lmtp service timeouting even after receiving full message
Arkadiusz Miśkiewicz
arekm at maven.pl
Thu Mar 22 23:33:16 EET 2018
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
--
Arkadiusz Miśkiewicz, arekm / ( maven.pl | pld-linux.org )
More information about the dovecot
mailing list