dccifd(8)

NAME

dccifd - Distributed Checksum Clearinghouse Program Inter

face

SYNOPSIS

dccifd  [-VdbxANQ]  [-G  on  |  off  | noIP | IPmask/xx] [-h
homedir] [-p /sock | host,port,rhost/bits] [-o /sock | host,port]
       [-D  local-domain]  [-r  rejection-msg]  [-m map] [-w
whiteclnt]    [-U    userdirs]     [-a     IGNORE|REJECT]     [-t
type,[log-thold,]rej-thold]
       [-g  [not-]type]  [-S header] [-l logdir] [-R rundir]
[-T tmpdir]
       [-j     maxjobs]      [-B      dnsbl-option]      [-L
ltype,facility.level]

DESCRIPTION

Dccifd is a daemon intended to connect spam filters such as

SpamAssasin and mail transfer agents (MTAs) other than sendmail

to DCC servers. The MTA or filter dccifd which in turn reports

related checksums to the nearest DCC server. DCCIFD then adds an

X-DCC SMTP header line to the message. The MTA is told to reject

the message if it is unsolicited bulk.

Dccifd is similar to the DCC sendmail milter interface, dc

cm(8) and the DCC Procmail interface, dccproc(8). dccifd is more

efficient than dccproc but not restricted to use with sendmail.

All three send reports of checksums related to mail received by

DCC clients and queries about the total number of reports of par

ticular checksums.

MTA programs generally use a simple ASCII protocol to send a

mail message including its SMTP envelope to the daemon. Dccifd

responds with an indication of whether the message is unsolicited

bulk and an optional copy of the message with an X-DCC header

added. The protocol is described below and in the

include/dccif.h file in the DCC source. There is a sample C in

terface routine in the dcclib/dccif.c file in the DCC source and

the dcclib.a library generated from the source. A Perl version

of the interface routine is in dccifd/dccif.pl. Test or demon

stration programs in the style of dccproc(8) that use those in

terface routines are in dccifd/dccif-test.

A subset of ESMTP can be used instead of the ASCII protocol

to connect dccifd to postfix as a "Before-Queue Content Filter."

See the -o flag.

Since the checksums of messages that are whitelisted locally

by the -w whiteclnt file are not reported to the DCC server,

dccifd knows nothing about the total recipient counts for their

checksums and so cannot add X-DCC header lines to such messages.

The list of servers that dccifd contacts is in a memory

mapped file shared by local DCC clients. The file is maintained

with cdcc(8). Turn on the daemon and put its parameters in the

dcc_conf. Start the daemon with the start-dccifd script.

OPTIONS

The following options are available:

-V displays the version of the DCC program interface.

-d enables debugging output from the DCC client library.

Additional -d options increase the number of messages. A single

-d causes aborted SMTP transactions to be logged.

-b causes the daemon to not detach itself from the con

trolling tty and put itself into the background.

-x causes the daemon to try "extra hard" to contact a DCC

server. Since it is usually more important to deliver mail than

to report its checksums, dccifd normally does not delay too long

while trying to contact a DCC server. It also will not try again

for several seconds after a failure. With -x, it will always try

to contact the DCC server and it will tell the MTA to answer the

DATA command with a 4yz temporary failure.

-A adds to existing X-DCC headers in the message instead

of replacing existing headers of the brand of the current server.

-N neither adds, deletes, nor replaces existing X-DCC

headers in the message. Each message is logged, rejected, and

otherwise handled the same.

-Q only queries the DCC server about the checksums of mes

sages instead of reporting and querying. This is useful when

dccifd is used to filter mail that has already been reported to a

DCC server by another DCC client. This can also be useful when

applying a private white or black list to mail that has already

been reported to a DCC server. No single mail message should be

reported to a DCC server more than once per recipient, because

each report will increase the apparent "bulkness" of the message.

-G on | off | noIP | IPmask/xx

controls greylisting. At least one working greylist

server must be listed in the map file in the DCC home directory.

If more than one is named, they must "flood" or change checksums

and they must use the same -G parameters. See dccd(8). Usually

all DCC client processes of dccm or dccifd should use the same -G

parameters.

IPmask/xx and noIP remove part or all of the IP address

from the greylist triple. The CIDR block size, xx, must be be

tween 1 and 128. 96 is added to block sizes smaller than 33 to

make them appropriate for the IPv6 addresses used by the DCC.

IPmask/96 differs from noIP because the former retains the IPv4

to IPv6 mapping prefix.

-W turns off DCC filtering by default to ease managing

systems where only a minority of users want unsolicited bulk mail

to be rejected. This is equivalent to a option dcc-off line in

the main -w whiteclnt file. When DCC filtering is off, the DCC

server is queried and the X-DCC header is added but the message

is marked to be delivered regardless of target counts and thresh

olds.

DCC filtering is enabled for a mailbox when -W is not

used and there is no option dcc-off line in the main or per-user

whiteclnt file or there is a option dcc-on pine in the per-user

whiteclnt file for the mailbox. DCC filtering can also be en

abled with an "OK2" entry for the fully qualified mailbox in the

main or per-user whiteclnt file.

Messages sent only to target addresses that are listed

in the global or relevant per-user -w whiteclnt file with "OK"

are not reported to the DCC server and so are not rejected and do

not receive X-DCC headers.

-h homedir

overrides the default DCC home directory, which is of

ten /var/dcc.

-p /sock/name | host,port,rhost/bits

overrides the default address at which programs contact

dccifd. The default is a UNIX domain socket named dccifd in the

DCC home directory.

The second form specifies a local host name or IP ad

dress, a local TCP port number, and the host names or IP address

es of computers that can use dccifd. 127.0.0.1 or localhost are

common choices for host. The string @ specifies IN_ADDRANY or

all local IP addresses. 127.0.0.0/8 is a common choice for

rhost/bits.

-o /sock | host,port

enables SMTP proxy mode instead of the ASCII protocol

and specifies the address of the SMTP server for which dccifd

acts as SMTP client. When /sock is /var/null, dccifd acts as if

there were downstream SMTP server that always answers "250 ok".

The string @ specifies the same IP address as the incoming TCP

connection.

See below concerning the subset of ESMTP used in this

mode.

-m map

specifies a name or path of the memory mapped parameter

file instead of the default map in the DCC home directory. It

should be created with the cdcc(8) command.

-w whiteclnt

specifies an optional file containing SMTP client IP

addresses, SMTP envelope values, and header values of mail that

is not spam, does not need a X-DCC header, and whose checksums

should not be reported to the DCC server. Local whitelist env_To

values are handy for whitelisting or exempting destination ad

dresses such as Postmaster from filtering and for blacklisting or

marking addresses that should never receive mail. Mail sent to

blacklisted addresses or with other blacklisted values such as

From or env_From values is reported to the DCC server as spam or

with target counts of millions.

If the pathname whiteclnt is not absolute, it is rela

tive to the DCC home directory. The format of the dccifd white

clnt file is the same as the whitelist files used by dbclean(8)

and the whiteclnt file used by dccproc(8). See dcc(8) for a de

scription of DCC white and blacklists. Because the contents of

the whiteclnt file are used frequently, a companion file is auto

matically created and maintained. It has the same pathname but

with an added suffix of .dccw and contains a memory mapped hash

table of the main file.

A local whitelist entry ("OK") or two or more semi

white listings ("OK2") for one of the message's checksums pre

vents all of the message's checksums from being reported to the

DCC server and the addition of a X-DCC header line by dccifd (ex

cept for env_To checksums or when -W is used). A local whitelist

entry for a checksum also prevents rejecting the message based on

DCC recipient counts as specified by -t. Otherwise, one or more

checksums with blacklisting entries ("MANY") cause all of the

message's checksums to be reported to the server with an ad

dressee count of "MANY".

If the message has a single recipient, an env_To local

whiteclnt entry of "OK" for the checksum of its recipient address

acts like any other whiteclnt entry of "OK." When the SMTP mes

sage has more than one recipient, the effects can be complicated.

When a message has several recipients with some but not all list

ed in the whiteclnt file, dccifd tries comply with the wishes of

the users who want filtering as well as those who don't by

silently not delivering the message to those who want filtering

(i.e. are not whitelisted) and delivering the message to don't

want filtering.

Consider the -W option for implicitly or by default

whitelisting env_to values.

-U userdirs

enables private whitelists and log files. Each target

of a message can have a directory of log files named

userdirs/addr/log where addr is the local user or mailbox name

computed by the MTA. The name of each user's log directory must

be log. If it is not absolute, userdirs is relative to the DCC

home directory. The sub-directory prefixes for -l logdir are not

honored. The directory containing the log files must be named

log and it must be writable by the dccifd process. Each log di

rectory must exist or logging for the corresponding is silently

disabled. The files created in the log directory are owned by

the UID of the dccifd process, but they have group and other read

and write permissions copied from the corresponding log directo

ry. To ensure the privacy of mail, it may be good to make the

directories readable only by owner and group, and to use a cron

script that changes the owner of each file to match the grandpar

ent addr directory.

There can also be a whitelist named

userdirs/addr/whiteclnt for each address addr. The name of the

file must be whiteclnt. Any checksum that is not white- or

blacklisted by an individual addressee's whitelist is checked in

the -w -whiteclnt list. A missing per-address whiteclnt file is

the same as an empty file. Relative paths for whitelists includ

ed in per-address files are resolved in the DCC home directory.

The whiteclnt files and the addr directories containing them must

be writable by the dccifd process.

-a IGNORE | REJECT

specifies the action taken when dccifd is in proxy mode

with -o and the DCC server counts or -t thresholds say that a

message is unsolicited bulk. IGNORE causes the message to be un

affected except for adding the X-DCC header line to the message.

This turns off DCC filtering.

Spam can also be REJECTed. The default is REJECT.

With an action of REJECT, spam sent to both white-list

ed targets and non-white-listed targets is delivered to white

listed targets and if possible, silently discarded for non-white

listed targets. This is not possible if there are too many non

white-listed targets to be saved in a buffer of about 500 bytes.

The effects of the -w whiteclnt are not affected by -a.

-t type,[log-thold,]rej-thold

sets logging and "spam" thresholds for checksum type.

The checksum types are IP, env_From, From, Message-ID, Received,

Body, Fuz1, and Fuz2. The string ALL sets thresholds for all

types, but is unlikely to be useful except for setting logging

thresholds. The string CMN specifies the commonly used checksums

Body, Fuz1, and Fuz2. Rej-thold and log-thold must be numbers,

the string NEVER, or the string MANY indicating millions of tar

gets. Counts from the DCC server as large as the threshold for

any single type are taken as sufficient evidence that the message

should be logged or rejected.

Log-thold is the threshold at which messages are

logged. It can be handy to log messages at a lower threshold to

find solicited bulk mail sources such as mailing lists. If no

logging threshold is set, only rejected mail and messages with

complicated combinations of white and blacklisting are logged.

Messages that reach at least one of their rejection thresholds

are logged regardless of logging thresholds.

Rej-thold is the threshold at which messages are con

sidered "bulk," and so should be rejected if not whitelisted.

The checksums of locally whitelisted messages are not

checked with the DCC server and so only the number of targets of

the current instance of a whitelisted message are compared

against the thresholds.

The default is -t ALL,NEVER, so that nothing is reject

ed or logged. A common choice is -t CMN,25,50 to reject mail

with common bodies except as overridden by the whitelist of the

DCC server and local -g, and -w.

-g [not-]type

indicates that whitelisted, OK or OK2, counts from the

DCC server for a type of checksum are to be believed. They

should be ignored if prefixed with not-. Type is one of the same

set of strings as for -t. Only IP, env_From, and From are likely

choices. By default all three are honored, and hence the need

for not-.

-S hdr

adds to the list of substitute or locally chosen head

ers that are checked with the -w whiteclnt file and sent to the

DCC server. The checksum of the last header of type hdr found in

the message is checked. Hdr can be HELO to specify the SMTP en

velope HELO value. Hdr can also be mail_host to specify the host

name from the Mail_from value in the SMTP envelope. As many as 6

different substitute headers can be specified, but only the

checksum of the first of the 6 will be sent to the DCC server.

-l logdir

specifies a directory in which files containing copies

of messages processed by dccifd are kept. All messages logged

are copied to the -l logdir directory. They can also be copied

to per-user directories specified with -U. Information about

other recipients of a message is deleted from the per-user

copies.

If logdir starts with D?, log files are put into subdi

rectories of the form logdir/JJJ where JJJ is the current julian

day. H?logdir puts logs files into subdirectories of the form

logdir/JJJ/HH where HH is the current hour. M?logdir puts log

files into subdirectories of the form logdir/JJJ/HH/MM where MM

is the current minute. See the FILES section below concerning

the contents of the files.

The directory is relative to the DCC home directory if

it is not absolute

-R rundir

specifies the "run" directory where the UNIX domain

socket and file containing the daemon's process ID are stored.

The default value is often /var/run/dcc.

-T tmpdir

changes the default directory for temporary files from

the default. The default is the directory specified with -l or

the system default if there -l is not used. The system default

is often /tmp.

-D local-domain

specifies a host name by which the system is known.

There can be several -D settings.

To find the per-user log directory and whitelist for

each mail recipient, dccifd must know each recipient's user name.

The default ASCII protocol includes an optional user name with

each recipient SMTP address. When that user name is absent or

when the subset of ESMTP enabled with -o is used, each mail ad

dress is checked against the list of -D local-domains. If there

is at least one match, the part of the recipient address remain

ing after matching the longest local-domain is taken as the user

name. The matching is anchored at the right or the end of the

recipient address. It must start at a period (.) or at-sign (@)

in the domain name part of the address.

-r rejection-msg

specifies the rejection message for unsolicited bulk

mail or for mail temporarily blocked by greylisting when -G is

specified. The first rejection-msg replaces the default bulk

mail rejection message, "5.7.1 550 mail %s from %s rejected by

DCC" The second replaces "4.2.1 452 mail %s from %s greylist tem

porary embargoed". There can be zero, one, or two "%s" strings.

The first is replaced an empty string and the second is replaced

by the IP address of the SMTP client.

A common alternate for the bulk mail rejection message

is "4.7.1 451 Access denied by DCC" to tell the sender to contin

ue trying. Use a 4yz response with caution, because it is likely

to delay for days a delivery failure message for false positives.

If the bulk mail rejection message does not start with a recog

nized error type and number, type 5.7.1 and 550 or 4.2.1 and 452

are used.

-j maxjobs

limits the number of simultaneous requests that will be

processed. The default value is the maximum number that seems to

be possible given the number of open files, select() bit masks,

and so forth that are available.

-B dnsbl-option

enables DNS blacklist checks of the SMTP client IP ad

dress, SMTP envelope Mail_From sender domain name, and of host

names in URLs in the message body. Body URL blacklisting has far

too many false positives to use on abuse mailboxes. It is less

effective than greylisting with dccm(8) or dccifd(8) but can be

useful in situations where greylisting cannot be used.

Dnsbl-option is either of the form set:option or of the

form domain[,IPaddr[,bltype]]. Domain is a DNS blacklist domain

such as example.com that will be searched. IPaddr is the IP ad

dress in the DNS blacklist that indicates that the mail message

is spam. 127.0.0.1 is assumed if IPaddr is absent. IPv6 ad

dresses can be specified with the usual colon (:) notation.

Names can be used instead of numeric addresses. The type of DNS

blacklist is specified by bltype as name, IPv4, or IPv6. Given

an envelope sender domain name or a domain name in a URL of

spam.domain.org and a blacklist of type name, spam.domain.org.ex

ample.com will be tried. Blacklist types of IPv4 and IPv6 re

quire that the domain name in a URL be resolved into an IPv4 or

IPv6 address. The address is then written as a reversed string

of decimal octets to check the DNS blacklist, as in 2.0.0.127.ex

ample.com,

More than one blacklist can be specified. They are

searched in order. All searching is stopped at the first posi

tive result. Positive results are ignored after being logged un

less an option DNSBL-on line appears in the global or per-user

whiteclnt file.

-B set:debug sends more messages about all DNS resolu

tions to the system log.

-B set:msg-secs=S limits dccifd to S seconds total for

checking all DNS blacklists. The default is 20.

-B set:URL-secs=S limits dccifd to at most S seconds

resolving and checking any single URL. The default is 5. Some

spam contains dozens of URLs and that some "spamvertised" URLs

contain host names that need minutes to resolve. Busy mail sys

tems cannot afford to spend minutes checking each incoming mail

message. In order to use typical single-threaded DNS resolver

libraries, dccm(8) and dccifd(8) use fleets of helper processes.

-B set:no-envelope says that SMTP client IP addresses

and sender Mail_From domain names should not be checked in the

following blacklists. -B set:envelope restores the default for

subsequently named blacklists.

-B set:no-body says that URLs in the message body

should not be checked in the in the following blacklists. -B

set:body restores the default for later blacklists.

-B set:no-MX says MX servers of sender Mail_From domain

names and host names in URLs should not be checked in the follow

ing blacklists. -B set:MX restores the default.

-L ltype,facility.level

specifies how messages should be logged. Ltype must be

error or info to indicate which of the two types of messages are

being controlled. Level must be a syslog(3) level among EMERG,

ALERT, CRIT, ERR, WARNING, NOTICE, INFO, and DEBUG. Facility

must be among AUTH, AUTHPRIV, CRON, DAEMON, FTP, KERN, LPR, MAIL,

NEWS, USER, UUCP, and LOCAL0 through LOCAL7. The default is

equivalent to

-L info,MAIL.NOTICE -L error,MAIL.ERR

dccifd normally sends counts of mail rejected and so forth

the system log at midnight. The SIGUSR1 signal sends an immedi

ate report to the system log. The reports will be repeated every

24 hours at the same minute as the signal instead of at midnight.

Protocol

Dccifd uses a simple ASCII protocol to receive mail messages

to be checked and to return results. For each message, the MTA

must open a connection to the interface daemon, send options, en

velope recipients, and the message, receive the results, and

close the connection.

Instead of the ASCII protocol, a subset of ESMTP is enabled

by -o. Only the familiar HELO, EHLO, Mail, Rcpt, DATA, RSET, and

QUIT commands and the Postfix extensions XFORWARD and XCLIENT are

honored. Since SMTP has no provisions for user names, the proto

col enabled by -o depends on a list of local domain names speci

fied with -D to find per-user log directories and whitelist

files. If neither XFORWARD nor XCLIENT are used, dccifd uses the

IP address of the MTA and the value of the HELO command.

In the ASCII protocol, each of the following lines are sent

in order to dccifd. Each ends with a newline ('0) character.

options zero or more blank-separated strings among:

spam the message is already known to

be spam
body return all of the headers with

the added X-DCC header line and the body
header return the X-DCC header query ask the DCC server about the

message without reporting it as if dccifd were running with -Q. grey-query only query the greylist server

for this message. -G on must be in use.
no-reject suppress the overall, one char

acter line 'R' result. This can be useful when using dccifd only

for greylisting. ')

after the IP address. The client IP address must be present and

non-null if the host name is present. If the client IP address

is absent, then the IP address and host name are taken from the

first Received header if it has the standard "name (name [IP ad

dress])..." format.

client IP address of the SMTP client in a "dotted" or

"coloned" ASCII string and reverse-DNS host name. If the host

name is present, it must follow a carriage return character ('
HELO SMTP HELO value or nothing, followed by a new

line character.
sender or SMTP Mail From command value

'). A local user name can be null if it is not known. Recipi

ents that lack local user names will lack per-user log files and

will not invoke a per-user whitelist.
recipients or SMTP Rcpt To values followed by correspond

ing local user names, one pair to a line. Each optional local

user name is separated from the corresponding recipient address

by a carriage return ('

The last recipient-user name pair is followed by an empty

line and the headers and body of the message. The end of the

body of the mail message is signaled by the MTA half-closing the

connection. See shutdown(2).

Dccifd responds with three things. First is a one character

line of the overall result advising the MTA to

A accept the message for all recipients and answer the

SMTP DATA command with a 2yz result.
G answer with a 4yz result to embargo the message for

greylisting.
R reject the message and answer the DATA command with a

5yz result.
S accept the message for some recipients and so answer

the DATA command with a 2yz result.
T temporary failure by the DCC system and so answer

with a 4yz result.

Second is a line of 'A', 'G', and 'R' characters indicating

that the message should be accepted and delivered or discarded

for each corresponding recipient. Limitations in the SMTP proto

col allows only a single result for the DATA command for all re

cipients that were not rejected before body of the message was

offered with the DATA command. To accept the message for some

recipients and reject it for others, the MTA must tell the SMTP

client it is accepting the message for all recipients and then

discard it for those that would reject it.

Finally, if the body or header strings are in the first line

of options sent by the MTA to the daemon, then the X-DCC header

line or the entire body with the X-DCC header line follows.

FILES

/var/dcc is the DCC home directory in which other files

are found.
libexec/start-dccifd

is a script often used to the daemon.

dcc/dcc_conf

contains parameters used by the scripts to start

DCC daemons and cron jobs.

logdir is an optional directory specified with -l and

containing marked mail. Each file in the directory contains one

message, at least one of whose checksums reached its -t thresh

olds or that is interesting for some other reason. Each file

starts with lines containing the date when the message was re

ceived, the IP address of the SMTP client, and SMTP envelope val

ues. Those lines are followed by the body of the SMTP message

including its header as it was received. Only approximately the

first 32 KBytes of the body are recorded unless modified by

./configure --with-max-log-size=xx The checksums for the message

follow the body. They are followed by lines indicate that one of

the checksums is white- or blacklisted by the -w whiteclnt file.

Each file ends with the X-DCC header line added to the message

and the disposition of the message.
map is the memory mapped file of information con

cerning DCC servers in the DCC home directory.
whiteclnt contains the client whitelist in the format de

scribed in dcc(8).
whiteclnt.dccw

is a memory mapped hash table of the whiteclnt

file.

dccifd.pid in the -R rundir directory contains daemon's

process ID.

SEE ALSO

cdcc(8), dbclean(8), dcc(8), dccd(8), dblist(8), dccm(8),

dccproc(8), dccsight(8),

HISTORY

Implementation of dccifd was started at Rhyolite Software in

2002. This describes version 1.2.74.

BUGS

dccifd uses -t where dccproc(8) uses -c.

Systems without setrlimit(2) and getrlimit(2) can have prob

lems with the default limit on the number of simultaneous jobs,

the value of -j. Every job requires four open files. These

problems are usually seen with errors messages that say something

like

dccifd[24448]: DCC: accept() returned invalid socket

A fix is to use a smaller value for -j or to allow dccifd to

open more files.

BSD December 8, 2007

BSD