New in the FAQ: A misfeature introduced in recent versions of smail breaks fetchmail, but there is a workaround. See question T3.
Before reporting any bug, please read G3 for advice on how to include diagnostic information that will get your bug fixed as quickly as possible.
If you have a question or answer you think ought to be added to this FAQ list, mail it to fetchmail's maintainer, Eric S. Raymond, at esr@snark.thyrsus.com.
Rob Funk is fetchmail's designated backup maintainer. Other backup maintainers may be added in the future, in order to ensure continued support should Eric S. Raymond and/or Rob Funk drop permanently off the net for any reason.
There is a fetchmail-friends list for people who want to discuss fixes and improvements in fetchmail and help co-develop it. It's at fetchmail-friends@thyrsus.com and is a SmartList reflector; sign up in the usual way with a message containing the word "subscribe" in the subject line sent to to fetchmail-friends-request@thyrsus.com. (Similarly, "unsubscribe" in the Subject line unsubscribes you, and "help" returns general list help)
There is a fetchmail home page at http://www.ccil.org/~esr.
Fetchmail is not a toy or a coder's learning exercise, but an industrial-strength tool capable of transparently handling every retrieval demand from those of a simple single-user ISP connection up to mail retrieval and rerouting for an entire client domain. Fetchmail is easy to configure, unobtrusive in operation, powerful, feature-rich, and well documented. Extensive testing by a large, multi-platform user community has shown that it is as near bulletproof as the underlying protocols permit.
If you found this FAQ in the distribution, see the README for fetchmail's full feature list.
A text dump of this FAQ is included in the fetchmail distribution. Because it freezes at distribution release time, it may not be completely current.
A transcript of the failed session with -v on is almost always useful. If the bug involves a core dump or hang, a gdb stack trace is good to have. (Bear in mind that you can attach gdb to a running but hung process by giving the process ID as a second argument.) You will need to reconfigure with
Best of all is a mail file which, when fetched, will reproduce the bug under the latest (current) version.
Any bug I can reproduce will usually get fixed very quickly, often within 48 hours. Bugs I can't reproduce are a crapshoot. If the solution isn't obvious when I first look, it may evade me for a long time (or to put it another way, fetchmail is well enough tested that the easy bugs have long since been found). So if you want your bug fixed rapidly, it is not just sufficient but nearly necessary that you give me a way to reproduce it.
You can do spam filtering better with procmail or mailagent on the
server side and (if you're the server sysadmin) sendmail.cf domain
exclusions. You can do other policy things better with the
mda option and script wrappers around fetchmail. If
it's a prime-time-vs.-non-prime-time issue, ask yourself whether a
wrapper script called from crontab would do the job.
I'm not going to do these; fetchmail's job is transport, not policy, and I refuse to change it from doing one thing well to attempting many things badly. One of my objectives is to keep fetchmail simple so it stays reliable.
All that said, if you have a feature idea that really is about a transport problem that can't be handled anywhere but fetchmail, lay it on me. I'm very accommodating about good ideas.
The experiment was a success. I wrote a paper about it titled The Cathedral and the Bazaar which was first presented at Linux Kongress '97 in Bavaria and very well received there.
If you're reading a non-HTML dump of this FAQ, you can find the paper on the Web with a search for that title.
This error can be triggered by having a user option such as `keep'
or `fetchall' before the first explicit username. For
example, if you write
poll openmail protocol pop3 keep user "Hal DeVore" there is hdevore herethe `
keep' option will generate an entire user entry with the default
username (the name of fetchmail's invoking user).The popclient compatibility syntax was removed in 4.0. It complicated the configuration file grammar and confused users.
Also, the `interface', `monitor' and
`batchlimit' options changed after 2.8.
They used to be global options with `set' syntax like the batchlimit and logfile options. Now they're per-server options, like `protocol'.
If you had something like
set interface = "sl0/10.0.2.15"in your .fetchmailrc file, simply delete that line and insert `interface sl0/10.0.2.15' in the server options part of your `defaults' declaration.
Do similarly for any `monitor' or `batchlimit' options.
The configuration file parser treats any all-numeric token as a number, which will confuse it when it's expecting a name. String quoting forces the token's class.
You can work around this easily. Just put string quotes around your token.
I haven't fixed this because there is no good fix for it short of implementing a token pushback stack in the lexer. That's more additional complexity than I'm willing to add to banish a very marginal bug with an easy workaround.
(cd ~; mv ~/.poprc ~/.fetchmailrc)in order to migrate. Be aware that some of popclient's unnecessary options have been removed (see the NOTES file in the distribution for explanation). You can't deliver to a local mail file anymore or to standard output any more, and using an MDA for delivery is discouraged. If you throw those options away, fetchmail will now forward your mail into your system's normal Internet-mail delivery path.
Actually, using an MDA is now almost always the wrong thing; the MDA facility has been retained only for people who can't or won't run a sendmail-like SMTP listener on port 25. The default, SMTP forwarding to port 25, is better for at least two major reasons. One: it feeds retrieved POP and IMAP mail into your system's normal delivery path along with local mail and normal Internet mail, so all your normal filtering/aliasing/forwarding setup for local mail works. Two: because the port 25 listener returns a positive acknowledge, fetchmail can be sure you're not going to lose mail to a disk-full or some other resource-exhaustion problem.
If you used to use -mda "procmail -d
<you>" or something similar, forward to port
25 and do "| procmail -d <you>" in
your ~/.forward file.
As long as your new .fetchmailrc file does not use the removed
`localfolder' option or `limit' (which now takes a maximum byte size
rather than a line count), a straight move or copy of your .poprc will
often work. (The new run control file syntax also has to be a little
stricter about the order of options than the old, in order to support
multiple user descriptions per server; thus you may have to rearrange
things a bit.)
Run control files in the minimal .poprc format (without the `username'
token) will trigger a warning. To eliminate this warning, add the
`username' keyword before your first user entry per server (it is
already required before second and subsequent user entries per server.
In some future version the `username' keyword will be required.
On the machine where I'm the only real user, I run fetchmail as root from a cron job, like this:
fetchmail -u "itz" -p POP3 -s bolero.rahul.net
This used to work as is (with no .fetchmailrc file in root's home
directory) with the last version I had (1.7 or 1.8, I don't
remember). But with 2.0, it RECPs all mail to the local root user,
unless I create a .fetchmailrc in root's home directory containing:
skip bolero.rahul.net proto POP3
user itz is itz
It won't work if the second line is just "user itz". This is silly.It seems fetchmail decides to RECP the `default local user' (ie. the uid running fetchmail) unless there are local aliases, and the `default' aliases (itz->itz) don't count. They should.
Answer:
No they shouldn't. I thought about this for a while, and I don't much like the conclusion I reached, but it's unavoidable. The problem is that fetchmail has no way to know, in general, that a local user `itz' actually exists.
"Ah!" you say, "Why doesn't it check the password file to see if the remote name matches a local one?" Well, there are two reasons.
One: it's not always possible. Suppose you have an SMTP host declared that's not the machine fetchmail is running on? You lose.
Two: How do you know server itz and SMTP-host itz are the same person? They might not be, and fetchmail shouldn't assume they are unless local-itz can explicitly produce credentials to prove it (that is, the server-itz password in local-itz's .fetchmailrc file.).
Once you start running down possible failure modes and thinking about ways to tinker with the mapping rules, you'll quickly find that all the alternatives to the present default are worse or unacceptably more complicated or both.
Under bash, you can arrange this by putting `fetchmail -q' in the file `~/.bash_logout'. Most csh variants execute `~/.logout' on logout. For other shells, consult your shell manual page.
First, you may not need to use --interface at all. If your machine only ever does SLIP or PPP to one provider, it's almost certainly by a point to point modem connection to your provider's local subnet that's pretty secure against snooping (unless someone can tap your phone or the provider's local subnet!). Under these circumstances, specifying an interface address is fairly pointless.
What the option is really for is sites that use more than one provider. Under these circumstances, typically one of your provider IP addresses is your mailserver (reachable fairly securely via the modem and provider's subnet) but the others might ship your packets (including your password) over unknown portions of the general Internet that could be vulnerable to snooping. What you'll use --interface for is to make sure your password only goes over the one secure link.
To determine the device:
interface "sl0/205.164.136.0/255.255.255.0"would work. To range over any value of the last two octets (65536 addresses) you would use
interface "sl0/205.164.0.0/255.255.0.0"
First, a lightly edited version of a recipe from Masafumi NAKANE:
1. You must have ssh (the ssh client) on the local host and sshd (ssh server) on the remote mail server. And you have to configure ssh so you can login to the sshd server host without a password. (Refer to ssh man page for several authentication methods.)
2. Add something like following to your .fetchmailrc file:
poll mailhost port 1234 via localhost with pop3:
preconnect "ssh -f -L 1234:mailhost:110 mailhost sleep 20 /dev/null";
(Note that 1234 can be an arbitrary port number. Privileged ports can
be specified only by root.) The effect of this ssh command is to
forward connections made to localhost port 1234 (in above example) to
mailhost's 110.This configuration will enable secure mail transfer. All the conversation between fetchmail and remote pop server will be encrypted.
If sshd is not running on the remote mail server, you can specify intermediate host running it. If you do this, however, communication between the machine running sshd and the POP server will not be encrypted. And the preconnect line would be like this:
preconnect "ssh -f -L 1234:mailhost:110 sshdhost sleep 20 /dev/null"You can work this trick with IMAP too, but the port number 110 in the above would need to become 143.
Second, a recipe frm Charlie Brady <cbrady@ind.tansu.com.au>. Charlie says: "The [previous] recipe certainly works, but the solution I post here is better in a few respects":
command="socket localhost 110",no-port-forwarding 1024 ......where "
1024 ......" is the content of noddy's identity.pub file.
#! /bin/sh exec ssh -q -C -l your.login.id -e none mailhost socket localhost 110
1234 stream tcp nowait noddy /usr/sbin/tcpd /usr/local/bin/ssh.fm
Basically you need to use the "check_*" rules in sendmail. These are rules introduced since version 8.8.2
The idea is to generate a list of domains and addresses that are placed into a file - I call mine "sendmail.rej" and you place just one domain or email address on each line. During the SMTP transaction, this file is checked and if there is a match, the message is refused, with a suitable "Service not available" message sent back to the sender.
With the feature enabled in fetchmail, the mail is simply deleted, with no further processing.
The only drawback when blocking spam with fetchmail is that you do not get the satisfaction of sending an error back to the sender.
To actually use the check_mail rules in sendmail 8.8.2 or better, you need to know how to generate a sendmail.cf file from the m4 config files distributed with sendmail.
The actual rules can be found at the following URLS:
http://www.informatik.uni-kiel.de/%7Eca/email/check.html
This one is by Claus Aßman, who has documented more of sendmail then I can digest! The actual setup I used though was by David Begley, who has put together a WWW page describing how to quickly implement these rules yourself.
http://www.nepean.uws.edu.au/users/david/pe/blockmail.html
David's pages could be moving shortly. I will post an update if it happens.
Remember, when copying these rulesets off the web, that there are tabs embedded in them, that may not be preserved. You must reintroduce these tabs into the rules to make them work properly.
Once you have your ruleset in place, and have generated a nice sendmail.cf file, and the list of blocked sites, try telneting to your SMTP port to test it, and send a message with a blocked address in it.
You should see a message similar to:
"571 unsolicited email is refused"
Next, if you have access to a host that you can send mail from, that
is not your mail host, add that host to your spamlist and
restart sendmail.Send a message to your mailing address from that host and then pop off the message with fetchmail, using the -v argument. You can monitor the SMTP transaction, and when the FROM address is parsed, if sendmail sees that it is an address in spamlist, fetchmail will flush and delete it.
Under no circumstances put your mailhost or any host you accept mail from using fetchmail into your reject file. You will lose mail if you do this!!!
The check_ rules work, and they work well. Coupled with fetchmail's ability to respond to the appropriate error messages, you can be assured of never seeing a spam from any address you put in the reject list.
The only thing that is missing, as mentioned previously, is the ability to allow sendmail to process the message further and generate an error message to the sender.
forcecr option; qmail's listener mode doesn't like
header or message lines terminated with bare linefeeds.(This information is thanks to Robert de Bath <robert@mayday.cix.co.uk>.)
If a mailhost is using the qmail package (see http://pobox.com/~djb/qmail.html) then, providing the local hosts are also using qmail, it is possible to setup one fetchmail link to be reliably collect the mail for an entire domain.
One of the basic features of qmail is the `Delivered-To:' message header. Whenever qmail deliver a message to a local mailbox it puts the username and hostname of the envelope recipient on this line. The major reason for this is to prevent mail loops.
To set up qmail to batch mail for a disconnected site the isp-mailhost will have normally put that site in its `virtualhosts' control file so it will add a prefix to all mail addresses for this site. This results in mail sent to 'username@userhost.userdom.dom.com' having a 'Delivered-To:' line of the form:
Delivered-To: mbox-userstr-username@userhost.userdom.dom.com
A single host maildrop will be slightly simpler:
Delivered-To: mbox-userstr-username@userhost.dom.com
The ISP can make the 'mbox-userstr-' prefix anything they choose
but a string matching the user host name is likely.To use this line you must:
| ../bin/qmail-inject -a -f"$SENDER" "${LOCAL#mbox-userstr-}@$HOST}"
Note this does require a modern /bin/sh.
By default, the exim listener enforces the the RFC1123 requirement that MAIL FROM addresses you pass to it have to be canonical (e.g. with a fully qualified hostname part).
This is a potential problem if the MTAs upstream from your fetchmail
don't necessarily pass canonicalized From and Return-Path addresses,
and fetchmail's rewrite option is off. The specific case
where this has come up involves bounce messages generated by sendmail
on your mailer host, which have the (un-canonicalized) origin address
MAILER-DAEMON.
The right way to fix this is to enable the rewrite option and
have fetchmail canonicalize From and Return-Path addresses with the
mailserver hostname before exim sees them.
If you must run with rewrite off, there is a switch in exim's
configuration files that allows it to accept domainless MAIL FROM
addresses; you will have to flip it by putting the line
sender_unqualified_hosts = localhost
in the main section of the exim configuration file. Note that this
will result in such messages having an incorrect domain name attached
to their return address (your SMTP listener's hostname rather than
that of the remote mail server).
Smail 3.2 is very nearly plug-compatible with sendmail, and may work fine out of the box.
We have one report that when processing multiple messages from a single fetchmail session, smail sometimes delivers them in an order other than received-date order. This can be annoying because it scrambles conversational threads. This is not fetchmail's problem, it is an smail "feature" and has been reported to the maintainers as a bug.
Very recent smail versions require an -smtp_hello_verify
option in the smail config file. This overrides smail's check to see
that the HELO address is actually that of the client machine, which
is never going to be the case when fetchmail is in the picture.
According to RFC1123 an SMTP listener must allow this
mismatch, so smail's new behavior (introduced sometime between
3.2.0.90 and 3.2.0.95) is a bug.
The Lotus Notes SMTP gateway tries to deduce when it should convert \n to \r\n, but its rules are not intuitive. Use `forcecr'.
Or maybe you're trying to run fetchmail in multidrop mode as root without a .fetchmailrc file. This doesn't do what you think it should; see question C1.
Or you may not be connecting to the SMTP listener. Run fetchmail -v and see the next question.
The first thing to check is if you can telnet to port 25 on your smtp host (which is normally `localhost' unless you've specified an smtp option in your .fetchmailrc or on the command line) and get a greeting line from the listener. If the SMTP host is inaccessible or the listener is down, fix that first.
If the listener seems to be up when you test with telnet, the most benign and typical problem is that the listener had a momentary seizure due to resource exhaustion while fetchmail was polling it -- process table full or some other problem that stopped the listener process from forking. If your SMTP host is not `localhost' or something else in /etc/hosts, the fetchmail glitch could also have been caused by transient nameserver failure.
Try running fetchmail -v again; if it succeeds, you had one of these kinds of transient glitch. You can ignore these hiccups, because a future fetchmail run will get the mail through.
If the listener tests up, but you have chronic failures trying to connect to it anyway, your problem is more serious. One way to work around chronic SMTP connect problems is to use --mda. But this only attacks the symptom; you may have a DNS or TCP routing problem. You should really try to figure out what's going on underneath before it bites you some other way.
We have one report (from toby@eskimo.com) that you can sometimes solve
such problems by doing an smtp declaration with an IP
address that your routing table maps to something other than the
loopback device (he used ppp0).
We had another report from a Linux user of fetchmail 2.1 who solved his SMTP connection problem by removing the reference to -lresolv from his link line and relinking. Apparently in some recent Linux distributions the libc bind library version works better.
As of 2.2, the configure script has been hacked so the bind library is linked only if it is actually needed. So under Linux it won't be, and this particular cause should go away.
Try sending yourself test mail and retrieving it using the
command-line options `-k -m cat'. This will dump exactly what
fetchmail retrieves to standard output.
If the dump doesn't match what shows up in your mailbox when you configure an MDA, your MDA is mangling the message. If it doesn't match what you sent, then fetchmail or something on the server is broken.
Workaround: link with GNU malloc rather than the stock C library malloc.
We're told there is some problem with the malloc() code in that version which makes it fragile in the presence of multiple free() calls on the same pointer (the malloc arena gets corrupted). Unfortunately it appears from doing gdb traces that whatever free() calls producing the problem are being made by the C library itself, not the fetchmail code (they're all from within fclose, and not an fclose called by fetchmail, either).
However, POP3 has a design problem in that its servers mark a message `seen' as soon as the fetch command to get it is sent down. If for some reason the message isn't actually delivered (you take a line hit during the download, or your port 25 listener can't find enough free disk space, or you interrupt the delivery in mid-message) that `seen' message can lurk invisibly in your server mailbox forever.
Workaround: add the `fetchall' keyword to your POP3 fetch options.
Solution: switch to an IMAP server.
You can work around this by disabling optimization.
There may be an actual bug here that the optimizer exposes; the stack trace says the segfault is in free() and has all the earmarks of a heap- corruption screw. But the symptom doesn't reproduce under Linux with the same .fetchmailrc and .netrc.
Many POP servers, if an interruption occurs, will restore the whole mail queue after about 10 minutes. Others will restore it right away. If you have an interruption and don't see it right away, cross your fingers and wait ten minutes brfore retrying.
Some servers (such as Microsoft's NTMail) are mis-designed to restore
the entire queue, including messages you have deleted. If you have
one of these and it flakes out on you a lot, try setting a small
--fetchlimit value. This will result in more IP connects
to the server but will mean it actually executes changes to the queue
more often.
Qualcomm's qpopper, used at many BSD Unix sites, is better behaved. If its connection is dropped, it will first execute all DELE commands (as though you had issued an QUIT -- this is a technical violation of the RFCs, but a good idea in a world of flaky phone lines). Then it will re-queue any message that was being downloaded at hangup time. Still, qpopper may require a noticeable amount of time to do deletions and clean up its queue. (Fetchmail waits a bit before retrying in order to avoid a `lock busy' error.)
The easiest workaround is to write enough aka declarations to cover all of your mailserver's aliases, then say `no dns'. This will take DNS out of the picture (though it means mail may be uncollected if it's sent to an alias of the server that you don't have listed).
It would be better to fix your DNS, however. DNS problems can hurt you in lots of ways, for example by making your machines intermittently or permanently unreachable to the rest of the net.
In general, this is not really a good idea. It would be smarter to just let the mail sit in the mailserver's queue and use fetchmail's ETRN mode to trigger SMTP sends periodically (of course, this means you have to poll more frequently than the mailserver's expiry period). If you can't arrange this, try setting up a UUCP feed.
If neither of these alternatives is available, multidrop mode may do
(though you are going to get hurt by some mailing list
software; see the caveats under THE USE AND ABUSE OF MULTIDROP
MAILBOXES on the man page). If you want to try it, the way to do it
is with the `localdomains' option.
In general, if you use localdomains you need to make sure of two other things:
1. You've actually set up your .fetchmailrc entry to invoke multidrop mode.
Many people set a `localdomains' list and then forget
that fetchmail wants to see more than one name (or the wildcard `*')
in a `here' list before it will do multidrop routing.
2. You may have to set `no envelope'.
Normally, multidrop mode tries to deduce an envelope address from a message before parsing the To/Cc/Bcc lines (this enables it to avoid losing to mailing list software that doesn't put a recipient addess in the To lines).
Some ways of accumulating a whole domain's messages in a single server
mailbox mean it all ends up with a single envelope address that is
useless for rerouting purposes. You may have to set `no
envelope' to prevent fetchmail from being bamboozled by this.
If you use sendmail, you can check the list expansion with
sendmail -bv.
As of 2.2, the configure script has been hacked so the bind library is linked only if it is actually needed. So under Linux it won't be, and this problem should go away.
aka' option to pre-declare as many of your
mailserver's DNS names as you can. When an address's host part
matches an aka name, no DNS lookup needs to be done to check it.
If you're sure you've pre-declared all of your mailserver's DNS dames,
you can use the `no dns' option to prevent other hostname
parts from being looked up at all.
Sometimes delays are unavoidable. Some SMTP listeners try to call DNS on the From-address hostname as a way of checking that the address is valid.
Michael <michael@bizsystems.com> gave us a recipe for dealing with this case that pairs a run control file like this:
poll your.pop3.server proto pop3:
no envelope no dns
localdomains virtual.localdomain1.com virtual.localdomain2.com ...
user yourISPusername is root * here,
password yourISPpassword fetchall
with a hack on your local sendmail.cf like this:
############################################# # virtual info, local hack for ruleset 98 # ############################################# # domains to treat as direct mapped local domain CVvirtual.localdomain1.com virtual.localdomain2.com ... --------------------------- in ruleset 98 add ------------------------- # handle virtual users R$+ <@ $=V . > $: $1 < @ $j . > R< @ > $+ < @ $=V . > $: $1 < @ $j . > R< @ > $+ $: $1 R< error : $- $+ > $* $#error $@ $1 $: $2 R< $+ > $+ < @ $+ > $: $>97 $1This ruleset just strips virtual domain names off the addresses of incoming mail. Your sendmail must be 8.8 or newer for this to work. Michael says:
I use this scheme with 2 virtual domains and the default ISP user+domain and service about 30 mail accounts + majordomo on my inside pop3 server with fetchmail and sendmail 8.83
Some SMTP listeners get upset if you try to hand them a MAIL FROM address naming a different host than the originating site for your connection. This is a feature, not a bug -- it's supposed to help prevent people from forging mail with a bogus origin site. (RFC 1123 says you shouldn't do this exclusion...)
Since the originating site of a fetchmail delivery connection is localhost, this effectively means these picky listeners will barf on any MAIL FROM address fetchmail hands them with an @ in it!
In versions up to 1.9.9 this led to pesky errors at some sites. Because of this, I hacked 2.0 to just use the calling user ID as the MAIL FROM address.
Versions 2.1 and up try the header From address first and fall back to the calling-user ID. So if your SMTP listener isn't picky, the log will look right.
This is not fetchmail's problem. The first thing to try is installing
a current version of deliver. If this doesn't work, try to
figure out which other program in your mail path is inserting the
blank line and replace that. If you can't do either of these things,
pick a different MDA (such as procmail) and declare it with the
`mda' option.
The O'Reilly sendmail book does warn that IDA sendmail doesn't process X- headers correctly. If this is your problem, all I can suggest is replacing IDA sendmail, because it's broken and not RFC822 conformant.
Some POP daemons ignore Content-Length headers and split messages on From lines. We have one report that the 2.1 version of the BSD popper program (as distributed on Solaris 2.5 and elsewhere) is broken this way.
You can test this. Declare an mda of `cat' and send yourself one piece of mail containing "From" at start of a line. If you see a split message, your POP/IMAP server is at fault. Upgrade to a more recent version.
Sendmail and other SMTP listeners don't split RFC822 messages either. What's probably happening is either sendmail's local delivery agent or your mail reader are not quite RFC822-conformant and are breaking messages on what it thinks are Unix-style From headers. You can figure out which by looking at your client-side mailbox with vi or more. If the message is already split in your mailbox, your local delivery agent is the problem. If it's not, your mailreader is the problem.
If you can't replace the offending program, take a look at your sendmail.cf file. There will likely be a line something like
Mlocal, P=/usr/bin/procmail, F=lsDFMShP, S=10, R=20/40, A=procmail -Y -d $udescribing your local delivery agent. Try inserting the `E' option in the flags part (the F= string). This will make sendmail turn each dangerous start-of-line From into a >From, preventing programs further downstream from acting up.
biff nto turn it off. If this doesn't work, try the command
chmod -x `tty`which is essentially what
biff -n will do. If this
doesn't work, comment out any reference to ``comsat'' in your
/etc/inetd.conf file and restart inetd.In Slackware Linux distributions, the last line in /etc/profile is
biff yChange this to
biff nto solve the problem system-wide.
$Id: fetchmail-FAQ.html,v 1.49 1997/07/25 12:21:08 esr Exp $