innd - InterNetNews daemon

innd [-aCdfNrsu] [-c days] [-H count] [-i count] [-I address] [-l size] [-m mode] [-n flag] [-o count] [-p fd] [-P port] [-t timeout] [-T count] [-X seconds]

innd, the InterNetNews daemon, handles all incoming NNTP feeds, coordinates the storage, retransmission, and overview generation for all accepted articles, and manages the active(5) and history(5) databases. It handles incoming connections on the NNTP port, and also creates and listens to a local Unix-domain stream socket in order to receive articles from local processes such as nnrpd(8) and rnews(1).

As the master daemon, innd should generally be started at boot and be always running. It listens to a Unix-domain datagram socket for commands to control its activites, commands that can be sent using ctlinnd(8). The current status of innd can be obtained by running CWctlinnd mode, or for more detailed output, innstat(8).

innd can be in one of three operating modes: running, paused, or throttled. Running is the normal mode; when the server is throttled, it closes connections and rejects new ones. Paused is like a temporary throttle, suspending innd's activities but not causing the server to shut down existing connections. The mode is normally changed via ctlinnd(8), either by various automated processes (such as nightly article expiration) or manually by the news administrator, but innd will also throttle itself if it encounters ENOSPC errors in writing data or an excessive number of I/O errors (among other problems).

innd normally takes care of spawning nnrpd(8) to handle connections from news reading clients, but it can be run on a separate port from nnrpd(8) so that feed connections and news reading connections are handled separately (this can often be faster). Normally, innd listens on port 119, the assigned port for NNTP; if it is desireable to run innd and nnrpd(8) on separate ports, it's recommended that nnrpd(8) be given port 119 (since many news reading clients connect only to that port) and that port 433 be used for innd.

The primary configuration files that control innd's activities are incoming.conf, which specifies what remote sites innd will accept connections from, newsfeeds, which specifies what is to be done with incoming articles besides storing them, and inn.conf, which sets a wide variety of configuration parameters. Some parameters in inn.conf(5) can also be set with command-line flags; for these, the command-line flags take precedence if used.

innd should normally not run directly. It must run as the news user or all sorts of file ownership problems may result, and normally the port it listens on (119 or 433) is privileged and must be opened by root. Instead, innd should normally be started via inndstart(8), a small setuid-root program that opens the appropriate port, cleans up the environment, changes to the news user, and then runs innd, passing along any command-line arguments.

To use IPv6, innd must be started by inndstart.

For the options below that override inn.conf settings, see inn.conf(5) for the default values if neither the inn.conf setting nor the command-line option is given.

-a

By default, if a host connects to innd but is not listed in incoming.conf, the connection is handed off to nnrpd (or rejected if noreader is set in inn.conf). If -a is given, incoming.conf is ignored and any host can connect and transfer articles. This flag should never be used with an accessible server connected to Usenet; it would open the server up for all sorts of abuse.

-c days

innd normally rejects any article that is older (in days) than the value of artcutoff in inn.conf. This option, if given, overrides the value of that setting. If days is 0, this check is suppressed and innd will accept articles regardless of how old they are.

-C

This flag tells innd to accept and propagate but not actually process cancel or supersede messages. This is intended for sites concerned about abuse of cancels, or that wish to use another cancel mechanism with stronger authentication.

-d, -f

innd normally puts itself into the background, points its standard output and error to log files, and disassociates itself from the terminal. Using -d prevents all of this, resulting in log messages being written to standard output; this is generally useful only for debugging. Using -f prevents the backgrounding and disassociation but still redirects output; it may be useful if you want to monitor innd with a program that would be confused by forks.

-H count, -T count, -X seconds

These flags control the number of connections per minute that are allowed. This code is meant to protect your server from newsreader clients that make too many connections per minute (and therefore these flags are probably only useful when innd is spawning nnrpd). You probably should not use these options unless you're having problems. The table used for this check is fixed at 128 entries and is used as a ring; the size was chosen to make calculating the index easy and to be fairly sure that it won't run out of space. In practice, it is unlikely that even half the table will be used at any given moment. The -H flag limits the number of times a host is allowed to connect to the server per the time interval given by -X. The default is CW2. The -T flag limits the total number of incoming connections per the time interval given by -X. The maximum value is CW128, and the default is CW60.

-i count

innd normally allows a maximum number of concurrent NNTP connections given by the value of maxconnections in inn.conf. This option, if given, overrides the value of that setting. If count is CW0, this check is suppressed.

-I address

Normally if innd itself binds to a port, it lets the operating system pick the source IP address (unless bindaddress is set in inn.conf). If this option is given, it specifies the IP address that INN should bind as. This is only relevant for servers with multiple local IP addresses. The IP address must be in dotted quad (CWnnn.nnn.nnn.nnn) format. This option is rarely useful since innd should not be binding to a port itself. Instead, use inndstart(8) and its analgous -I option.

-l size

innd normally rejects any article larger than the value of maxartsize in inn.conf. This option, if given, overrides the value of that setting and specifies a maximum article size of size. If size is CW0, this check is suppressed.

-m mode

Normally innd starts in the CWrunning mode. If this option is given, it specifies what mode innd should start in. mode should begin with one of CWg, CWp, or CWt, and the starting mode will be set to CWrunning, CWpaused, or CWthrottled, respectively, based on that initial letter. (CWg is short for CWgo.)

-N

If this option is given, any filters (Perl, Tcl, or Python) are disabled before innd starts (normally, filters default to being enabled). The filters can be enabled after innd has started with ctlinnd(8).

-n flag

Whether innd allows (and hands off to nnrpd) reader connections while paused or throttled is normally determined by the value of readerswhenstopped in inn.conf). This option, if given, overrides that value. If flag is CWn, innd will not allow readers if it is paused or throttled. If flag is CWy, readers will be allowed regardless of innd's operating mode.

-o count

This flag limits the number of file descriptors that are available for outgoing file feeds. The default is the number of available file descriptors minus some reserved for internal use (which could potentially starve innd of descriptors to use for accepting new connections). If innd has more file feeds than count, some of them will be buffered and only written out periodically. Normally you never need to use this option, since the number of outgoing feeds is fixed, being the number of file feeds configured in newsfeeds, and is generally small (particularly given that innfeed(8) is now used for most outgoing feeds at large sites).

-p fd

If this flag is given, innd expects the file descriptor given by fd to already be open and bound to the appropriate local port and to be suitable for listening to for incoming connections. This is how inndstart tells innd which open file descriptor is the network connection. If this flag is not given, innd will attempt to open its network socket itself. inndstart always passes this flag to innd.

-P port

The port innd should listen on is normally given by the value of port in inn.conf. This option, if given, overrides that value and specifies the port that innd should bind to. This option is rarely useful since innd normally does not bind itself; instead the analgous -P option to inndstart(8) should be used. Since innd should never be run as root, port has to be a non-privileged port (one larger than 1024).

-r

Instructs innd to renumber the active file after starting, just as if a CWctlinnd renumber command were sent.

-s

Just check the syntax of the newsfeeds file and exit. innd will exit with a non-zero status if any errors are found; the actual errors will be reported via syslog(3).

-t seconds

Normally, innd will flush any changes to history and the active file after 300 seconds of inactivity. This option changes that timeout to seconds.

-u

The news log (the trace information for every article accepted by innd) is normally buffered. This option changes the log to be unbuffered.

Arriving articles that have a Control header are called control messages. Except for cancel messages, these messages are handled by controlchan(8) via a feed set up in newsfeeds.

(Cancel messages update the history database, so they must be handled internally; the cost of syncing, locking, then unlocking would be too high given the number of cancel messages that are received.)

The distribution of control messages is different than that of standard articles. Control messages are normally filed into the pseudo-newsgroup named CWcontrol regardless of which newsgroup they were actually posted to. If, however, a CWcontrol.command newsgroup exists that matches the control command, the control message will be filed into that group instead. For example, a newgroup control message will be filed in CWcontrol.newgroup if that group exists; otherwise, it will be filed in CWcontrol.

If you want to specifically feed all control messages to a given site regardless of whether the control messages would affect the newsgroups you're feeding that site, you can put the appropriate control newsgroup in the subscription list. For example, to feed all cancel messages to a given remote site (normally a bad idea), add CWcontrol.cancel to its subscription list. Normally it's best to exclude the control newsgroups from feeds to keep from sending your peers more control messages than they care about.

newgroup and rmgroup control messages receive additional special treatment. If one of these control messages is approved and posted to the newsgroup being created or removed, the message will be sent to all sites whose subscription patterns would cause them to receive articles posted to that group. For example, if a newgroup control message for a nonexistent newsgroup CWnews.admin.meow is received, it will be sent to any site whose subscription pattern would cause it to receive CWnews.admin.meow if that newsgroup existed (such as a pattern of CWnews.admin.*). For this reason, it is correct to post newgroup messages to the newsgroup that the control message would create. It is not generally correct to crosspost newgroup messages to some well-propagated newsgroup; not only will this not actually improve their propagation to sites that want such control messages, but it will also cause sites that do not want those control messages to receive them.

If a control message is posted to a group whose name ends with the four characters CW.ctl, this suffix is stripped off and the control message is propagated as if it were posted to the base group. For example, a cancel message posted to CWnews.admin.ctl will be sent to all sites that subscribe to CWcontrol.cancel (or CWcontrol if that newsgroup doesn't exist) or CWnews.admin. This behavior is present for historical compatibility reasons and should be considered obsolete; support for the CW.ctl suffix may be removed in a future version of INN.

Finally, articles posted to newsgroups beginning with CWto. are treated specially. Provided that either that newsgroup exists in the active file or mergetogroups is set in inn.conf, the remainder of the newsgroup is taken to be a site name, as configured in newsfeeds, and the article is sent to that site. If mergetogroups is set, the article will be filed in the group named CWto (which must exist in the active file). For example, with mergetogroups set, an article posted to CWto.uunet will be filed in CWto and sent to the site CWuunet.

innd implements the NNTP commands defined in RFC 977, with the following differences:

1.

The LIST command may be followed by an optional ACTIVE, ACTIVE.TIMES, or NEWSGROUPS. There is only basic support for LIST in innd since feeding peers normally don't need it; see nnrpd(8) for full support.

2.

The AUTHINFO USER and AUTHINFO PASS commands are implemented, although the authentication is currently limited to matching a password for a given peer specified in incoming.conf. These are based on the reference Unix implementation.

3.

A new command, MODE READER, is implemented. This command will cause the server to pass the connection to nnrpd.

4.

The streaming extension (MODE STREAM, CHECK, and TAKETHIS) is fully supported.

5.

A batch transfer command, XBATCH byte-count, is provided. This command will read byte-count bytes and store them for later processing by rnews(1) (which must be run separately, probably from cron). See innxbatch(8) and backends/sendxbatches for more details on this extension.

6.

innd implements a limited subset of the protocol useful for transferring news. The only other commands implemented are HEAD, HELP, IHAVE, STAT, and QUIT. The remaining commands are mostly only useful for readers and are implemented by nnrpd(8).

innd modifies as few article headers as possible, although it could be better in this area.

Empty headers and headers that consist of nothing but whitespace are dropped.

The local site's name (as set with the pathhost parameter in inn.conf) and an exclamation point are prepended to the Path: header, provided the first site name in the Path: header is different from the local one. In addition, pathalias may be similarly prepended to the Path: header; see inn.conf(5) for the details.

The Xref: header is removed and a new one created.

A Lines: header will be added if the article was missing one.

innd does not rewrite incorrect headers. For example, it will not replace an incorrect Lines header, though it may reject such an article depending on the value of linecountfuzz in inn.conf.

In order to efficiently apply a large number of local cancels (such as from processing NoCeMs or from some other external source), INN supports a special feed mode available only to connections to the local Unix domain socket (not to connections to any network sockets).

To enter this mode, connect to the Unix domain socket (pathrun/nntpin) and send the command MODE CANCEL. The response will have code CW284. Every subsequent line sent on that connection should consist of a single message ID. An attempt will be made to cancel that message ID, and the server will reply CW289 for success or CW484 for failure. (Failure can occur, for example, if the server is paused or throttled, or the Message-ID is corrupt. Failure does not occur if the article to be cancelled does not exist.)

innd reports all incoming articles in its log file (pathlog/news). This is a text file with a variable number of space-separated fields in one of the following formats:

    mon dd hh:mm:ss.mmm + feed <message-id> site ...
    mon dd hh:mm:ss.mmm j feed <message-id> site ...
    mon dd hh:mm:ss.mmm c feed <message-id> Cancelling <message-id>
    mon dd hh:mm:ss.mmm - feed <message-id> reason
    mon dd hh:mm:ss.mmm ? feed <message-id> reason

There may also be hostname and/or size fields after the message ID depending on the settings of nntplinklog and logartsize in inn.conf.

The first three fields are the date and time to millisecond resolution. The fifth field is the site that sent the article (based on the Path header) and the sixth field is the article's message ID; they will be a question mark if the information is not available.

The fourth field indicates whether the article was accepted or not. If it is a plus sign, then the article was accepted. If it is the letter CWj then the article was accepted, but all of the newsgroups to which the article was posted were set to mode CWj in the active file (or not listed in the active file and wanttrash was set in inn.conf) so the article was filed into the CWjunk newsgroup. In both of these cases, the article has been accepted and the CWsite ... field contains the space-separated list of sites to which the article is being sent.

If the fourth field is the letter CWc, then a cancel message was accepted before the original article arrived, and a history entry for the cancelled message was created so that innd will reject that message if it arrives later.

If the fourth field is a minus sign, then the article was rejected. The reasons for rejection generated by innd include:

    "%s" header too long
    "%s" wants to cancel <%s> by "%s"
    Article exceeds local limit of %s bytes
    Article posted in the future -- "%s"
    Bad "%s" header
    Can't write history
    Duplicate
    Duplicate "%s" header
    EOF in headers
    Linecount %s != %s +- %s
    Missing %s header
    No body
    No colon-space in "%s" header
    No space
    Space before colon in "%s" header
    Too old -- "%s"
    Unapproved for "%s"
    Unwanted newsgroup "%s"
    Unwanted distribution "%s"
    Whitespace in "Newsgroups" header -- "%s"

where CW%s, above, is replaced by more specific information. (The Perl, Python, andr Tcl filters, if used, may reject articles with other reasons.)

If the fourth field is the letter CW?, the article contains strange strings, such as CR without LF or LF without CR. (These characters should never occur in isolation, only together as CRLF to indicate the end of a line.) This log message is just informational, to give an idea of how widespread such articles are; innd does not reject such articles.

Note that when wanttrash is set to true in inn.conf and an article is received that isn't posted to any valid newsgroups, it will be accepted and logged with two lines, a CWj line and a minus sign line.

innd also makes extensive reports through syslog(3). The first word of the log message will be the name of the site if the entry is site-specific (such as a connected message). The first word will be CWSERVER if the message relates to the server itself, such as when a read error occurs.

If the second word is the four letters CWcant, then an error is being reported. (The absence of an apostrophe is intentional; it makes it easier to grep from the command line and easier to find error messages in FAQs using a search engine.) In this case, the next two words generally name the system call or library routine that failed and the object upon which the action was being performed. The rest of the line may contain other information.

In other cases, the second word attempts to summarize what change has been made, while the rest of the line gives more specific information. The word CWinternal generally indicates an internal logic error.

innd will catch SIGTERM and SIGHUP and shut down. If -d is used, SIGINT will also be caught and will result in an orderly shutdown.

innd will catch the SIGUSR1 signal and recreate the control channel used by ctlinnd(8).

innd normally attempts to strip IP options from incoming connections, since it uses IP-based authentication and source routing can confuse that. However, this doesn't work on all systems, and it doesn't work at all in the presence of IPv6 support (and is disabled in that case). Hence, if using innd with IPv6 support, make sure that your kernel or router disables source routing.

Written by Rich CW$alz <rsalz@uunet.uu.net> for InterNetNews.

$Id: innd.8,v 1.37 2003/03/19 00:19:59 vinocur Exp $

active(5), ctlinnd(8), dbz(3), history(5), incoming.conf(5), inn.conf(5), newsfeeds(5), nnrpd(8), rnews(1), syslog(3).