating dot
wanted\n", 53) >> 53 <0.000022> 13413 22:07:24 lseek(3, 19,
SEEK_SET)   = 19 <0.000015>
>> [...]
>> 13413 22:07:24 write(2, "writing data block fd=6 size=6585
timeout=300\n",
>> 46) = 46 <0.000019> 13413 22:07:24 alarm(300)               = 0
<0.000013>
>> 13413 22:07:24 write(6, "Return-path:
>> [...]
>> \nwrap\"><br></span></div>.\n", 6585) =
6585 <0.000023>
>>
>> and then exim waits for dovecot lmtp to say it accepted message but
lmtp
>> never does that.
>>
>> Isn't .\n always enough for dovecot to signal end of message?
Because exim
>> wrote full message to dovecot lmtp unix socket.
>>
>> Now the interesting this - exim does only one write() with 6585 bytes
to
>> write that message to lmtp. If exim needs more than one write() to do
that
>> (message is larger by some buffer) then it succeeds, dovecot accepts
it.
>>
>> Could dovecot lmtp handling for single write() case be broken?
>>
>> dovecot 2.2.35
> And reproducer. Both messages have ".\n" as ending but one has
\r\n before it.
>
> #!/usr/bin/python3
>   
> import socket
>
> c = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
> c.connect("/var/run/dovecot/lmtp")
> print(c.recv(1024))
> print("LHLO")
> c.send(b"LHLO mbox8\r\n")
> print(c.recv(1024))
> print("MAIL FROM")
> c.send(b"MAIL FROM:<someone at example.com>\r\n")
> print(c.recv(1024))
> print("RCPT TO")
> c.send(b"RCPT TO:<arekm at example.com>\r\n")
> print(c.recv(1024))
> print("DATA")
> c.send(b"DATA\r\n")
> print(c.recv(1024))
> print("body...")
>
> msg_fail = b"testtest<br></span></div>.\n"
> msg_ok = b"01, 41.82, 30.14\r\n.\n"
>
> #c.send(msg_ok)
> c.send(msg_fail)
>
> print(c.recv(1024))
>
> c.close()
>
Shouldn't the last dot be own it's own line?

Aki

On Friday 23 of March 2018, Aki Tuomi wrote:
> 
> Shouldn't the last dot be own it's own line?
Oh, right. So exim fault (spool wireformat feature breaks this).
> Aki
Thanks,
-- 
Arkadiusz Mi?kiewicz, arekm / ( maven.pl | pld-linux.org )

On Friday 23 of March 2018, Arkadiusz Mi?kiewicz wrote:
> On Friday 23 of March 2018, Aki Tuomi wrote:
> > Shouldn't the last dot be own it's own line?
> 
> Oh, right. So exim fault (spool wireformat feature breaks this).
And just got fixed:

https://github.com/Exim/exim/commit/f64e8b5f3b01b5285bb1f9172da7950e7f000c22
> 
> > Aki
> 
> Thanks,

-- 
Arkadiusz Mi?kiewicz, arekm / ( maven.pl | pld-linux.org )