NAME

dccd — Distributed Checksum Clearinghouse Daemon

SYNOPSIS

dccd [-dVbfFQ] -i server-ID [-n brand] [-h homedir] -I [host-ID][,user]
[-a [server-addr][,server-port]] [-q qsize]
[-G [on,][weak-body,][weak-IP,][embargo][,window][,white]]
[-W [rate][,chg][,dbsize]] [-K [no-]type] [-T [no-]tracemode] [-u anon-delay[,inflate]] [-C dbclean] [-L ltype,facility.level]
[-R [RL_SUB],[RL_ANON],[RL_ALL_ANON],[RL_BUGS]]

DESCRIPTION

Dccd receives reports of checksums related to mail received by DCC clients and queries about the total number of reports of particular checksums. A DCC server never receives mail, address, headers, or other information from clients, but only cryptographically secure checksums of such information. A DCC server cannot determine the text or other information that corresponds to the checksums it receives. It only acts as a clearinghouse of total counts of checksums computed by clients.

Each DCC server is identified by a numeric server-ID. Each DCC client is identified by a client-ID, either explicitly listed in the /var/lib/dcc/ids file or the special anonymous client-ID. Many computers are expected to share a single client-ID. A server-ID is between 100 and 32768 while a client-ID is between 32768 and 16777215. DCC server-IDs need be known only to DCC servers and the people running them. The passwords associated with DCC server-IDs should be protected, because DCC servers listen to commands authenticated with server-IDs and their associated passwords. Each client that does not use the anonymous ID must know the client-ID and password used by each of its servers. A single client computer can use different passwords with different server computers. See the /var/lib/dcc/ids file.

A /var/lib/dcc/whitelist of known good (or bad) sources of email prevents legitimate mailing lists from being seen as unsolicited bulk email by DCC clients. The whitelist used by a DCC server is built into the database when old entries are removed by dbclean(8). Each DCC client has its own, local whitelist, and in general, whitelists work better in DCC clients than servers.

A dccd /var/lib/dcc/whitelist file containing IP addresses that should be in client whiteclnt files is useful. When -T WLIST tracing is enabled (as it is by default), dccd complains to the system log when an authenticated client reports mail from IP addresses listed as OK, MX, or MXDCC. It is often useful to have a /var/lib/dcc/whitecommon file containing whitelisted IP addresses.

The effectiveness of a Distributed Checksum Clearinghouse increases as the number of subscribers increases. Flooding reports of checksums among DCC servers increases the effective number of subscribers to each server. Each dccd daemon tries to maintain TCP/IP connections to the other servers listed in the /var/lib/dcc/flod file, and send them reports containing checksums with total counts exceeding thresholds. Changes in the flod file are noticed automatically within minutes.

Controls on report flooding are specified in the flod file. Each line specifies a hostname and port number to which reports should be flooded, a server-ID to identify and authenticate the output stream, a server-ID to identify and authenticate an input stream from the same server, and flags with each ID. The ability to delete reports of checksums is handy, but could be abused. If del is not present among the in-opts options for the incoming ID, incoming delete requests are logged and then ignored. Floods from DCC "brands" that count only mail to spam traps and whose servers use the -Q option to count extremely bulk mail should be marked with traps. They can be seen as counting millions of targets, so the traps flag on their /var/lib/dcc/flod file entry changes their incoming flooded reports counts to many.

Dccd automatically checks its /var/lib/dcc/flod and /var/lib/dcc/ids files periodically. Cdcc(8) has the commands new ids and flood check to tell dccd to check those two files immediately. Both files are also checked for changes after the SIGHUP signal.

OPTIONS

The following options are available. Most of them should set by changing the /var/lib/dcc/dcc_conf control file.

-d

enables debugging output. Additional -d options increase the number of messages.

-V

displays the version of the DCC server daemon. Two or more -V options show the options with which it was built.

-b

causes the server to not detach itself from the controlling tty or put itself into the background.

-F

uses write() instead of mmap() and msync() in some cases to modify the DCC database. It is the default on Solaris except when the database is in a memory mapped file system. See -f.

-f

uses mmap() and msync() to modify the DCC database. See -F.

-Q

causes the server to treat reports of checksums as queries except from DCC clients marked trusted in the /var/lib/dcc/ids file with rpt-ok. See -u to turn off access by anonymous or unauthenticated clients.

-i server-ID

specifies the ID of this DCC server. Each server identifies itself as responsible for checksums that it forwards to other servers.

-n brand

is an arbitrary string of letters and numbers that identifies the organization running the DCC server. The brand is required, and appears in the SMTP X-DCC headers generated by the DCC.

-h homedir

overrides the default DCC home directory, /var/lib/dcc.

-I [host-ID][,user]

sets the UID and GID of the process or the server's name for assertions of its -i server-ID flooded to peers. The default name is the first 16 characters of the host name. If present, user must be a valid user name.

-a [server-addr][,server-port]

adds an hostname or IP address to the list of local IP addresses that the server answers. Multiple -a options can be used to specify a subset of the available network interfaces or to use more than one port number. The default without any -a options is to listen on all local IP addresses. It can be useful to list some of the IP addresses of multi-homed hosts to deal with firewalls. By default server-port is 6277 for DCC servers and 6276 for greylist servers. It is the UDP port at which DCC requests are received and the TCP port for incoming floods of reports.

If server-addr is absent and if the getifaddrs(8) function is supported, separate UDP sockets are bound to each configured network interface so that each DCC clients receives replies from the IP addresses to which corresponding request are sent. If dccd is started before all network interfaces are turned on or there are interfaces that are turned on and off or change their addresses such as PPP interfaces, then the special string @ should be used to tell dccd to bind to an INADDR_ANY UDP socket.

Outgoing TCP connections to flood checksum reports to other DCC servers used the IP address of a single -a option, but only if there is single option that is not localhost. See also the /var/lib/dcc/flod file.

-q qsize

specifies the maximum size of the queue of requests from anonymous or unauthenticated clients. The default value is the maximum DCC RTT in seconds times 200 or 1000.

-G [on,][weak-body,][weak-IP,][embargo][,window][,white]

changes dccd to a greylist server for dccm(8) or dccifd(8). Greylisting consists of temporarily rejecting or embargoing mail from unfamiliar combinations of SMTP client IP address, SMTP envelope sender, and SMTP envelope recipient. If the SMTP client persists for embargo seconds and so is probably not an open proxy, worm-infected personal computer, or other transient source of spam, the triple of (IP address,sender,recipient) is added to a database similar to the usual DCC database. If the SMTP client does not try again after embargo seconds and before window seconds after the first attempt, the triple is forgotten. If the SMTP client persists past the embargo, the triple is added to the database and becomes familiar and the message is accepted. Familiar triples are remembered for white seconds after the last accepted mail message. The triple is forgotten if it is ever associated with unsolicited bulk email.

All three durations can be a number of minutes, hours, days, or weeks followed by MINUTES, M, HOURS, H, DAYS, D, WEEKS or W. The default is -G 270seconds,7days,63days. The first duration or the embargo should be longer than open proxies can linger retransmitting. The second window time should be as long as legitimate mail servers persist in retransmitting to recognize embargoed messages whose retransmissions were not received because of network or other problems. The white time should be long enough to recognize and not embargo messages from regular senders.

Usually the DCC greylist system requires that an almost identical copy of the message be retransmitted during the embargo. If weak-body is present, any message with the same triple of sender IP address, sender mail address, and target mail address ends the embargo, even if the body of the message differs.

If weak-IP is present, all mail from an SMTP client at an IP address is accept after any message from the same IP address has been accepted.

Unlike DCC checksums, the contents of greylist databases are private and do not benefit from broad sharing. However, large installations can use more two or more greylist servers flooding triples among themselves. Flooding among greylist servers is controlled by the /var/lib/dcc/grey_flod file.

All greylist cooperating or flooding greylist servers must use the same -G values.

Clients of greylist servers cannot be anonymous and must have client-IDs and passwords assigned in the /var/lib/dcc/ids file. This implies that cdcc commands directed to greylist servers must specify the server-ID.

White- and blacklists are honored by the DCC clients. whitelisted messages are embargoed or checked with a greylist server. The greylist triples of blacklisted messages, messages whose DCC counts make them spam, and other messages known to be spam are sent to a greylist server to be removed from the greylist database and cause an embargo on the next messages with those triples.

Messages whose checksums match greylist server whitelists are not embargoed and the checksums of their triples are not added to the greylist database.

The target counts of embargoed messages are reported to the DCC network to improve the detection of bulk mail.

-W [rate][,chg][,dbsize]

controls quick database cleaning. If the database is larger than dbsize in MBytes, the database has not recently been cleand and is not about to be cleaned, and dccd is receiving fewer than rate requests per second, or if telling DCC clients that the database is about to be cleaned reduces the requests/second by chg, then dccd starts dbclean(8) for a quick database cleaning. The cleaning is abandoned if it takes too long.

The defaults are equivalent to -W 1.0,40.0,RSS where RSS is the maximum dccd resident set size displayed in the system log when the database is opened. A rate of -W 0.0 disables quick cleanings.

-K [no-]type

marks checksums of type (not) be kept or counted in the database (unless they appear in the /var/lib/dcc/whitelist file). Explicit settings add to or remove from the initial contents of the list, which is equivalent to -K Body -K Fuz1 -K Fuz2.

-T [no-]tracemode

causes the server to trace or record some operations. tracemode must be one of the following:

ADMN: administrative requests from the control program, cdcc(8)
ANON: errors by anonymous clients
CLNT: errors by authenticated clients
RLIM: rate-limited messages
QUERY: all queries and reports
RIDC: some messages concerning the report-ID cache that is used to detect duplicate reports from clients
FLOOD1: messages about inter-server flooding connections
FLOOD2: messages about flooded reports
BL: requests from clients in the /var/lib/dcc/blacklist file.
DB: odd database events including long chains of duplicate checksums
WLIST: reports of whitelisted checksums from authenticated, not anonymous DCC clients

The default is ANON CLNT WLIST except for a greylist server which uses ANON CLNT WLIST IDS.

-u anon-delay[,inflate]

changes the number of milliseconds anonymous or unauthenticated clients must wait for answers to their queries and reports. The purpose of this delay is to discourage large anonymous clients. The anon-delay is multiplied by 1 plus the number of recent anonymous requests from IPv4 addresses in a /24 block or IPv6 addresses a /56 block divided by the inflate value.

The string FOREVER turns off all anonymous or unauthenticated access not only for checksum queries and reports but also cdcc(8) stats requests. A missing value for inflate turns off inflation.

The default value is 50, except when -G is used in which case FOREVER is assumed and required.

-C dbclean

changes the default name or path of the program used to rebuild the hash table when it becomes too full. The default value is /usr/lib/dcc/dbclean. The value can include arguments as in -C '/usr/lib/dcc/dbclean -F'.

Dbclean should not be run by dccd except in emergencies such as database corruption or hash table overflow. Dbclean(8) should be run daily with the /usr/lib/dcc/cron-dccd cron script

-L ltype,facility.level

specifies how messages should be logged. Ltype must be error, info, or off to indicate which of the two types of messages are being controlled or to turn off all syslog(3) messages from dccd. 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

-R [RL_SUB],[RL_ANON],[RL_ALL_ANON],[RL_BUGS]

sets one or more of the four rate-limits. RL_SUB limits the number of DCC transactions per second from subscribers or DCC clients with known client-IDs and passwords. This limit applies to each IP address independently.

RL_ANON limits the number of DCC transactions per second from anonymous DCC clients. This limit applies to each IP address independently. It is better to use -u than to change this value to exclude anonymous clients.

RL_ALL_ANON limits the number of DCC transactions per second from all anonymous DCC clients. This limit applies to all anonymous clients as a group, regardless of their IP addresses.

RL_BUGS limits the number of complaints or error messages per second for all anonymous DCC clients as a group as well as for each DCC client by IP address.

The default is equivalent to -R 400,50,2000,0.1

FILES

/var/lib/dcc

is the DCC home directory containing data and control files.

dcc_conf

is the DCC control file.

dcc_db

is the database of mail checksums.

dcc_db.hash

is the mail checksum database hash table.

grey_db

is the database of greylist checksums.

grey_db.hash

is the greylist database hash table.

flod

contains lines controlling DCC flooding of the form:

host[,rport][;src[,lport]] rem-ID [passwd-ID [o-opt [i-opt]]]

where absent optional values are signaled with "-" and

host

is the IP address or name of a DCC server and rport is the name or number of the TCP port used by the remote server.

src

and lport are the source IP address or host name and TCP port from which the outgoing flooding connection should come. The string * specifies any source IP address. Incoming flooding connections must arrive at an address and port specified with -a.

rem-id

is the server-ID of the remote DCC server.

passwd-ID

is a server-ID that is not assigned to a server, but whose first password is used to sign checksum reports sent to the remote system. Either of its passwords are required with incoming reports. If it is absent or "-", outgoing floods are signed with the first password of the local server in the ids file and incoming floods must be signed with either password of the remote server-ID.

i-opt and o-opt

are comma separated lists of

off: turns off flooding to the remote or local system.
no-del: says checksum delete requests are refused by the remote or local server and so turns off sending or accepting delete requests, respectively. By default, delete requests are sent to remote servers and accepted in incoming floods if and only if the peers are exchanging DCC reputations.
del: says delete requests are accepted by the remote or local server.
no-log-del: turns off logging of incoming requests to delete checksums.
passive: is used to tell a server outside a firewall to expect a peer inside to create both of the pair of input and output TCP connections used for flooding. The peer inside the firewall should use SOCKS or NAT on its flod file entry for this system.
SOCKS: is used to tell a server inside a firewall that it should create both of the TCP connections used for flooding and that SOCKS protocol should be used. The peer outside the firewall should use passive on its flod file entry for this system.
NAT: differs from SOCKS only by not using the SOCKS protocol.
IDS->result: converts server-IDs in flooded reports. IDS may be the string ‘self’ to specify the server's own ID. IDS can instead be the string ‘all’ to specify all server-IDs or a pair of server-IDs separated by a dash to specify an inclusive range. result can be the string ‘self’ to translate to the server's own ID. ‘ok’ sends or receives reports without translation. The string ‘reject’ to not send outgoing or refuse incoming reports. Only the first matching conversion is applied. For example, when ‘self->ok,all->reject’ is applied to a locally generated report, the first conversion is made and the second is ignored.
leaf=path-len: does not send reports with paths longer than path-len server-IDs. A path-len of 0 blocks reports from this server.
IPv4: requires only IPv4 addresses to connect to this flooding peer.
IPv6: requires only IPv6 addresses to connect to this flooding peer.
vers: specifies the version of the DCC flooding protocol used by the remote DCC server with a string such as ‘version2’.
trace1: sends information about a single peer like the cdcc(8) command trace FLOOD1 on does for all peers.
trace2: sends information about individual flooded reports like the cdcc(8) command trace FLOOD2 on does for all peers.

grey_flod

is the equivalent of the /var/lib/dcc/flod file used by dccd when it is a greylist server.

flod.map

is an automatically generated file in which dccd records its progress sending or flooding reports to DCC peers.

grey_flod.map

is the equivalent of the /var/lib/dcc/flod.map file used by dccd when it is a greylist server.

ids

contains the IDs and passwords known by the DCC server. An ids file that can be read by others cannot be used. It contains blank lines, comments starting with "#" and lines of the form:

id[,rpt-ok][,no-reps][,trace][,delay=ms[*inflate]] pass1 [pass2]

where

id: is a DCC client-ID or server-ID.
trace: logs activity from clients and flooding peers using the ID.
rpt-ok: overrides -Q by saying that this client is trusted to report only checksums for unsolicited bulk mail.
no-reps: disables DCC Reputation reports from and responses to clients using this ID.
delay=ms[*inflate]: delays answers to systems using the client id. The delay in milliseconds is multiplied by 1 plus the number of recent requests from an IP address using id divided by the inflate value. See -u.
pass1: is the password currently used by clients with identifier id. It is a 1 to 32 character string that does not contain blank, tab, newline or carriage return characters.
pass2: is the optional next password that those clients will use. A DCC server accepts either password if both are present in the file.

Both passwords can be absent if the entry not used except to tell dccd that server-IDs in the flooded reports are valid. The string unknown is equivalent to the null string.

whitelist

contains the DCC server whitelist. It is not used directly but is loaded into the database when dbclean(8) is run.

grey_whitelist

contains the greylist server whitelist. It is not used directly but is loaded into the database when dbclean(8) is run with -G.

blacklist

if present, contains a list of IP addresses and blocks of IP addresses of DCC clients and flooding peers that are ignored. Each line in the file should be blank, a comment starting with '#', or an IP address or block of IP addresses in the form

[trace,][ok,][bad,][no-anon][flood-ok] address

Addresses are single IPv4 or IPv6 addresses, CIDR blocks in the usual form, or a pair of addresses separated by a hyphen (-) specifying an inclusive range. The last line in the file that cover an address applies. Changes to the file are automatically noticed within a few minutes. An IP address is assigned all of the flags from all of lines that apply to it. Addresses or blocks of addresses can be preceded with ok to "punch holes" in blacklisted blocks or specify tracing without blacklisting. Trace logs activity. No-anon blacklists clients only when they use the anonymous client-ID. flood-ok says that DCC flood peering is allowed. Bad is assumed when only an address is present. This mechanism is intended for no more than a few dozen blocks of addresses.

dccd_clients

contains client IP addresses and activity counts.

grey_clients

contains greylist client IP addresses and activity counts.

EXAMPLES

dccd is usually started with other system daemons with something like the script /usr/lib/dcc/rcDCC. That scripts uses values in /var/lib/dcc/dcc_conf to start the server. With the argument stop, /usr/lib/dcc/rcDCC can be used to stop the daemon.

The database grows too large unless old reports are removed. dbclean(8) should be run daily with the /usr/lib/dcc/cron-dccd cron(8) script

HISTORY

dccd is based on an idea from Paul Vixie. It was designed and written at Rhyolite Software, starting in 2000. This document describes version 2.3.167.

dccd(8)