Allow access and error logging to stdout by configuring a path of "-".
Thanks to @firstname.lastname@example.org for the suggestion.
|1 month ago|
|contrib/init||7 months ago|
|.gitignore||1 year ago|
|LICENSE||1 year ago|
|README.md||1 month ago|
|certificate.go||8 months ago|
|config.go||8 months ago|
|dirlist.go||7 months ago|
|dynamic.go||2 months ago|
|example.conf||8 months ago|
|handler.go||1 month ago|
|logging.go||8 months ago|
|main.go||1 month ago|
The Unsinkable Molly Brown is a full-featured Gemini server implemented in Go.
For more information on the Gemini protocol see:
Molly Brown is intended to be a full-featured Gemini server which is suitable for use in pubnix or similar shared-hosting environments, where users can upload their content but do not have access to the main configuration file (of course, it is also perfectly suitable for single user environments, but its multi-user supports sets it apart from many other Gemini servers).
Molly Brown features:
text/geminicontent in place of filenames.
text/xml. The file extension for
gmi, but this can be overrideen too.
.mollyfiles, analogous to Apache’s
The follow features are planned for the future:
Molly Brown is known to run on:
Please let us know if you get it to work on some other platform!
Molly Brown only has a single dependency beyond the Go standard library, which is this TOML parsing library.
The easiest way for now to install Molly Brown is to use the standard
go (note I said “easiest”, not “easy” - this is still a
pretty clunky manual process, sorry). Unfortunately, you have to do a
little bit of preparation for this to work (unless you’re a Go
developer yourself in which case you surely already have this done)…
(you can in fact put your $GOPATH anywhere you like, but
~/go is the
go get tildegit.org/solderpunk/molly-brown. If everything goes
well, the end result of this will be that you’ll have the Molly Brown
source code sitting in
and an executable binary sitting at
~/go/bin/molly-brown. If it
makes you happier or your life easier, you can copy that binary to
/usr/sbin/ or anywhere else.
Molly Brown can run without a configuration file, in which case it
will use compiled-in default settings. However, these settings are
oriented toward quick test runs with all files in the current
working directory. For regular use, you will want to override these
defaults with more suitable settings from a config file. An example
config file showing the syntax for all settings can be found in the
~/go/src/tildegit.org/solderpunk/molly-brown/ directory with the
example.conf. You can copy this file to
and edit it to suit your environment. All the options are explained
further below. If you put your configuration file somewhere other
/etc/molly.conf, you will need to use Molly Brown’s
command line option to tell Molly Brown where to find it.
Molly Brown does not handle details like daemonising itself, changing the user it runs as, etc. You will need to take care of these tasks by, e.g. integrating Molly Brown with your operating system’s init system. Some limited instructions on how to do this for common systems follows.
You can always use a tool like daemon
to take care of daemonising the Molly Brown process, changing the user
it runs as, chrooting it to a particular location, etc. You can call
/etc/rc.local (if your OS still supports it) to start
it on system boot.
An example systemd unit file for Molly Brown, named
molly-brown.service.example, can be found in the
directory of the Molly Brown source directory. After copying this
/usr/lib/systemd/system/molly-brown.service (consult your
system’s documentation for the appropriate choice) and making any
necessary changes for your environment, you can run the follow
commands as root to start Molly Brown and make sure it starts
automatically on system boot.
# systemctl daemon-reload # systemctl enable molly-brown.service # systemctl start molly-brown.service
An example OpenRC initscript for Molly Brown, named
molly-brown.openrc.example, can be found in the
directory of the Molly Brown source directory.
More detailed instructions on OpenRC setup are welcome!
An example OpenBSD initscript for Molly Brown, named
molly-brown.openbsd.example, can be found in the
directory of the Molly Brown source directory. After copying this
/etc/rc.d/mollybrownd, you can add the
daemon to your system startup with
rcctl or by manually adding
mollybrownd to your
/etc/rc.conf.local configuration. The
following lines in
rc.conf.local will autostart your
mollybrownd daemon as the user
Be sure that the user running your
mollybrownd daemon has
read access to
/etc/molly.conf and all of the files and
directories listed in
/etc/molly.conf. That user will
also need write access to the configured log file locations.
You can start your
mollybrownd daemon with
rcctl start mollybrownd
The following sections detail all the options which can be set in
/etc/molly.conf or any other configuration file specified with the
The format of the configuration file is
TOML, which bares some similarity
to the “INI” format. Remember that you can check
examples of the appropriate syntax.
Port: The TCP port to listen for connections on (default value
Hostname: The hostname to respond to requests for (default value
localhost). Requests for URLs with other hosts will result in a status 53 (PROXY REQUEST REFUSED) response.
CertPath: Path to TLS certificate in PEM format (default value
KeyPath: Path to TLS private key in PEM format (default value
DocBase: Base directory for Gemini content (default value
/var/gemini/). Only world-readable files stored in or below this directory will be served by Molly Brown.
HomeDocBase: Requests for paths beginning with
~/username/will be looked up relative to
users). Note that Molly Brown does not look inside user’s actual home directories like you may expect based on experience with other server software. Of course, you can symlink
/home/gus/public_gemini/if you want.
AccessLog: Path to access log file (default value
access.log, i.e. in the current wrorking directory). Note that all intermediate directories must exist, Molly Brown won’t create them for you. Set to
-for logging to
ErrorLog: Path to error log file (default value
error.log, i.e. in the current wrorking directory). Note that all intermediate directories must exist, Molly Brown won’t create them for you. Set to
-for logging to
GeminiExt: Files with this extension will be served with a MIME type of
MimeOverrides: In this section of the config file, keys are path regexs and values are MIME types. If the path of a file which is about to be served matches one the regexs, the corresponding MIME type will be used instead of one inferred from the filename extension.
DefaultLang: If this option is set, it will be served as the
langparameter of the MIME type for all
Molly Brown will automatically generate directory listings for
world-readable directories under
DocBase which do not contain an
index.gmi file. Only world-readable files and directories will be
listed. If a world-readable file named
.mollyhead is found in a
directory, it’s contents will be inserted above the directory listing
instead of the default “Directory listing” title.
The following options allow users to configure various aspects of the directory listing:
DirectorySort: A string specifying how to sort files in automatically generated directory listings. Must be one of “Name”, “Size” or “Time” (default value “Name”).
DirectoryReverse(boolean): if true, automatically generated directory listings will list files in descending order of whatever
DirectorySortis set to (default value false).
DirectoryTitles(boolean): if true, automatically generated directory listings will use the first top-level heading (i.e. line beginning with “# “) in files with an extension of
GeminiExtinstead of the filename (default value false).
TempRedirects: In this section of the config file, keys are regular expressions which the server will attempt to match against the path component if incoming request URLs. If a match is found, Molly Brown will serve a redirect to a new URL derived by replacing the path component with the value corresponding to the matched key. Within the replacement values, $1, $2, etc. will be replaced by the first, second, etc. submatch in the regular expression. Named captures can also be used for more sophisticated redirect logic - see the documentation for the Go standard library’s
regexppackage for full details.
PermRedirects: As per
TempRedirectsabove, but Molly Brown will use the 31 status code instead of 30.
Molly Brown supports dynamically generated content using an adaptation of the CGI standard, and also the SCGI standard.
stdout of CGI processes will be sent verbatim as the response to
the client, and CGI applications are responsible for generating their
own response headers. CGI processes must terminate naturally within
10 seconds of being spawned to avoid being killed. Details about the
request are available to CGI applications through environment
variables, generally following RFC 3875. In particular, note that if
a request URL includes components after the path to an executable
cgi-bin/script.py/foo/bar/baz) then the environment variable
SCRIPT_PATH will contain the part of the URL path mapping to the
/var/gemini/cgi-bin/scripty.py) while the variable
PATH_INFO will contain the remainder (e.g.
It is very important to be aware that programs written in Go are unable to reliably change their UID once started, due to how goroutines are implemented on unix systems. As an unavoidable consequence of this, CGI processes started by Molly Brown are run as the same user as the server process. This means CGI processes necessarily have read and write access to the server logs and to the TLS private key. There is no way to work around this. As such you must be extremely careful about only running trustworthy CGI applications, ideally only applications you have carefully written yourself. Allowing untrusted users to upload arbitrary executable files into a CGI path is a serious security vulnerability.
SCGI applications must be started separately (i.e. Molly Brown expects them to already be running and will not attempt to start them itself), and as such they can run e.g. as their own user and/or chrooted into their own filesystem, meaning that they are less of a security threat than CGI applications (in addition to avoiding the overhead of process startup, database connection etc. on each request).
CGIPaths: A list of filesystem paths, within which world-executable files will be run as CGI processes. The paths act as prefixes, i.e. if
/var/gemini/cgi-binis listed then
/var/gemini/cgi-bin/subdir/subsubdir/script.pywill both be run. The paths may include basic wildcard characters, where
?matches a single non-separator character and
*matches a sequence of them - if wildcards are used, the path should not end in a trailing slash
SCGIPaths: In this section of the config file, keys are URL path prefixes and values are filesystem paths to unix domain sockets. Any request for a URL whose path begins with one of the specified prefixes will cause an SCGI request to be sent to the corresponding domain socket. Anything sent back from a program listening on the other end of the socket will be sent as the response to the client. SCGI applications are responsible for generating their own response headers.
Molly Brown allows you to use client certificates to restrict access
to certain resources (which may be static or dynamic). The overall
workflow is highly reminiscent of OpenSSH’s
CertificateZones: In this section of the config file, keys are path regexs and values are lists of hex-encoded SHA256 fingerprints of client certificates. Any requests whose path matches one of the regexs will only be served as normal if the request is made with a client certificate whose fingerprint is in the corresponding list. Requests made without a certificate will cause a response with a status code of 60. Requests made with a certificate not in the list will cause a response with a status code of 60.
In order to allow users of shared-hosting who do not have access to
the main Molly Brown configuration file to customise some aspects of
their Gemini site, Molly Brown features functionality much like
.htaccess files. If the main configuration file contains
ReadMollyFiles = true, then each directory in the path to a
resource will be checked for a file named
.molly. These files
should be in exactly the same format as the main configuration file,
an their contents will override (some) settings from the main file.
.molly file will override settings specified in
from higher directories.
E.g. when handling a request which maps to
/var/gemini/.molly, if it exists, will override those in
/var/gemini/foo/.molly, if it exists, will override those in
/var/gemini/foo/bar/.molly, if it exists, will override those in
/var/gemini/foo/bar/baz/.molly, if it exists, will override those in
Only the following settings can be overriden by
.molly files. Any
other settings in
.molly files will be ignored:
Margaret Brown was an American philanthropist and socialite who survived the sinking of the RMS Titanic, leading to a Broadway musical and later a film about her life being titled “The Unsinkable Molly Brown”. The “unsinkable” moniker inspired NASA astronaut Gus Grissom to name the Gemini 3 capsule he commanded “Molly Brown” - Grissom had almost drowned a few years earlier when his Mercury 4 capsule “Liberty Bell” sank after splashdown.