reimeika

login

the koi cms

introduction

koi is a content management system (CMS) written in python (3.8.2+) using the bottle microframework.

The koi CMS uses JSON files to store information. It can be fully curated and managed from the back-end command-line, but for convenience it also features a simple web-based markdown article editor which is additionally able to manipulate image galleries (an anti-CSFR mechanism is built-in, as are anti-XSS measures). A search functionality is included, as is multi-user support with optional email-based two-factor authentication.

To use koi to its fullest extent from the start the following should be run (as root, tested on Ubuntu 20.04):

apt install python3-markdown python3-whoosh python3-passlib python3-bleach \
python3-pil python3-natsort
local installation

koi can be downloaded from reimeika.ca. Unzip the file koi.zip and type cd koi. The file config.py contains detailed explanations of all configuration options and should be reviewed. In particular session_cookie_sig must be set. Once this is done, running ./koi.py will make this tutorial accessible at http://localhost:8080.

serving files

By default web pages are in the pages directory. To create a new web page first make a sub-directory test and copy a file into it, e.g. from inside the koi directory:

mkdir pages/test
cp logo.txt pages/test

The file is now located at http://localhost:8080/pages/test/logo.txt but clicking the link will complain about a missing .koi file and return a 403 error. In order to serve requests every page must have an associated template in the dir_templates directory. Which template is used is determined by the name of a JSON .koi file inside the web page directory so that, for example, article.koi will use the template file article.tpl. The most basic template is files.tpl and the simplest document possible is storing the string {} into a file called files.koi within test/ as so:

echo '{}' > pages/test/files.koi

Once this is done the files.tpl template will be used to serve the logo.txt file which can now be retrieved by clicking the link above. Note that, for extra security, the directive force_acl can be set to True, in which case all pages must have a properly-defined ACL if they are to be served. If that's the case the above command should be replaced by:

echo '{"acl": {"files.koi": {"users": "*", "groups": [], "ips": "*", "time": 0}}}' > pages/test/files.koi
creating a web page

Although the files inside test can be now served, the web page http://localhost:8080/pages/test itself is empty. The files.tpl template can display the contents of a key called body, so replacing the empty JSON object with {"body": "Hello world!"} in files.koi will now show a "Hello world" web page at the above link:

echo '{"body": "Hello world!"}' >! pages/test/files.koi

Editing JSON files is not very practical and so koi includes a simple web-based markdown article editor. koi articles can also be created/edited from the command-line using the koiedit.py script by simply creating a directory/slug (with the appropriate permissions) and running (if in the koi top-level directory):

mkdir pages/<slug>
./koiedit.py pages/<slug>/article.koi
./acledit.py pages/<slug>/article.koi

Also included is the html2koi.py script, which allows to convert existing HTML files into .koi format.

searching

The default search engine is powered by whoosh which is a non-standard but readily-available module which might be have to be installed (as root), as well as markdown in order to parse most pages:

apt install python3-markdown python3-whoosh

Main features of the search function are:

  • query words can be separated by AND (the default), OR, NOT, ANDNOT and ANDMAYBE
  • fuzzy queries e.g. grafiti~ will find graffiti (see more)
  • phrase search using double quotes
  • field searches on title, body, keywords, author, and creator

A complex search could be crafted as follows:

author:yuma OR grafiti~ ANDNOT title:lavender

The index is dynamically updated any time a new article or page is added, deleted, or modified. Note that both this and the simple search engine (see below) filter their results based on access control lists (ACLs) so that matches of restricted pages are not shown unless the user (or visitor) performing the search has access to them.

Only the title, body, keywords, author, and creator fields of .koi files are indexed. Thus, a text file upload paper.txt will not be found in a search even if a keyword matches within. Furthermore, the index will only be updated if a .koi file timestamp field has been properly set. The search index can be regenerated at any time by deleting the cache (stored in search/.index by default).

adding users

The basic functionality for creating websites through the back-end is to simply write templates and add content via their corresponding .koi JSON file. A simpler approach is to use the built-in article editor. However, in order to do so users must be added to the system. Note that this requires passlib so running apt install python3-passlib may be necessary.

For increased security all user management is deliberately done from the back-end, although a template could in principle be written to accomplish this over the web. Running the script ./accounts.py offers a simple menu-driven command-line interface which allows, amongst other things, adding, deleting, listing, and modifying user accounts. Of note is the fact that the hashing algorithm is compatible with the Linux /etc/shadow file, and thus offers the possibility of importing existing user accounts. User accounts are stored in JSON files inside the directory specified by dir_accounts_fp in config.py so it's important to review this setting before proceeding.

Adding an account is straightforward and can be done following the steps set by the script. User names are case-insensitive and restricted by the user_re configuration setting, which by default allows alphanumeric strings and email addresses in any language (so that "AIKA", "sora@remeika.ca", and "ゆま" are all valid, albeit not necessarily a prudent mix). If desired an email for two-factor authentication may be recorded, but care must be taken to configure the format of the message (twoF_msg) and SMTP settings in config.py.

Listing a user's profile via accounts.py shows the basic structure of the account and some information about their last login, logout, IP used, "locked" status, roles, and groups they belong to. Note that "roles" consists of lists that can be assigned on a per-template basis, and so having the "editor" role for the "edit" template allows authoring articles via the web browser as explained in the next section.

editing articles

By default koi has no users at all. Assuming a user has been created using accounts.py they need to have an "editor" role to edit articles. Note that the editor requires the bleach module to operate so apt install python3-bleach may be necessary.

Once a user has been created they can log in at http://localhost:8080/pages/login. If a two-factor email has been recorded a numerical token will be required to validate the login. After authenticating users can click on the editor button (which is only shown to editors) to create a new page or see a listing of the pages they can edit and delete (hovering over an entry provides details about the web page). Only pages using the article.tpl or gallery.tpl templates are supported.

The editor is mostly self-explanatory. Articles are written in markdown (help for which is available from the collapsed section at the bottom of the editing page) and consist of a title, a slug (see below), a space-separated list of keywords, and a body (most elements will provide a helpful pop-up if hovered on). Controls for editing an article are:

  • webpage opens the current web page in a new tab (must be refreshed after saving edits)
  • files file manager to review and delete files, and also manage ACLs (see below)
  • uploader for uploading files unto the web page
  • articles to return to the article listing
  • save to save the current article

Files (including images) can be linked and embedded in web pages using markdown syntax (the file manger provides the link/embed code for each file which can be copied and pasted into the article).

By default web pages are stored in directories named after the creation time-stamp e.g. /pages/1593798867, but can be renamed using the slug field after first saved. All article revisions are stored in hidden files as .article.koi-rev# (where rev# is the revision number) within the web page, so it's feasible to recover prior versions (though only from the back-end). Deleted articles and their files are also backed-up as hidden directories (e.g. /pages/.name-rev#-timestamp), although individually-deleted files are permanently removed.

curating a gallery

The editor offers a second function: that of creating image galleries. This feature requires PIL, and optionally natsort, and hence apt install python3-pil python3-natsort may be necessary. The default image formats supported are .jpg and .png.

As with articles, galleries have a title, a slug, and keywords. Two viewing modes are available, grid (overview of the entire image set) and slide (single picture view, with a ▸ Details section at the bottom offering extra image information). By default a new gallery has the same curator (editor) and ACL permissions as new articles i.e. it's restricted to the user who created it.

After creating a new gallery images can be uploaded to the page when in grid view. The ACL (see next section) of individual images can only be set from the back-end, but a show/hide mechanism is available. When an image is uploaded it is automatically tagged as hidden. In grid and slide views these images look slightly washed-out to the curator, and visitors will need to append ?unhide at the end of the gallery URL to view them. A toggle button can be found under each image in slide view, and so a curator can control what images are always shown and which ones require the special URL. The unhide_tag is configurable in edit.koi from the back-end, as is the default state of the images upon upload (currently all hidden unless hide_new is set to False). Keep in mind that hiding images does not block a direct link to the image file (an ACL would be needed for that), but it provides a simple filter depending on the link followed to the gallery.

The last feature of the editor is ACL manipulation, explained in the section below.

access control lists (acls)

Articles and galleries have two types of restrictions: who can edit/curate them and who can view them. Newly-created articles and galleries can only be modified and viewed by their original author, but other editors can be added by clicking on the ▸ Access control list for <name> (or ▸ gallery ACL in grid view) expandable section and modifying the list of users who can edit the page (note that it's impossible to remove oneself). Adding a wildcard * allows any logged-in user to edit the article/gallery. Group-based editing controls are not currently implemented, nor is blocking with the ! prefix (see below). ACL controls for articles and galleries are equivalent, so any reference to articles below applied also to galleries.

The access control list for viewing articles can restrict access to web pages and files on a per-user, per-IP, and date-time basis (per-group is also supported but only through the back-end). This limits who can view a web page or download a file when clicking on a link. By default new articles can only be accessed by the user who originally wrote them, from any IP, starting from the creation date-time. To make an article universally available suffice to make the user ACL equal to *, the IP ACL equal to *, and leave the timestamp as-is. Other ACL features are:

  • Adding a * to the user list will give access permissions to all logged-in users
  • Subnets (possibly in combination with an IP address list) can be specified as xxx.yyy
  • Setting a future release date will only allow access from that date-time onward
  • To block, prepend ! to a user or IP/subnet (overrides any conflicting allow directive)

Access control lists can be manipulated in exactly the same way from the file manager on a per-file basis (by default uploaded files inherit the ACL of the web page). Galleries only offer a per-image ACL via the back-end.

trusted articles, making forms

To thwart XSS attacks user input is sanitized using bleach and context-based allowlists, and further escaped upon display unless used in an HTML context. This, however, strips most HTML code which is sometimes undesirable. koi supports the concept of trusted articles which allow full HTML-editing using the web editor. This setting can be toggled via the back-end by changing trusted to True in the .koi file, which in turn will add the tag "(trusted)" next to the ACL section near the top of the article editor. It is strongly encouraged that only select editors be allowed to modify such articles.

Forms can then be included in templates as well as trusted .koi files (or .html files converted using html2koi.py). koi provides an anti-CSFR measure via a token which is tied to the user session and which must be used when making a submission (in fact, koi forbids logged-in users to submit any data — including uploads — unless a valid anti-CSFR token is present).

For example, the code to add the following search field to an article:

is:

<form action="/pages/search" method="post">
    {{!×CSRF}}
    <input name="search_query" size="12" type="text">
    <button class="button" type="submit">search</button>
</form>

depending on how various variables have been defined in config.py i.e. search_query is actually the value of CONFIG['search_var'] (note that "x" has been replaced with "×" for the anti-CSRF variable name, otherwise it'd be replaced in the snippet above). The anti-CSRF measure must be coded in the template, which in the case of article.tpl is done as so:

% if user := PROFILE.get('user', ''):
%   xCSRF = f'<input name="xCSRF" type="hidden" value="{PROFILE["xCSRF"]}">'
% else:
%   xCSRF = ''
% end
% content = PAGE['body'].replace('{{!×CSRF}}', xCSRF)
templates

koi includes a few templates in the templates directory (configurable via dir_templates) which can be studied for reference. Other than the special login.tpl, all templates are provided the following dictionaries:

  • BOTTLE: the WSGI environment
  • CONFIG: all parameters defined in config.py
  • INDEX: the website index, each key being the page name and the corresponding .koi data
  • FORMS: all form values, decoded
  • ME: properties of the current page
  • PAGE: the .koi JSON dictionary
  • PROFILE: the current user's JSON profile (an empty dictionary if no session is ongoing)
  • QUERY: the combined GET and POST dictionary, decoded
  • TREE: the website tree, each key being the page name with the equivalent ME dictionary
  • UPLOAD: a dictionary of uploaded files (with number keys 0 up to upload_max_files)
  • USERS: overview of all users, each key being a user name and its PROFILE dictionary

The main difference between FORMS and QUERY is that the latter returns a dictionary with one key per form input while the former retrieves a FormsDict which can contain repeated keys with different values as a list (resulting, for example, from a selector element). A form with an input named animal can have its value retrieved as so:

QUERY['animal'] returns "lion"

while the input from a selector element named colours can be obtained as so:

FORMS.getall('colours') returns "['red', 'green', 'yellow']"

Detailed dictionary keys:

  • ME: page, path, koi_fp, template, uri, and files (a list of all non-hidden files/dirs in the directory)
  • PROFILE:
    • auth: hash, 2f_email
    • id: user, uid, groups, name, roles
    • session: ip, token, nonce, xCSRF, login, logout, locked
    • misc: koi_version
    • unused: data
  • TREE[page]: path, template, uri, files, and serve (which is True if the current user can access the page)
  • UPLOAD[N]: OK, status, content_type, raw_filename, safe_name, and the upload in file_data.
  • USERS[user]: same as PROFILE

Note that for performance reasons INDEX and TREE are only provided if get_index is set to True in the .koi file, and similarly USERS is only provided if get_users is also True. Otherwise the dictionaries are present but empty (as is UPLOAD if no uploads are found). User names have their case preserved, but should be lower-cased for internal usage in templates i.e. use user.lower().

uploads

The following code adds an upload form/button in a template (note that the names action and file_upload are reserved in koi and must be used in the form below):

<form id="upload" action="{{ME['uri']}}" enctype="multipart/form-data" method="post">
    {{!×CSRF}}
</form>
<input form="upload"  name="file_upload" type="file" multiple>
<input form="upload" class="button-primary" name="action" type="submit" value="upload">

Uploads are stored in an UPLOAD dictionary with numerical entries UPLOAD[0], UPLOAD[1] ... UPLOAD[config.upload_max_files-1]. They can be stored via a template using the following code:

% for key, up in UPLOAD.items():
%    if up['OK']:
%       with open(os.path.join(ME['path'], up['safe_name']), 'wb') as fd:
%          fd.write(up['file_data'])
%       end
%    end
% end

Change ME['path'] to save the file in a different directory, and check for I/O errors as needed.

scripts and back-end usage

As mentioned earlier in this guide, koi can be readily used and managed from the back-end (which offers much more control and security at the expense of convenience). No users or editors are really necessary, and all GUI components can be disabled via ACL restrictions. For this purpose the following scripts are provided inside the koi directory:

  • accounts.py: add, delete, modify and list user accounts with batch-import support
  • acledit.py: manipulate a page's access control list
  • html2koi.py: convert an HTML file into an article.koi file
  • koiedit.py: edit the contents of a .koi file using a text editor
  • mkgallery.py: create a gallery.koi file

For example, to create a new gallery from the back-end the following procedure would be followed (starting from the koi directory). First, create the gallery:

./mkgallery.py

Set proper ACL permissions:

./acledit.py

Move gallery.koi to its page directory and populate the gallery:

mkdir pages/portfolio
mv gallery.koi pages/portfolio
cp -p /mnt/camera/*.jpg pages/portfolio/

Upon visiting the page for the first time at http://localhost:8080/pages/portfolio?unhide the gallery will automatically be generated (and updated if images are added or deleted).

apache configuration

To run koi on an apache web server via mod_wsgi it may be necessary to run:

apt install libapache2-mod-wsgi-py3

Assuming a working SSL-enabled web server is already running, the first step is to move the koi directory into a suitable location, say /www/wsgi, and modify the file koi.wsgi to set sys.path accordingly e.g.

sys.path = ['/www/wsgi/koi/'] + sys.path

Ownership (perhaps www-data) and permissions of files (600) and directories (700) should be reviewed, as should the settings in config.py, particularly dir_accounts_fp (say, /usr/local/etc) and a new session_cookie_sig. Setting force_ssl to True is highly recommended, and care should be taken not to run in DEBUG mode.

Adding the following code to an ssl VirtualHost may then suffice:

DocumentRoot /www/wsgi/koi
<Directory /www/wsgi/koi>
    Options None
    AllowOverride None
    Require all granted
</Directory>
WSGIProcessGroup koi
WSGIDaemonProcess koi user=www-data group=www-data
WSGIScriptAlias / /www/wsgi/koi/koi.wsgi

Remember to touch koi.wsgi after adding a new template or making changes to existing ones in order to update the cache.

odds and ends

— Since koi can be fully managed from the back-end it may be desirable (for enhanced security, say) to disable login access. This can be achieved by hiding the login page (or whatever page_login is set to in config.py):

mv pages/login pages/.login

and then editing config.py appropriately e.g. page_login='.login'. Editing and searching can also be disabled in a similar manner (note that login, edit, and search can each be toggled this way independently of each other).

— URL slugs in the article editor are, by default, restricted to the following regular expression in config.py: r'^[a-zA-Z0-9][a-zA-Z0-9_-]{0,75}$'. This allows for a mix of up to seventy-six underscores, dashes, and alphanumeric ASCII characters. This rule is not enforced by the koi core (which doesn't rely on the editor) but by the template edit.tpl. Another example which disallows some misleading page names could be r'(?i)^(?!home|download|admin)[a-z0-9][a-z0-9_-]{0,75}$'.

— koi includes two search engines, a simple one in template ssearch.tpl and the more advanced wsearch.tpl. Which one is used is a simple matter of making a symlink of the preferred template file to search.tpl. The simple search engine has no external dependencies and requires no index, but does no more than a case-insensitive AND search of the submitted words with no concept of query analysis or scoring. The whoosh-based engine wsearch.tpl is the default.

— If error messages are deemed too informative they can be tweaked in the error.tpl template, in particular the following snippet:

% if CODE in [400, 403, 404, 413, 500]:
    <p><font class="error">[{{CODE}}] {{DETAILS}}</font></p>
% end

can be customized as desired (by, say, getting rid of {{DETAILS}} in extreme cases).

— .koi files can always be (carefully) edited directly using a text editor, or from within the python interpreter (but care must be taken not to delete standard fields which may be required by the editor):

import json
with open("pages/article/article.koi", "r") as fd:
    art = json.load(fd)

... do stuff to "art" ...

with open("pages/article/article.koi", "w") as fd:
    json.dump(art, fd, ensure_ascii=False)

— For advanced users who prefer using their text editor of choice, the best way to edit .koi files (article files in particular) is to use the included koiedit.py script. The editor of choice can be set via the EDITOR variable at the top of the script. The default value is emacs -nw, which combined with the "Save Place" directive ((save-place-mode 1) in the .emacs file) allows to conveniently cycle between editing, saving/quitting, and reloading the file in a browser. The koiedit.py command can take the .koi file as an argument.

— Symlinked files will only be served within the page directory (and sub-directories).

— While multi-lingual web page content should be fine, creating non-ASCII slugs and file names from the back-end will likely cause issues and should be avoided.

— An example of a complex ACL:

{"article.koi":{"users": "*", "groups": [], "ips": "*", "time": 0},
 "araara.doc": {"users": ["クレア", "*"], "groups": [], "ips": ["8.8.8.8"], "time": 0}
 "kira.tex":   {"users": ["AIKA"], groups: ["av"], "ips": ["128.100"], "time": 0}
 "bday.pdf":   {"users": ["yuma", "!sora"], "groups": ["sod"], "ips": "*", "time": 0}
 "shibu.jpg":  {"users": ["*"], "groups": ["!moodyz"], "ips": "*", "time": 1604973574}}

(so that, for example, the file shibu.jpg can be downloaded by any logged-in user after 2020-11-10T01:59:34+00:00 who does not belong to the moodyz group).

These ACLs are not exclusive to articles and can be applied to any web page regardless of the template (for example, the login page, or the search engine). The lack of an ACL entry in the .koi file is equivalent to making the web page and its files available to anyone.

colophone

koi and documentation is released under the 3-clause BSD license

bottle is distributed under the MIT license

CSS used is a customized version of skeleton, distributed under the MIT license

The koi logo is public domain