B. The GNATS network server -- gnatsd

This section describes in details how the GNATS network daemon works. This information is mainly assumed to be useful for developers of GNATS client software.

B.1 Description of gnatsd
B.2 gnatsd options
B.3 gnatsd command protocol
B.4 gnatsd commands
B.5 gnatsd environment variables

B.1 Description of gnatsd

The gnatsd network daemon is used to service remote GNATS requests such as querying PRs, PR creation, deletion, and editing, and miscellaneous database queries. It uses a simple ASCII-based command protocol (similar to SMTP or POP3) for communicating with remote clients.

It also provides a security model based either on IP-based authentication (generally considered very weak) or username/passwords, where passwords may be in cleartext, UNIX crypt or MD5 hash format. Access through gnatsd is granted according to certain predefined access levels. Access levels are further discussed in Controlling access to databases. It should be emphasized that security has not been a focus of development until now, but future versions are expected to address this more thoroughly.

All of the GNATS clients are capable of communicating via the GNATS remote protocol to perform their functions.

gnatsd is usually started from the inetd facility and should run as the `gnats' user (the actual username of this user is configurable during installation, see section Configuring and compiling the software for details.)

B.2 gnatsd options

The daemon supports the following command-line options:

gnatsd [--database database | -d database] [--not-inetd | -n] [--max-access-level level | -m level] [--version | -V] [--help | -h]

-V, --version

Prints the program version to stdout and exits.

-h, --help

Prints a short help text to stdout and exits.

-d, --database

Specifies the default database which is to be serviced by this invocation of gnatsd. (The selected database may be changed via the CHDB command; the name set with this option is simply the default if no CHDB command is issued.) If no database is specified, the database named default is assumed. This option overrides the database specified in the GNATSDB environment variable.

-n, --not-inetd

As its name suggests, indicates that gnatsd is not being invoked from inetd. This can be used when testing gnatsd, or if it is being run via ssh or some other mechanism.

This has the effect of using the local hostname where gnatsd is being invoked for authentication purposes, rather than the remote address of the connecting client.

--max-access-level, -m

Specifies the maximum access level that the connecting client can authenticate to. Authentication is as normal but if the user or host authenticates at a higher level, access level is still forced to this level. See Controlling access to databases for details on access levels.

B.3 gnatsd command protocol

Commands are issued to gnatsd as one or more words followed by a carriage-return/linefeed pair. For example, the CHDB (change database) command is sent as `CHDB database<CR><LF>' (the CRLF will not be explicitly written for future examples.)

Replies from gnatsd are returned as one or more response lines containing a 3-digit numeric code followed by a human-readable string; the line is terminated with a <CR><LF> pair. For example, one possible response to the CHDB command above would be:

210 Now accessing GNATS database 'database'.

The three-digit code is normally followed by a single ASCII space (character 0x20). However, if additional response lines are to be returned from the server, there will be a single dash `-' instead of the space character after the three-digit code.

Response code values are divided into ranges. The first digit reflects the general type of response (such as "successful" or "error"), and the subsequent digits identify the specific type of response.

Codes 200-299

Positive response indicating that the command was successful. No subsequent data will be transmitted with the response. In particular, code 210 (CODE_OK) is used as the positive result code for most simple commands.

Commands that expect additional data from the client (such as SUBM or VFLD) use a two-step mechanism for sending the data. The server will respond to the initial command with either a 211 (CODE_SEND_PR) or 212 (CODE_SEND_TEXT) response line, or an error code if an error occurred with the initial command. The client is then expected to send the remaining data using the same quoting mechanism as described for server responses in the 300-349 range. The server will then send a final response line to the command.

Codes 300-399

Positive response indicating that the query request was successful, and that a PR or other data will follow. Codes 300-349 are used when transmitting PRs, and 350-399 are used for other responses.

Codes in the 300-349 range are followed by a series of CRLF-terminated lines containing the command response, usually a PR. The final line of the result is a single period `.'. Result lines that begin with a period have an extra period prepended to them.

Codes in the 350-399 range use a different scheme for sending their responses. The three-digit numeric code will be followed by either a dash `-' or a single space. If the code is followed by a dash, that indicates that another response line will follow. The final line of the response has a single space after the three-digit code.

In previous versions of the protocol the first line of a CODE_INFORMATION (310) response was to be ignored. This is no longer the case. Instead, any lines marked with code CODE_INFORMATION_FILLER (351) are to be ignored. This allows the server to transmit additional headers or other human-readable text that can be safely ignored by the clients.

Codes 400-599

An error occurred, usually because of invalid command parameters or invalid input from the client, missing arguments to the command, or a command was issued out of sequence. The human-readable message associated with the response line describes the general problem encountered with the command.

Multiple error messages may be returned from a command; in this case the `-' continuation character is used on all but the last response line.

Codes 600-799

An internal error occurred on the server, a timeout occurred reading data from the client, or a network failure occurred. These errors are of the "this should not occur" nature, and retrying the operation may resolve the problem. Fortunately, most GNATS transactions are idempotent; unfortunately, locking the database or a PR are not repeatable actions (we cannot determine if an existing lock is the one we originally requested, or someone else's).

B.4 gnatsd commands

Note that the set of GNATS commands and their responses is somewhat inconsistent and is very much in flux. At present the GNATS clients are rather simple-minded and not very strict about processing responses. For example, if the server were to issue a code 300 (CODE_PR_READY) response to a CHDB command, the client would happily expect to see a PR appear (and would print it out if one was sent).

It is thus suggested that any clients that use the GNATS protocol be equally flexible about the way received responses are handled; in particular, only the first digit of the response code should be assumed to be meaningful, although subsequent digits are needed in some cases (codes 300-399). No attempt should be made to parse the message strings on error response lines; they are only intended to be read by humans, and will be changed on a regular basis.

Almost every command may result in the response 440 (CODE_CMD_ERROR). This indicates that there was a problem with the command arguments, usually because of insufficient or too many arguments being specified.

Access to most gnatsd commands requires a certain access level. For details of this, see Privileged gnatsd commands.

USER [userid password]

Specifies the userid and password for database access. Either both a username and password must be specified, or they both may be omitted; in the latter case, the current access level is returned.

The possible server responses are:

350 (CODE_INFORMATION)
The current access level is specified.

422 (CODE_NO_ACCESS)
A matching username and password could not be found.

200 (CODE_OK)
A matching username and password was found, and the login was successful.

QUIT

Requests that the connection be closed. Possible responses:

201 (CODE_CLOSING)
Normal exit.

The QUIT command has the dubious distinction of being the only command that cannot fail.

LIST list type

Describes various aspects of the database. The lists are returned as a list of records, one per line. Each line may contain a number of colon-separated fields.

Possible values for list type include

Categories
Describes the legal categories for the database.

Submitters
Describes the set of submitters for the database.

Responsible
Lists the names in the responsible administrative file, including their full names and email addresses.

States
Lists the states listed in the state administrative file, including the state type (usually blank for most states; the closed state has a special type).

Classes
Lists the set of PR classes and their associated human-readable descriptions.

FieldNames
Lists the entire set of PR fields.

InitialInputFields
Lists the fields that should be provided when a PR is initially entered.

Databases
Lists the set of databases.

The possible responses are:

301 (CODE_TEXT_READY)
Normal response, followed by the records making up the list as described above.

416 (CODE_INVALID_LIST)
The requested list does not exist.

FTYP field [field ...]

Describes the type of data held in the field(s) specified with the command. The currently defined data types are:

Text
A plain text field, containing exactly one line.

MultiText
A text field possibly containing multiple lines of text.

Enum
An enumerated data field; the value is restricted to one entry out of a list of values associated with the field.

MultiEnum
The field contains one or more enumerated values. Values are separated with spaces or colons `:'.

Integer
The field contains an integer value, possibly signed.

Date
The field contains a date.

TextWithRegex
The value in the field must match one or more regular expressions associated with the field.

The possible responses are:

350 (CODE_INFORMATION)
The normal response; the supplied text is the data type.

410 (CODE_INVALID_FIELD_NAME)
The specified field does not exist.

If multiple field names were given, multiple response lines will be sent, one for each field, using the standard continuation protocol; each response except the last will have a dash `-' immedately after the response code.

FTYPINFO field property

Provides field-type-related information. Currently, only the property `separators' for MultiEnum fields is supported. When `separators' is specified, the possible return codes are:

350 (CODE_INFORMATION) A proper MultiEnum field was specified and the returned text is the string of separators specified for the field in the dbconfig file (see section 4.3.3 Field datatypes) quoted in ''s.

435 (CODE_INVALID_FTYPE_PROPERTY) The `separators' property is not defined for this field, i.e. the specified field is not of type MultiEnum.

Currently, specifying a different property than `separators' results in return code 435 as above.

FDSC field [field ... ]

Returns a human-readable description of the listed field(s). The possible responses are:

350 (CODE_INFORMATION) The normal response; the supplied text is the field description.

410 (CODE_INVALID_FIELD_NAME) The specified field does not exist.

Like the FVLD command, the standard continuation protocol will be used if multiple fields were specified with the command.

FIELDFLAGS field [field ... ]

Returns a set of flags describing the specified field(s). The possible responses are either

410 (CODE_INVALID_FIELD_NAME)
meaning that the specified field is invalid or nonexistent, or

350 (CODE_INFORMATION)
which contains the set of flags for the field. The flags may be blank, which indicate that no special flags have been set for this field.

Like the FDSC and FTYP commands, multiple field names may be listed with the command, and a response line will be returned for each one in the order that the fields appear on the command line.

The flags include:

textsearch
The field will be searched when a text field search is requested.

allowAnyValue
For fields that contain enumerated values, any legal value may be used in the field, not just ones that appear in the enumerated list.

requireChangeReason
If the field is edited, a reason for the change must be supplied in the new PR text describing the reason for the change. The reason must be supplied as a multitext PR field in the new PR whose name is field-Changed-Why (where field is the name of the field being edited).

readonly
The field is read-only, and cannot be edited.

FVLD field

Returns one or more regular expressions or strings that describe the valid types of data that can be placed in field. Exactly what is returned is dependent on the type of data that can be stored in the field. For most fields a regular expression is returned; for enumerated fields, the returned values are the list of legal strings that can be held in the field.

The possible responses are:

301 (CODE_TEXT_READY)
The normal response, which is followed by the list of regexps or strings.

410 (CODE_INVALID_FIELD_NAME) The specified field does not exist.

VFLD field

VFLD can be used to validate a given value for a field in the database. The client issues the VFLD command with the name of the field to validate as an argument. The server will either respond with 212 (CODE_SEND_TEXT), or 410 (CODE_INVALID_FIELD_NAME) if the specified field does not exist.

Once the 212 response is received from the server, the client should then send the line(s) of text to be validated, using the normal quoting mechanism described for PRs. The final line of text is followed by a line containing a single period, again as when sending PR text.

The server will then either respond with 210 (CODE_OK), indicating that the text is acceptable, or one or more error codes describing the problems with the field contents.

INPUTDEFAULT field [field ... ]

Returns the suggested default value for a field when a PR is initially created. The possible responses are either 410 (CODE_INVALID_FIELD_NAME), meaning that the specified field is invalid or nonexistent, or 350 (CODE_INFORMATION) which contains the default value for the field.

Like the FDSC and FTYP commands, multiple field names may be listed with the command, and a response line will be returned for each one in the order that the fields appear on the command line.

RSET

Used to reset the internal server state. The current query expression is cleared, and the index of PRs may be reread if it has been updated since the start of the session. The possible responses are:

200 (CODE_OK)
The state has been reset.

440 (CODE_CMD_ERROR)
One or more arguments were supplied to the command.

6xx (internal error)
There were problems resetting the state (usually because the index could not be reread). The session will be immediately terminated.

LKDB

Locks the main GNATS database. No subsequent database locks will succeed until the lock is removed. Sessions that attempt to write to the database will fail. The possible responses are:

200 (CODE_OK) The lock has been established.

440 (CODE_CMD_ERROR) One or more arguments were supplied to the command.

431 (CODE_GNATS_LOCKED) The database is already locked, and the lock could not be obtained after 10 seconds.

6xx (internal error) An internal error occurred, usually because of permission or other filesystem-related problems. The lock may or may not have been established.

UNDB

Unlocks the database. Any session may steal a database lock; no checking of any sort is done. The possible responses are:

200 (CODE_OK)
The lock has been removed.

432 (CODE_GNATS_NOT_LOCKED)
The database was not locked.

440 (CODE_CMD_ERROR)
One or more arguments were supplied to the command.

6xx (internal error)
The database lock could not be removed, usually because of permissions or other filesystem-related issues.

LOCK PR user [pid]

Locks the specified PR, marking the lock with the user name and the optional pid. (No checking is done that the user or pid arguments are valid or meaningful; they are simply treated as strings.)

The EDIT command requires that the PR be locked before it may be successfully executed. However, it does not require that the lock is owned by the editing session, so the usefulness of the lock is simply as an advisory measure.

The APPN and REPL commands lock the PR as part of the editing process, and they do not require that the PR be locked before they are invoked.

The possible responses are:

440 (CODE_CMD_ERROR)
Insufficient or too many arguments were specified to the command.

300 (CODE_PR_READY)
The lock was successfully obtained; the text of the PR (using the standard quoting mechanism for PRs) follows.

400 (CODE_NONEXISTENT_PR)
The PR specified does not exist.

430 (CODE_LOCKED_PR)
The PR is already locked by another session.

6xx (internal error)
The PR lock could not be created, usually because of permissions or other filesystem-related issues.

UNLK PR

Unlocks PR. Any user may unlock a PR, as no checking is done to determine if the requesting session owns the lock.

The possible responses are:

440 (CODE_CMD_ERROR)
Insufficient or too many arguments were specified to the command.

200 (CODE_OK)
The PR was successfully unlocked.

433 (CODE_PR_NOT_LOCKED)
The PR was not locked.

6xx (internal error)
The PR could not be unlocked, usually because of permission or other filesystem-related problems.

DELETE PR

Deletes the specified PR. The user making the request must have admin privileges (see section Controlling access to databases). If successful, the PR is removed from the filesystem and the index file; a gap will be left in the numbering sequence for PRs. No checks are made that the PR is closed.

The possible responses are:

200 (CODE_OK)
The PR was successfully deleted.

422 (CODE_NO_ACCESS)
The user requesting the delete does not have admin privileges.

430 (CODE_LOCKED_PR)
The PR is locked by another session.

431 (CODE_GNATS_LOCKED)
The database has been locked, and no PRs may be updated until the lock is cleared.

6xx (internal error)
The PR could not be successfully deleted, usually because of permission or other filesystem-related problems.

CHEK [initial]

Used to check the text of an entire PR for errors. Unlike the VFLD command, it accepts an entire PR at once instead of the contents of an individual field.

The initial argument indicates that the PR text to be checked is for a PR that will be newly created, rather than an edit or replacement of an existing PR.

After the CHEK command is issued, the server will respond with either a 440 (CODE_CMD_ERROR) response indicating that the command arguments were incorrect, or a 211 (CODE_SEND_PR) response code will be sent.

Once the 211 response is received from the server, the client should send the PR using the normal PR quoting mechanism; the final line of the PR is then followed by a line containing a single period, as usual.

The server will then respond with either a 200 (CODE_OK) response, indicating there were no problems with the supplied text, or one or more error codes listing the problems with the PR.

EDIT PR

Verifies the replacement text for PR. If the command is successful, the contents of PR are completely replaced with the supplied text. The PR must previously have been locked with the LOCK command.

The possible responses are:

431 (CODE_GNATS_LOCKED)
The database has been locked, and no PRs may be updated until the lock is cleared.

433 (CODE_PR_NOT_LOCKED)
The PR was not previously locked with the LOCK command.

400 (CODE_NONEXISTENT_PR)
The specified PR does not currently exist. The SUBM command should be used to create new PRs.

211 (CODE_SEND_PR)
The client should now transmit the replacement PR text using the normal PR quoting mechanism. After the PR has been sent, the server will respond with either 200 (CODE_OK) indicating that the edit was successful, or one or more error codes listing problems either with the replacement PR text or errors encountered while updating the PR file or index.

EDITADDR address

Sets the e-mail address of the person communicating with gnatsd. The command requires at least the edit access level.

The possible responses are:

200 (CODE_OK)
The address was successfully set.

440 (CODE_CMD_ERROR)
Invalid number of arguments were supplied.

APPN PR field

REPL PR field

Appends to or replaces the contents of field in PR with the supplied text. The command returns a 201 (CODE_SEND_TEXT) response; the client should then transmit the new field contents using the standard PR quoting mechanism. After the server has read the new contents, it then attempts to make the requested change to the PR.

The possible responses are:

200 (CODE_OK)
The PR field was successfully changed.

400 (CODE_NONEXISTENT_PR)
The PR specified does not exist.

410 (CODE_INVALID_FIELD_NAME)
The specified field does not exist.

402 (CODE_UNREADABLE_PR)
The PR could not be read.

431 (CODE_GNATS_LOCKED)
The database has been locked, and no PRs may be updated until the lock is cleared.

430 (CODE_LOCKED_PR)
The PR is locked, and may not be altered until the lock is cleared.

413 (CODE_INVALID_FIELD_CONTENTS)
The supplied (or resulting) field contents are not valid for the field.

6xx (internal error)
An internal error occurred, usually because of permission or other filesystem-related problems. The PR may or may not have been altered.

SUBM

Submits a new PR into the database. The supplied text is verified for correctness, and if no problems are found a new PR is created.

The possible responses are:

431 (CODE_GNATS_LOCKED)
The database has been locked, and no PRs may be submitted until the lock is cleared.

211 (CODE_SEND_PR)
The client should now transmit the new PR text using the normal quoting mechanism. After the PR has been sent, the server will respond with either a 200 (CODE_OK) response indicating that the new PR has been created (and mail sent to the appropriate persons), or one or more error codes listing problems with the new PR text.

CHDB database

Switches the current database to the name specified in the command.

The possible responses are:

422 (CODE_NO_ACCESS)
The user does not have permission to access the requested database.

417 (CODE_INVALID_DATABASE)
The database specified does not exist, or one or more configuration errors in the database were encountered.

220 (CODE_OK)
The current database is now database. Any operations performed will now be applied to database.

DBLS

Lists the known set of databases.

The possible responses are:

6xx (internal error)
An internal error was encountered while trying to obtain the list of available databases, usually due to lack of permissions or other filesystem-related problems, or the list of databases is empty.

301 (CODE_TEXT_READY)
The list of databases follows, one per line, using the standard quoting mechanism. Only the database names are sent.

The gnatsd access level `listdb' denies access until the user has authenticated with the USER command. The only other command available at this access level is DBLS. This access level provides a way for a site to secure its GNATS databases while still providing a way for client tools to obtain a list of the databases for use on login screens etc. See section Controlling access to databases.

DBDESC database

Returns a human-readable description of the specified database.

Responses include:

6xx (internal error)
An internal error was encountered while trying to read the list of available databases, usually due to lack of permissions or other filesystem-related problems, or the list of databases is empty.

350 (CODE_INFORMATION)
The normal response; the supplied text is the database description.

417 (CODE_INVALID_DATABASE)
The specified database name does not have an entry.

EXPR query expression

Specifies a query expression used to limit which PRs are returned from the QUER command. The expression uses the normal query expression syntax, (see section 2.4.3 Query expressions).

Multiple EXPR commands may be issued; the expressions are boolean ANDed together.

Expressions are cleared by the RSET command.

Possible responses include:

415 (CODE_INVALID_EXPR)
The specified expression is invalid, and could not be parsed.

200 (CODE_OK)
The expression has been accepted and will be used to limit the results returned from QUER.

QFMT query format

Use the specified query format to format the output of the QUER command. The query format may be either the name of a query format known to the server (see section 4.3.5 Named query definitions), or an actual query format (see section 2.4.2 Formatting query-pr output). The possible responses are:

200 (CODE_OK)
The normal response, which indicates that the query format is acceptable.

440 (CODE_CMD_ERROR)
No query format was supplied.

418 (CODE_INVALID_QUERY_FORMAT)
The specified query format does not exist, or could not be parsed.

QUER [PR] [PR] [...]

Searches the contents of the database for PRs that match the (optional) specified expressions with the EXPR command. If no expressions were specified with EXPR, the entire set of PRs is returned.

If one or more PRs are specified on the command line, only those PRs will be searched and/or output.

The format of the output from the command is determined by the query format selected with the QFMT command.

The possible responses are:

418 (CODE_INVALID_QUERY_FORMAT)
A valid format was not specified with the QFMT command prior to invoking QUER.

300 (CODE_PR_READY)
One or more PRs will be output using the requested query format. The PR text is quoted using the normal quoting mechanisms for PRs.

220 (CODE_NO_PRS_MATCHED)
No PRs met the specified criteria.

ADMV field key [subfield]

Returns an entry from an administrative data file associated with field. key is used to look up the entry in the data file. If subfield is specified, only the value of that subfield is returned; otherwise, all of the fields in the adm data file are returned, separated by colons `:'.

The responses are:

410 (CODE_INVALID_FIELD_NAME) The specified field does not exist.

221 (CODE_NO_ADM_ENTRY) An adm entry matching the key was not found, or the field does not have an adm file associated with it.

350 (CODE_INFORMATION) The normal response; the supplied text is the requested field(s).

B.5 gnatsd environment variables

gnatsd supports the GNATSDB environment varable in almost the same way as the GNATS tools do. This variable is used to determine which database to use. For a local database, it contains the name of the database to access. gnatsd cannot service remote databases (though it might be interesting if it could) so the database is always assumed to be local.

If GNATSDB is not set and the --database option is not supplied, it is assumed that the database is local and that its name is `default'.