diff --git a/README.md b/README.md index 13af995..58c509e 100644 --- a/README.md +++ b/README.md @@ -280,6 +280,7 @@ Every setting the code reads, with its default: | `SWW_WAF_EXEMPT_PATHS` | empty | Path prefixes the Core Rule Set does not inspect. | | `SWW_WAF_BODY_LIMIT` | `off` | How much of a request body the Core Rule Set reads, a size up to `64K`; `off` reads none. | | `SWW_CORE_RULE_SET_EXCLUSIONS_FILE` | unset | A file of Core Rule Set exclusions, as Coraza directives, that switch rules off for some fields or paths only. | +| `SWW_APPS` | unset | The apps each hostname serves, `drupal`, `gitea`, `mediawiki` or `wordpress`, such as `git.example.com:gitea;blog.example.com:wordpress`; only those apps' fixes apply there. Unset, every app's fixes apply to every hostname. | | `SWW_BLOCKLIST_URLS` | empty | Blocklists to fetch, text files of addresses and netblocks one to a line, such as `https://www.spamhaus.org/drop/drop.txt`. | | `SWW_BLOCKLIST_REFRESH` | `24h` | How long after a list was last tried it is fetched again, `1h` or more. | | `SWW_BLOCKLIST_ACTION` | `deny` | What a client on a blocklist gets: `deny`, `limit:` of every limit, or `log`. | @@ -581,22 +582,24 @@ client's limits by tag. See "Request tags" in [`SPEC.md`](SPEC.md). headers of every request, and no body unless `SWW_WAF_BODY_LIMIT` is set. In front of a code forge, bodies carry what people write and publish, which the rules take for attacks. -- `smallwebwaf` changes the Core Rule Set in eleven ways so that ordinary - requests get through to a code forge, to WordPress, to Drupal and to - MediaWiki. They include git over HTTP and sign-in through Git Credential - Manager on a code forge, and saving a post from the block editor on WordPress. - On gitea's own paths, its searches, file names and branch names get through - too. On WordPress's own paths, its searches, names and file names do. On - Drupal's own paths, the page to return to, its searches and its names get - through too. On MediaWiki's own paths, its searches, page titles, names and - wikitext do. No setting undoes the changes; "Attack detection" in +- `smallwebwaf` changes the Core Rule Set in six ways so that it lets ordinary + requests to a code forge through, such as git over HTTP and sign-in through + Git Credential Manager. No setting undoes the changes; "Attack detection" in [`SPEC.md`](SPEC.md) lists them. -- These apps work out of the box at the default settings, and their recorded - ordinary requests pass the Core Rule Set: - - gitea; - - WordPress; - - Drupal; - - MediaWiki. +- Each supported app also has its own fixes, which let its ordinary requests + through. Its recorded ordinary requests pass the Core Rule Set at the default + settings. The supported apps, by their names in `SWW_APPS`, are: + - `gitea`, whose fixes let its searches, file names and branch names + through; + - `wordpress`, whose fixes let through its searches, its names, its file + names and a post saved from the block editor; + - `drupal`, whose fixes let through the page to return to, its searches and + its names; + - `mediawiki`, whose fixes let through its searches, its page titles, its + names and its wikitext. +- `SWW_APPS` names the apps each hostname serves. A hostname gets only its own + apps' fixes, and a hostname it does not list gets none. Unset, every app's + fixes apply to every hostname. - `SWW_CORE_RULE_SET_EXCLUSIONS_FILE` switches single rules off for an app's own fields or paths. - Rule files are plain text, one rule to a line, edited by hand and read again diff --git a/SPEC.md b/SPEC.md index 72f3f09..c886798 100644 --- a/SPEC.md +++ b/SPEC.md @@ -81,8 +81,8 @@ to them. of bans is full. - `internal/rules`: reads the rule files at start and again when they change, and tells which rules a request matches. - - `internal/coreruleset`: the Core Rule Set, with the eleven changes listed - under "Attack detection", as Coraza directives. + - `internal/coreruleset`: the Core Rule Set, with the six changes and the + apps' fixes listed under "Attack detection", as Coraza directives. - `internal/lookup`: looks clients up through GeoJS or in the lookup database. - `internal/reputation`: fetches the lists, asks the DNS blocklists and @@ -676,8 +676,8 @@ files, the limits and the bans still apply to the client that sends it. - `SWW_CORE_RULE_SET_EXCLUSIONS_FILE` names a file of exclusions, written as Coraza directives. An exclusion switches one rule off for some fields or paths only, where `SWW_WAF_DISABLED_RULES` switches it off everywhere. The file - holds an app's own exclusions; the eighth to eleventh changes below hold - gitea's, WordPress's, Drupal's and MediaWiki's. + holds an app's own exclusions; the apps' fixes below hold gitea's, + WordPress's, Drupal's and MediaWiki's. - `discriminator` loads the file after `smallwebwaf`'s own setup rules and before the Core Rule Set's rules. The Core Rule Set loads its own `REQUEST-900-EXCLUSION-RULES-BEFORE-CRS.conf` at the same point. @@ -723,9 +723,9 @@ files, the limits and the bans still apply to the client that sends it. scanner from an ordinary client. In front of a code forge, registry and git clients get `404` and `401` by design, hundreds of times a minute. -`smallwebwaf` changes the Core Rule Set 4.25.0 in eleven ways, since in front of -gitea, WordPress, Drupal and MediaWiki it would otherwise refuse ordinary -requests. The changes hold in front of every app, and no setting undoes them: +`smallwebwaf` changes the Core Rule Set 4.25.0 in six ways, since in front of +gitea it would otherwise refuse ordinary requests. The changes hold in front of +every app, and no setting undoes them: 1. `PUT`, `PATCH` and `DELETE` are allowed methods besides `GET`, `HEAD`, `POST` and `OPTIONS`. APIs, container image pushes and package uploads use them. @@ -742,8 +742,52 @@ requests. The changes hold in front of every app, and no setting undoes them: and tea, which gitea registers for OAuth sign-in, ask to be sent back to `http://127.0.0.1` on the user's own machine. The server never fetches that address. -4. Ten query parameters are not checked against the Core Rule Set's lists of - system files (930120), shell paths (932160) and command names (932260): +4. `Referer` is not checked for a Unix command given without arguments (932340), + for Java starting a process (944110) or for a Java class name (944130). + `Referer` names the page a request comes from, on everything that page loads + and on every link followed from it. The Core Rule Set takes some page + addresses for attacks: + - a search for `env`; + - OpenJDK's `Runtime.java`, whose path holds both `runtime` and `java.`; + - a file named `java.lang.String.md`. + + Every request from such a page would be refused. `Referer` keeps the other + rules, except after some gitea searches (gitea's fixes). An app that passes + `Referer` to a shell must check it itself. + +5. Responses are not inspected. A raw file from a repository, such as a shell + script, looks to the response rules like source code leaking from the server. +6. The `from` parameter of a container registry's upload path, + `/v2//blobs/uploads/`, is not checked for command names (932260). When + a push reuses a layer, `from` names the image the layer comes from. An image + such as `owner/python3-app` would have that push refused. + +Each supported app also has its own fixes, which let its ordinary requests +through. `SWW_APPS` says which hostnames get which app's fixes: + +- `SWW_APPS` holds entries separated by semicolons. Each is a hostname, a colon, + then one or more apps separated by commas, for example + `git.example.com:gitea;blog.example.com:wordpress`. +- The request's `Host` is matched by exact name, without its port and without + regard to case. There are no wildcards. +- A hostname not listed gets no app's fixes, so every rule checks it in full. + Set empty, `SWW_APPS` lists no hostname. +- Unset, every app's fixes hold for every hostname. An app's fixes then weaken + the check of any other app that serves the same paths or fields. +- Any of these mistakes stops the start: + - an unknown app; + - a hostname that is not a DNS name, such as `*.example.com`; + - a hostname listed twice; + - an app listed twice for one hostname; + - an entry without a colon. + +The apps supported are `drupal`, `gitea`, `mediawiki` and `wordpress`. Each +app's fixes are in one file, such as `internal/coreruleset/gitea.conf`, and one +entry of `apps` in `internal/coreruleset/coreruleset.go`. gitea's fixes are: + +1. Ten query parameters are not checked against the Core Rule Set's lists of + system files (930120), shell paths (932160) and command names (932260), on + any path: - `path`, `files`, `skip-to` and `sub_path`; - `ref`, `sha`, `branch` and `workflow`; - `artifactName` and `redirect_to`. @@ -756,42 +800,19 @@ requests. The changes hold in front of every app, and no setting undoes them: catch reaches the app, such as the command `|cat /etc/passwd`. Path traversal, SQL and script injection, and PHP, Java and Node.js code are still refused there. Other parameters keep all three rules, except on the - gitea, WordPress, Drupal and MediaWiki paths the eighth to eleventh changes - name. An app that uses one of these parameters as a file on the server, or - passes it to a shell, must check the value itself. A rule file can refuse it - instead. + paths named in gitea's third fix, in WordPress's fixes, in Drupal's fixes + and in MediaWiki's fixes. -5. Two cookies are left out, and `Referer` keeps fewer rules: - - The Core Rule Set reads the request without the `gitea_flash` and - `redirect_to` cookies. gitea puts a message naming what was just done into - `gitea_flash`. It puts into `redirect_to` the page to return to after - signing in. The Core Rule Set takes file names in them, such as - `package.json`, for attacks. A browser sends such a cookie with every - request until gitea replaces it. gitea cannot replace it while those - requests are refused, so one refusal would keep the browser out of the - whole site. gitea shows the message with any script removed and returns - only to its own pages. Every other cookie is read in full. An app that - reads a cookie of either name must check it itself. - - `Referer` is not checked for a Unix command given without arguments - (932340), for Java starting a process (944110) or for a Java class name - (944130). `Referer` names the page a request comes from, on everything - that page loads and on every link followed from it. The Core Rule Set - takes some page addresses for attacks: - - a search for `env`; - - OpenJDK's `Runtime.java`, whose path holds both `runtime` and `java.`; - - a file named `java.lang.String.md`. - - Every request from such a page would be refused. `Referer` keeps the - other rules, except after some gitea searches (the eighth change). An - app that passes `Referer` to a shell must check it itself. - -6. Responses are not inspected. A raw file from a repository, such as a shell - script, looks to the response rules like source code leaking from the server. -7. The `from` parameter of a container registry's upload path, - `/v2//blobs/uploads/`, is not checked for command names (932260). When - a push reuses a layer, `from` names the image the layer comes from. An image - such as `owner/python3-app` would have that push refused. -8. On gitea's paths, some fields skip the rules that refused ordinary gitea +2. The Core Rule Set reads the request without the `gitea_flash` and + `redirect_to` cookies. gitea puts a message naming what was just done into + `gitea_flash`. It puts into `redirect_to` the page to return to after signing + in. The Core Rule Set takes file names in them, such as `package.json`, for + attacks. A browser sends such a cookie with every request until gitea + replaces it. gitea cannot replace it while those requests are refused, so one + refusal would keep the browser out of the whole site. gitea shows the message + with any script removed and returns only to its own pages. Every other cookie + is read in full. +3. On gitea's paths, some fields skip the rules that refused ordinary gitea requests there. They are the fields where gitea sends a search, a file name or a branch name. `internal/coreruleset/gitea.conf` lists each rule and the request it refused. The fields are: @@ -809,8 +830,7 @@ requests. The changes hold in front of every app, and no setting undoes them: - packages. A user's or an organisation's page is any path of one segment, so on - every such path, on any app, `q` skips these three rules, for example on - `/search`. + every such path `q` skips these three rules, for example on `/search`. Some of these searches also skip the rule for Java class names (944130), which refused `org.apache.commons.logging`. @@ -848,231 +868,230 @@ requests. The changes hold in front of every app, and no setting undoes them: `actions/upload-artifact@v3` (920450). Elsewhere these fields keep every rule. The paths are those of gitea at the - site root. An app other than gitea that serves one of them must check these - fields itself. + site root. -9. On WordPress's paths, some fields skip the rules that refused ordinary - WordPress requests there. `internal/coreruleset/wordpress.conf` lists each - rule and the request it refused. The REST API has four addresses: - - `/wp-json/...` - - `/index.php/wp-json/...` - - `/index.php?rest_route=/...` - - `/?rest_route=/...` +WordPress's fixes: on WordPress's paths, some fields skip the rules that refused +ordinary WordPress requests there. `internal/coreruleset/wordpress.conf` lists +each rule and the request it refused. The REST API has four addresses: - The fields are: - - The header `X-HTTP-Method-Override` on any address of the REST API - (920450). The block editor saves a post with `POST` and names `PUT` in - this header. - - Three groups of fields skip the rules on system files and commands - (930120, 932120, 932250, 932260). The fields hold searches, names and - addresses of WordPress's own pages, and any of them can hold any word. The - groups are: - - On all of `/wp-admin/` and on `/wp-login.php`: - - the searches `s`, `search`, `q` and `term`; - - the names `tag`, `category_name`, `plugin`, `checked[]`, `theme` - and `stylesheet`; - - the name of a plugin's settings page in `page`; - - the addresses `url`, `return`, `redirect_to`, `wp_http_referer` - and `_wp_http_referer`. - - On every REST API route under `/wp/v2/` and on the route - `/oembed/1.0/embed`: `search`, `term`, `slug` and `url`. - - On `/`, `/index.php`, `/page/N/` and `/index.php/page/N/`: `s` and - `tag`. +- `/wp-json/...` +- `/index.php/wp-json/...` +- `/index.php?rest_route=/...` +- `/?rest_route=/...` - The addresses of the first group also skip the rule on Unix commands - (932235), because an address repeats the search it came from. The other - fields of the groups keep that rule. +The fields are: - - On all of `/wp-admin/` and on `/wp-login.php`, these names skip the lists - of PowerShell commands and command names (932120, 932260): - - the name of a template in `p` and `postId` of the site editor; - - the name of a plugin to install in `slug`. +- The header `X-HTTP-Method-Override` on any address of the REST API (920450). + The block editor saves a post with `POST` and names `PUT` in this header. +- Three groups of fields skip the rules on system files and commands (930120, + 932120, 932250, 932260). The fields hold searches, names and addresses of + WordPress's own pages, and any of them can hold any word. The groups are: + - On all of `/wp-admin/` and on `/wp-login.php`: + - the searches `s`, `search`, `q` and `term`; + - the names `tag`, `category_name`, `plugin`, `checked[]`, `theme` and + `stylesheet`; + - the name of a plugin's settings page in `page`; + - the addresses `url`, `return`, `redirect_to`, `wp_http_referer` and + `_wp_http_referer`. + - On every REST API route under `/wp/v2/` and on the route + `/oembed/1.0/embed`: `search`, `term`, `slug` and `url`. + - On `/`, `/index.php`, `/page/N/` and `/index.php/page/N/`: `s` and `tag`. - WordPress writes these names as slugs, such as `htaccess` and `ls-la`. + The addresses of the first group also skip the rule on Unix commands + (932235), because an address repeats the search it came from. The other + fields of the groups keep that rule. - - `url` on the REST API routes `/wp-block-editor/v1/url-details` and - `/oembed/1.0/proxy`, for the lists of system files and PowerShell commands - (930120, 932120). The block editor's link previews and embeds send an - address in this field. +- On all of `/wp-admin/` and on `/wp-login.php`, these names skip the lists of + PowerShell commands and command names (932120, 932260): + - the name of a template in `p` and `postId` of the site editor; + - the name of a plugin to install in `slug`. - - `_wp_http_referer` on `/wp-admin/upload.php`, for a Unix command given - without arguments (932340). The media list's own address ends in `&ps`. - - `file` on `/wp-admin/plugin-editor.php` and `/wp-admin/theme-editor.php`, - for the lists of system files, shell paths, command names and JavaScript - names (930120, 932120, 932160, 932260, 941180). - - These fields of a body. Each applies only when the query has no parameter - of that name, so it matters only while `SWW_WAF_BODY_LIMIT` is set: - - `content` of a post, a page, a template or a template part saved - through the REST API. It holds HTML and any code it shows. - - `content` of a reply to a comment sent to `admin-ajax.php`. - - `newcontent` of a plugin's or a theme's file saved from its editor, - which is PHP. - - `pwd` when signing in, for the list of system files (930120). - - `title`, `excerpt` and `name` in a JSON body sent to any REST API - route under `/wp/v2/`, for the rules on system files and commands - (930120, 932120, 932250, 932260). - - `slug` in a JSON body sent to any REST API route under `/wp/v2/`, for - the lists of PowerShell commands and command names (932120, 932260). - - `url` on `/wp-comments-post.php`, the address of a commenter's site, - for the lists of system files and PowerShell commands (930120, - 932120). + WordPress writes these names as slugs, such as `htaccess` and `ls-la`. - Elsewhere this change does not apply. The paths are those of WordPress at - the site root. An app other than WordPress that serves one of them must - check these fields itself. +- `url` on the REST API routes `/wp-block-editor/v1/url-details` and + `/oembed/1.0/proxy`, for the lists of system files and PowerShell commands + (930120, 932120). The block editor's link previews and embeds send an address + in this field. -10. On Drupal's paths, some fields skip the rules that refused ordinary Drupal - requests there. `internal/coreruleset/drupal.conf` lists each rule and the - request it refused. These query fields skip the lists of system files and - command names (930120, 932120, 932250, 932260): - - On every page under `/admin`: `title`, `name`, `user`, `filename`, - `filemime`, `info`, `subject`, `author_name`, `search`, `q` and - `destination`. - - `destination` on the pages Drupal links to with one, such as these: - - `/node/1/edit`; - - `/user/login`; - - `/node/1/revisions/2/revert`; - - `/block/add/basic`; - - `/big_pipe/no-js`. - - `keys` on `/search` and the pages under it. - - `q` on these autocompletes: - - `/entity_reference_autocomplete/`; - - `/ckeditor5/{editor}/entity-link-suggestions`; - - `/block-category/autocomplete`. - - Every `filter[...]` field on `/jsonapi/`. - - `name` on `/user/password`, `/media-library` and `/views/ajax`, in the - query or a body. +- `_wp_http_referer` on `/wp-admin/upload.php`, for a Unix command given without + arguments (932340). The media list's own address ends in `&ps`. +- `file` on `/wp-admin/plugin-editor.php` and `/wp-admin/theme-editor.php`, for + the lists of system files, shell paths, command names and JavaScript names + (930120, 932120, 932160, 932260, 941180). +- These fields of a body. Each applies only when the query has no parameter of + that name, so it matters only while `SWW_WAF_BODY_LIMIT` is set: + - `content` of a post, a page, a template or a template part saved through + the REST API. It holds HTML and any code it shows. + - `content` of a reply to a comment sent to `admin-ajax.php`. + - `newcontent` of a plugin's or a theme's file saved from its editor, which + is PHP. + - `pwd` when signing in, for the list of system files (930120). + - `title`, `excerpt` and `name` in a JSON body sent to any REST API route + under `/wp/v2/`, for the rules on system files and commands (930120, + 932120, 932250, 932260). + - `slug` in a JSON body sent to any REST API route under `/wp/v2/`, for the + lists of PowerShell commands and command names (932120, 932260). + - `url` on `/wp-comments-post.php`, the address of a commenter's site, for + the lists of system files and PowerShell commands (930120, 932120). - `destination` also skips 932230 and 932235, because it holds the address of - a page of results. `url` on `/media/oembed`, the address of a video, skips - only 930120 and 932120. +Elsewhere WordPress's fixes do not apply. The paths are those of WordPress at +the site root. - These fields of a body skip the rules that refused them: - - On the node, comment and media forms, for 930120, 932120, 932250 and - 932260: - - the title; - - the summary; - - the tags; - - the address; - - the revision note; - - the comment subject; - - the media name; - - the image text. - - On the tag, custom block and contact forms, for the same four rules: - - the tag's name and description; - - the block description and body; - - the subject and message of a contact message. - - These fields through JSON:API, for the same four rules: - - `title`; - - `subject`; - - `body.summary`; - - `name`; - - `revision_log`; - - `info`; - - `description.value`. - - The title saved through Drupal's JSON format on `/node/{id}`, for the same - four rules. - - The body and comment body on the node and comment forms and through - JSON:API, which hold HTML and any code it shows. - - `name` and `pass` when signing in (930120, 932120, 932250, 932260). - - `name` when resetting a password through Drupal's JSON format on - `/user/password`, for the same four rules. - - `mail` in that same reset, for 932120 and 932260 only. - - `name` when signing up, editing an account and commenting without an - account, for the same four rules. - - `mail` on those forms, for 930120, 932120 and 932260 only. +Drupal's fixes: on Drupal's paths, some fields skip the rules that refused +ordinary Drupal requests there. `internal/coreruleset/drupal.conf` lists each +rule and the request it refused. These query fields skip the lists of system +files and command names (930120, 932120, 932250, 932260): - These exclusions matter only while `SWW_WAF_BODY_LIMIT` is set. None applies - when the query has a parameter of the same name. +- On every page under `/admin`: `title`, `name`, `user`, `filename`, `filemime`, + `info`, `subject`, `author_name`, `search`, `q` and `destination`. +- `destination` on the pages Drupal links to with one, such as these: + - `/node/1/edit`; + - `/user/login`; + - `/node/1/revisions/2/revert`; + - `/block/add/basic`; + - `/big_pipe/no-js`. +- `keys` on `/search` and the pages under it. +- `q` on these autocompletes: + - `/entity_reference_autocomplete/`; + - `/ckeditor5/{editor}/entity-link-suggestions`; + - `/block-category/autocomplete`. +- Every `filter[...]` field on `/jsonapi/`. +- `name` on `/user/password`, `/media-library` and `/views/ajax`, in the query + or a body. - Elsewhere these fields keep every rule. The paths are those of Drupal at the - site root, with or without a language prefix such as `/fr`. Such a prefix is - any first path segment of two or three lowercase letters, optionally - followed by a hyphen and more letters, so `/api` counts too. An app other - than Drupal that serves one of them must check these fields itself. +`destination` also skips 932230 and 932235, because it holds the address of a +page of results. `url` on `/media/oembed`, the address of a video, skips only +930120 and 932120. -11. On MediaWiki's paths, some fields skip the rules that refused ordinary - MediaWiki requests there. `internal/coreruleset/mediawiki.conf` lists each - rule and the request it refused. These fields skip the lists of system files - and command names (930120, 932120, 932250, 932260), in the query or the - body: - - On MediaWiki's pages, which are `/index.php` and the paths under - `/index.php/` and `/wiki/`: - - the searches `search` and `ilsearch`; - - the titles `title`, `page` and `target`; - - `from`, `to` and `prefix`, which limit a list of pages; - - the user names `user`, `username` and `wpTarget`; - - the file names `filename` and `wpDestFile`; - - `returnto` and `returntoquery`, the page and query to go back to after - signing in. - - On `/api.php`: - - the searches `search`, `srsearch` and `pssearch`; - - the titles `titles`, `title`, `page` and `bltitle`; - - `from` and `to`, the titles of a page being moved; - - the starts of lists `apfrom`, `apto`, `apprefix`, `aifrom`, - `aiprefix`, `aufrom` and `auprefix`; - - `letitle`, the page whose log is listed; - - `euquery`, an address; - - the fields a generator sends in place of the fields above, each with a - `g` in front: - - `gsrsearch` and `gpssearch`; - - `gapfrom`, `gapto` and `gapprefix`; - - `gaifrom` and `gaiprefix`; - - `gbltitle`; - - `geuquery`; - - the user names `ucuser`, `bkusers` and `user`; - - `summary`, `sectiontitle`, `filename`, `comment` and `reason`; - - `lgname`, `lgpassword`, `username` and `password`, for signing in. - - `q` on `/rest.php/` and the paths under it. +These fields of a body skip the rules that refused them: - `modules` on `/load.php` skips 930120 alone, because it lists the scripts - and styles a page loads. +- On the node, comment and media forms, for 930120, 932120, 932250 and 932260: + - the title; + - the summary; + - the tags; + - the address; + - the revision note; + - the comment subject; + - the media name; + - the image text. +- On the tag, custom block and contact forms, for the same four rules: + - the tag's name and description; + - the block description and body; + - the subject and message of a contact message. +- These fields through JSON:API, for the same four rules: + - `title`; + - `subject`; + - `body.summary`; + - `name`; + - `revision_log`; + - `info`; + - `description.value`. +- The title saved through Drupal's JSON format on `/node/{id}`, for the same + four rules. +- The body and comment body on the node and comment forms and through JSON:API, + which hold HTML and any code it shows. +- `name` and `pass` when signing in (930120, 932120, 932250, 932260). +- `name` when resetting a password through Drupal's JSON format on + `/user/password`, for the same four rules. +- `mail` in that same reset, for 932120 and 932260 only. +- `name` when signing up, editing an account and commenting without an account, + for the same four rules. +- `mail` on those forms, for 930120, 932120 and 932260 only. - These fields of a body skip the rules that refused them: - - On MediaWiki's pages, for the same four rules: - - `wpSummary`, the summary of an edit; - - `wpUploadDescription`, the description of an uploaded file; - - `wpName` and `wpPassword`, for signing in; - - `wpOldTitle`, `wpNewTitleMain` and `wpReason`, for moving a page. - - `comment` in the JSON body of `/rest.php/`, for the same four rules. - - On MediaWiki's pages, for 930120 and 932120 alone, the reasons on the - admin pages: - - `wpReason-other`, for blocking a user; - - `user-reason`, for changing a user's groups; - - `mwProtect-reason`, for protecting a page. - - On MediaWiki's pages, `wpTitles`, the list of watched pages that - `Special:EditWatchlist/raw` sends, one title per line. It always skips - 930120, 932120 and 932250. It skips 932125, 932230, 932235, 932260 and - 932380 only when it holds no `;`, `|`, `&`, backtick or `{`. A title - starting with a command name, such as `Python3 notes`, passes. So does a - command alone on a line, such as `wget http://example.com/x.sh`. A command - such as `; wget http://example.com/x.sh` is still refused. So is a list - that mixes a title holding one of these characters with a title starting - with a command name. - - The wikitext of a page, which holds templates, tables and any code it - shows: - - `wpTextbox1` on MediaWiki's pages; - - `text`, `appendtext` and `prependtext` on `/api.php`; - - `source` in the JSON body of `/rest.php/`. +These exclusions matter only while `SWW_WAF_BODY_LIMIT` is set. None applies +when the query has a parameter of the same name. - These body exclusions matter only while `SWW_WAF_BODY_LIMIT` is set. None - applies when the query has a parameter of the same name. +Elsewhere these fields keep every rule. The paths are those of Drupal at the +site root, with or without a language prefix such as `/fr`. Such a prefix is any +first path segment of two or three lowercase letters, optionally followed by a +hyphen and more letters, so `/api` counts too. - The names of uploaded files skip 932180 on MediaWiki's pages and on - `/api.php`, so a picture named `Notes on .htaccess.png` can be uploaded. - MediaWiki refuses a file whose extension it does not allow. +MediaWiki's fixes: on MediaWiki's paths, some fields skip the rules that refused +ordinary MediaWiki requests there. `internal/coreruleset/mediawiki.conf` lists +each rule and the request it refused. These fields skip the lists of system +files and command names (930120, 932120, 932250, 932260), in the query or the +body: - `/index.php` itself counts as MediaWiki's only when its query has a `title`, - as in `/index.php?title=Main_Page`. WordPress serves `/index.php` too. A - wiki without short URLs serves its pages under `/index.php/`, as in - `/index.php/Main_Page`, and its sign-in and upload forms post there. +- On MediaWiki's pages, which are `/index.php` and the paths under `/index.php/` + and `/wiki/`: + - the searches `search` and `ilsearch`; + - the titles `title`, `page` and `target`; + - `from`, `to` and `prefix`, which limit a list of pages; + - the user names `user`, `username` and `wpTarget`; + - the file names `filename` and `wpDestFile`; + - `returnto` and `returntoquery`, the page and query to go back to after + signing in. +- On `/api.php`: + - the searches `search`, `srsearch` and `pssearch`; + - the titles `titles`, `title`, `page` and `bltitle`; + - `from` and `to`, the titles of a page being moved; + - the starts of lists `apfrom`, `apto`, `apprefix`, `aifrom`, `aiprefix`, + `aufrom` and `auprefix`; + - `letitle`, the page whose log is listed; + - `euquery`, an address; + - the fields a generator sends in place of the fields above, each with a `g` + in front: + - `gsrsearch` and `gpssearch`; + - `gapfrom`, `gapto` and `gapprefix`; + - `gaifrom` and `gaiprefix`; + - `gbltitle`; + - `geuquery`; + - the user names `ucuser`, `bkusers` and `user`; + - `summary`, `sectiontitle`, `filename`, `comment` and `reason`; + - `lgname`, `lgpassword`, `username` and `password`, for signing in. +- `q` on `/rest.php/` and the paths under it. - Elsewhere these fields keep every rule. The paths are those of MediaWiki at - the site root or under `/w/`. An app other than MediaWiki that serves one of - them must check these fields itself. +`modules` on `/load.php` skips 930120 alone, because it lists the scripts and +styles a page loads. -The parameter and cookie names in these changes are matched without regard to -case, as Coraza matches them. +These fields of a body skip the rules that refused them: + +- On MediaWiki's pages, for the same four rules: + - `wpSummary`, the summary of an edit; + - `wpUploadDescription`, the description of an uploaded file; + - `wpName` and `wpPassword`, for signing in; + - `wpOldTitle`, `wpNewTitleMain` and `wpReason`, for moving a page. +- `comment` in the JSON body of `/rest.php/`, for the same four rules. +- On MediaWiki's pages, for 930120 and 932120 alone, the reasons on the admin + pages: + - `wpReason-other`, for blocking a user; + - `user-reason`, for changing a user's groups; + - `mwProtect-reason`, for protecting a page. +- On MediaWiki's pages, `wpTitles`, the list of watched pages that + `Special:EditWatchlist/raw` sends, one title per line. It always skips 930120, + 932120 and 932250. It skips 932125, 932230, 932235, 932260 and 932380 only + when it holds no `;`, `|`, `&`, backtick or `{`. A title starting with a + command name, such as `Python3 notes`, passes. So does a command alone on a + line, such as `wget http://example.com/x.sh`. A command such as + `; wget http://example.com/x.sh` is still refused. So is a list that mixes a + title holding one of these characters with a title starting with a command + name. +- The wikitext of a page, which holds templates, tables and any code it shows: + - `wpTextbox1` on MediaWiki's pages; + - `text`, `appendtext` and `prependtext` on `/api.php`; + - `source` in the JSON body of `/rest.php/`. + +These body exclusions matter only while `SWW_WAF_BODY_LIMIT` is set. None +applies when the query has a parameter of the same name. + +The names of uploaded files skip 932180 on MediaWiki's pages and on `/api.php`, +so a picture named `Notes on .htaccess.png` can be uploaded. MediaWiki refuses a +file whose extension it does not allow. + +`/index.php` itself counts as MediaWiki's only when its query has a `title`, as +in `/index.php?title=Main_Page`. WordPress serves `/index.php` too. A wiki +without short URLs serves its pages under `/index.php/`, as in +`/index.php/Main_Page`, and its sign-in and upload forms post there. + +Elsewhere these fields keep every rule. The paths are those of MediaWiki at the +site root or under `/w/`. + +Each app's fixes switch rules off for some of its fields and cookies. Any other +app on a hostname that gets those fixes must check those fields and cookies +itself. + +The parameter names in these changes and fixes are matched without regard to +case, and the cookie names exactly. ### Ban lengths, response and scope diff --git a/docs/settings.md b/docs/settings.md index 18b7dee..0e7001c 100644 --- a/docs/settings.md +++ b/docs/settings.md @@ -255,6 +255,24 @@ the request first. a range. So does a directive that removes or changes rules as the file loads, such as `SecRuleRemoveById`, since none of the Core Rule Set's rules is loaded yet. It has no `_FILE` form. +- `SWW_APPS` (default unset): the apps each hostname serves. A request gets the + fixes of its hostname's apps only (see "Attack detection" in + [`SPEC.md`](../SPEC.md)). + - Entries are separated by semicolons. Each is a hostname, a colon, then one + or more apps separated by commas, for example + `git.example.com:gitea;blog.example.com:wordpress`. + - The apps supported are `drupal`, `gitea`, `mediawiki` and `wordpress`. + - The request's `Host` is matched by exact name, without its port and + without regard to case. There are no wildcards. + - A hostname not listed gets no app's fixes, so every rule checks it in + full. Set empty, `SWW_APPS` lists no hostname. + - Unset, every app's fixes apply to every hostname. + - Any of these mistakes stops the start: + - an unknown app; + - a hostname that is not a DNS name, such as `*.example.com`; + - a hostname listed twice; + - an app listed twice for one hostname; + - an entry without a colon. - `SWW_ERROR_BURST_THRESHOLD` (default `30`): the most refusals a client may get in a minute from a rule file, a trap path or the Core Rule Set, a count or `off`. The next one bans the client (see "Attack detection" in `SPEC.md`). diff --git a/internal/config/config.go b/internal/config/config.go index 8cb82d1..95ec599 100644 --- a/internal/config/config.go +++ b/internal/config/config.go @@ -29,6 +29,7 @@ import ( "sneak.berlin/go/smallwebwaf/internal/alerts" "sneak.berlin/go/smallwebwaf/internal/anomaly" + "sneak.berlin/go/smallwebwaf/internal/coreruleset" "sneak.berlin/go/smallwebwaf/internal/remotelog" ) @@ -235,6 +236,11 @@ type Config struct { WAFExemptPaths []string WAFBodyLimit int64 CoreRuleSetExclusionsFile string + // Apps are the names of the apps each hostname serves, by hostname in + // lowercase (SWW_APPS): a request gets only its hostname's apps' fixes + // to the Core Rule Set. Apps is nil while SWW_APPS is unset, and every + // request gets every app's fixes. + Apps map[string][]string // TrapPaths are the paths a request for which is a clear sign of // attack (SWW_TRAP_PATHS), each starting with / and without a ?. TrapPaths []string @@ -404,6 +410,11 @@ var ( errNotTrapPath = errors.New( "is not a path starting with / and without a ?, such as /wp-login.php", ) + errNotAppsEntry = errors.New( + "is not a hostname, : and a list of apps, such as git.example.com:gitea", + ) + errNotHostname = errors.New("is not a DNS name such as git.example.com") + errNotApp = errors.New("is not an app sww has fixes for; the apps are") errNotWAFMode = errors.New("is not off, detect or block") errNotParanoiaLevel = errors.New("is not a paranoia level, from 1 to 4") errNotRuleID = errors.New( @@ -577,6 +588,7 @@ func FromEnvironment( WAFBodyLimit: env.wafBodyLimit("SWW_WAF_BODY_LIMIT", off), CoreRuleSetExclusionsFile: env.filePath( "SWW_CORE_RULE_SET_EXCLUSIONS_FILE"), + Apps: env.apps("SWW_APPS"), TrapPaths: env.trapPaths("SWW_TRAP_PATHS"), ErrorBurstThreshold: env.count("SWW_ERROR_BURST_THRESHOLD", "30"), LogRemoteURL: env.logRemoteURL("SWW_LOG_REMOTE_URL"), @@ -832,6 +844,25 @@ func (e *environment) pathPrefixes(name, defaultValue string) []string { return prefixes } +// apps reads the setting that is the apps each hostname serves. It is nil +// while the setting is unset, and logged as null, unlike a setting set +// empty. +func (e *environment) apps(name string) map[string][]string { + value, set := e.lookup(name) + if !set { + e.settings = append(e.settings, slog.Any(name, nil)) + + return nil + } + + e.settings = append(e.settings, slog.String(name, value)) + + apps, err := parseApps(value) + e.check(name, err) + + return apps +} + // trapPaths reads the setting that is the list of trap paths. It is empty // by default. func (e *environment) trapPaths(name string) []string { @@ -1849,6 +1880,65 @@ func parseTrapPaths(value string) ([]string, error) { return paths, nil } +// parseApps reads the apps each hostname serves: entries separated by +// semicolons, each a hostname, a colon, and a comma-separated list of the +// names coreruleset.AppNames returns, such as +// git.example.com:gitea;cloud.example.com:nextcloud,wordpress. Each +// hostname is a DNS name, without a wildcard. It returns them by hostname +// in lowercase, since a hostname is matched without regard to case. An +// empty value lists no hostname. +func parseApps(value string) (map[string][]string, error) { + apps := map[string][]string{} + + if strings.TrimSpace(value) == "" { + return apps, nil + } + + for entry := range strings.SplitSeq(value, ";") { + hostname, list, found := strings.Cut(entry, ":") + if !found { + return nil, fmt.Errorf("%q %w", entry, errNotAppsEntry) + } + + hostname = strings.ToLower(strings.TrimSpace(hostname)) + if hostname == "" { + return nil, fmt.Errorf("%q %w", entry, errNotAppsEntry) + } + + if !isZone(hostname) { + return nil, fmt.Errorf("%q %w", hostname, errNotHostname) + } + + if _, listed := apps[hostname]; listed { + return nil, fmt.Errorf("%q %w", hostname, errListedTwice) + } + + names, err := parseList(list) + if err != nil { + return nil, err + } + + if len(names) == 0 { + return nil, fmt.Errorf("%q %w", entry, errNotAppsEntry) + } + + for i, name := range names { + if !slices.Contains(coreruleset.AppNames(), name) { + return nil, fmt.Errorf("%q %w %s", name, errNotApp, + strings.Join(coreruleset.AppNames(), ", ")) + } + + if slices.Contains(names[:i], name) { + return nil, fmt.Errorf("%q %w", name, errListedTwice) + } + } + + apps[hostname] = names + } + + return apps, nil +} + // parseRuleIDs reads a comma-separated list of the ids of Core Rule Set // rules, each a whole number above zero and outside firstSetupRuleID to // lastSetupRuleID. @@ -2455,7 +2545,8 @@ func MaskZoneKey(zone string) string { return masked + "." + rest } -// isZone reports whether each label of zone is as parseZones takes it. +// isZone reports whether each label of zone is as parseZones takes it. It +// also checks the hostnames of SWW_APPS, which are DNS names too. func isZone(zone string) bool { for label := range strings.SplitSeq(zone, ".") { badChar := strings.ContainsFunc(label, func(char rune) bool { diff --git a/internal/config/config_test.go b/internal/config/config_test.go index 37e8b7b..c7e7605 100644 --- a/internal/config/config_test.go +++ b/internal/config/config_test.go @@ -89,6 +89,7 @@ const ( wafExemptPaths = "SWW_WAF_EXEMPT_PATHS" wafBodyLimit = "SWW_WAF_BODY_LIMIT" exclusionsFile = "SWW_CORE_RULE_SET_EXCLUSIONS_FILE" + apps = "SWW_APPS" trapPaths = "SWW_TRAP_PATHS" errorBurstThreshold = "SWW_ERROR_BURST_THRESHOLD" logRemoteURL = "SWW_LOG_REMOTE_URL" @@ -582,6 +583,100 @@ func TestCoreRuleSetSettings(t *testing.T) { } } +// gitea is gitea's name in SWW_APPS. +const gitea = "gitea" + +func TestApps(t *testing.T) { + t.Parallel() + + for _, tc := range []struct { + env environment + want map[string][]string + }{ + {environment{}, nil}, + {environment{apps: ""}, map[string][]string{}}, + { + environment{apps: "Git.Example.com:gitea; cloud.example.com : gitea "}, + map[string][]string{ + "git.example.com": {gitea}, "cloud.example.com": {gitea}, + }, + }, + { + environment{apps: "cloud.example.com:gitea,wordpress"}, + map[string][]string{"cloud.example.com": {gitea, "wordpress"}}, + }, + } { + cfg := fromEnvironment(t, tc.env) + if !reflect.DeepEqual(cfg.Apps, tc.want) { + t.Errorf("%v gave %#v, want %#v", tc.env, cfg.Apps, tc.want) + } + } +} + +// Unset and set empty, SWW_APPS has opposite effects, so it logs differently. +func TestUnsetAndEmptyAppsAreLoggedDifferently(t *testing.T) { + t.Parallel() + + for _, tc := range []struct { + env environment + want string + }{ + {environment{}, `"` + apps + `":null`}, + {environment{apps: ""}, `"` + apps + `":""`}, + } { + var out bytes.Buffer + + slog.New(slog.NewJSONHandler(&out, nil)). + Info("starting", "settings", fromEnvironment(t, tc.env)) + + if !strings.Contains(out.String(), tc.want) { + t.Errorf("%v logged %s, want %s in it", tc.env, out.String(), tc.want) + } + } +} + +func TestInvalidAppsStopsTheStart(t *testing.T) { + t.Parallel() + + const ( + notEntry = " is not a hostname, : and a list of apps, " + + "such as git.example.com:gitea" + notHostname = " is not a DNS name such as git.example.com" + ) + + for _, tc := range []struct{ value, want string }{ + {"git.example.com:gitea;blog.example.com", `"blog.example.com"` + notEntry}, + {"git.example.com:gitea;", `""` + notEntry}, + {":gitea", `":gitea"` + notEntry}, + {"git.example.com:", `"git.example.com:"` + notEntry}, + {"*.example.com:gitea", `"*.example.com"` + notHostname}, + {"git.example.com/:gitea", `"git.example.com/"` + notHostname}, + {"git example.com:gitea", `"git example.com"` + notHostname}, + { + "git.example.com:gitea,nextcloud", + `"nextcloud" is not an app sww has fixes for; ` + + `the apps are drupal, gitea, mediawiki, wordpress`, + }, + { + "git.example.com:gitea;GIT.example.com:gitea", + `"git.example.com" is listed twice`, + }, + {"git.example.com:gitea,gitea", `"gitea" is listed twice`}, + {"git.example.com:gitea,", `"gitea," has an empty item in its list`}, + } { + t.Run(tc.value, func(t *testing.T) { + t.Parallel() + + _, err := config.FromEnvironment(environment{apps: tc.value}.lookupEnv) + + want := apps + ": " + tc.want + if err == nil || err.Error() != want { + t.Errorf("error %v, want %s", err, want) + } + }) + } +} + func TestWAFBodyLimitOf64K(t *testing.T) { t.Parallel() @@ -2589,6 +2684,7 @@ func TestLogsEachSettingWithItsValue(t *testing.T) { wafExemptPaths: "", wafBodyLimit: off, exclusionsFile: "", + apps: "", // logged as null, which decodes as "" trapPaths: "", errorBurstThreshold: "30", logRemoteURL: "", diff --git a/internal/coreruleset/coreruleset.go b/internal/coreruleset/coreruleset.go index c5b4179..8329532 100644 --- a/internal/coreruleset/coreruleset.go +++ b/internal/coreruleset/coreruleset.go @@ -1,9 +1,9 @@ // Package coreruleset runs the OWASP Core Rule Set 4.25.0, through Coraza, on the // method, the URL with its query and the headers of a request, and on its -// body while SWW_WAF_BODY_LIMIT is set, with the eleven changes sww -// makes to it and the exclusions in SWW_CORE_RULE_SET_EXCLUSIONS_FILE, as -// "Attack detection" under "Settings" in SPEC.md describes them. It reads no -// response. +// body while SWW_WAF_BODY_LIMIT is set, with the six changes sww +// makes to it, the fixes for the apps SWW_APPS names, and the exclusions in +// SWW_CORE_RULE_SET_EXCLUSIONS_FILE, as "Attack detection" under "Settings" +// in SPEC.md describes them. It reads no response. // // sww writes only to its state directory, so Coraza is built with // its no_fs_access tag, as the Dockerfile and script/build build it: of a @@ -16,6 +16,8 @@ import ( "errors" "fmt" "io" + "maps" + "net" "net/http" "net/netip" "os" @@ -32,9 +34,8 @@ import ( // directives are the Core Rule Set as sww runs it, with the // paranoia level for %d, bodyDirectives for the first %s while -// SWW_WAF_BODY_LIMIT is set, the exclusions file for the second, -// giteaExclusions for the third, wordpressExclusions for the fourth, -// drupalExclusions for the fifth, and mediawikiExclusions for the sixth. +// SWW_WAF_BODY_LIMIT is set, the exclusions file for the second, and the +// directives of every app in apps for the third. // Each rule sww adds has an id from 900000 to 900999, the ids the // Core Rule Set keeps for the rules that set it up, which // SWW_WAF_DISABLED_RULES and the exclusions file refuse, so that no @@ -82,53 +83,28 @@ SecRule ARGUMENTS_LIMIT_REACHED "@eq 1" "id:900300,phase:2,pass,\ # before the rules of phase 1 too. %s -# The eighth change comes before the rules for the same reason. +# The apps' fixes come before the rules for the same reason. %s -# The ninth change comes before the rules for the same reason. -%s - -# So does the tenth. -%s - -# And the eleventh. -%s - -# The sixth: only the rules for requests are loaded, and no response is +# The fifth: only the rules for requests are loaded, and no response is # inspected. Include @owasp_crs/REQUEST-*.conf # The third: redirect_uri is not checked for a URL naming an IP address or -# localhost. Coraza matches a parameter name here, and in the fourth, -# without regard to case. ARGS holds the fields of a form data or multipart -# body Coraza reads as well as the query parameters, so a field of one of -# these names is left out too. +# localhost. Coraza matches a parameter name here without regard to case. +# ARGS holds the fields of a form data or multipart body Coraza reads as +# well as the query parameters, so a field of this name is left out too. SecRuleUpdateTargetById 931100 "!ARGS:redirect_uri" SecRuleUpdateTargetById 934110 "!ARGS:redirect_uri" -# The fourth: the query parameters in which gitea sends names within a -# repository or its own records, or a page of its own site, are not -# checked against the lists of system files, shell paths and command -# names. Coraza takes one rule id per directive. -SecRuleUpdateTargetById 930120 "!ARGS:path|!ARGS:files|!ARGS:skip-to|\ - !ARGS:sub_path|!ARGS:ref|!ARGS:sha|!ARGS:branch|!ARGS:workflow|\ - !ARGS:artifactName|!ARGS:redirect_to" -SecRuleUpdateTargetById 932160 "!ARGS:path|!ARGS:files|!ARGS:skip-to|\ - !ARGS:sub_path|!ARGS:ref|!ARGS:sha|!ARGS:branch|!ARGS:workflow|\ - !ARGS:artifactName|!ARGS:redirect_to" -SecRuleUpdateTargetById 932260 "!ARGS:path|!ARGS:files|!ARGS:skip-to|\ - !ARGS:sub_path|!ARGS:ref|!ARGS:sha|!ARGS:branch|!ARGS:workflow|\ - !ARGS:artifactName|!ARGS:redirect_to" - -# The fifth, for Referer: it is not checked for a Unix command without -# arguments, for Java starting a process, or for a Java class name. 944130 -# refused every page opened from the page of java.lang.String.md. The -# cookies are left out in Inspect. +# The fourth: Referer is not checked for a Unix command without arguments, +# for Java starting a process, or for a Java class name. 944130 refused +# every page opened from the page of java.lang.String.md. SecRuleUpdateTargetById 932340 "!REQUEST_HEADERS:Referer" SecRuleUpdateTargetById 944110 "!REQUEST_HEADERS:Referer" SecRuleUpdateTargetById 944130 "!REQUEST_HEADERS:Referer" -# The seventh: a container image push that reuses a layer names the image +# The sixth: a container image push that reuses a layer names the image # it takes the layer from in the query parameter from, which is not # checked for command names. 932260 refused the push of a layer from # owner/python3-app. @@ -178,39 +154,49 @@ SecRule MULTIPART_STRICT_ERROR "!@eq 0" "id:900450,phase:2,pass,\ severity:'CRITICAL',setvar:'tx.inbound_anomaly_score_pl1=+5'" ` -// giteaExclusions are the eighth change: the fields of gitea's own -// requests that skip the rules that refused them there, as Coraza -// directives. -// +// app is an app whose ordinary requests the Core Rule Set would refuse, +// with the fixes that let them pass. +type app struct { + // directives are Coraza exclusions, loaded before the Core Rule Set's + // rules. Their first rule skips to the marker they end with, unless + // Inspect sets the transaction variable named for the app, such as + // app_gitea. The skip reaches only phase 1, so a rule of a later phase + // must check that variable itself. + directives string + // cookies are the cookies the Core Rule Set reads a request without. + cookies []string +} + //go:embed gitea.conf -var giteaExclusions string +var giteaDirectives string -// wordpressExclusions are the ninth change: the fields of WordPress's own -// requests that skip the rules that refused them there, as Coraza -// directives. -// //go:embed wordpress.conf -var wordpressExclusions string +var wordpressDirectives string -// drupalExclusions are the tenth change: the fields of Drupal's own -// requests that skip the rules that refused them there, as Coraza -// directives. -// //go:embed drupal.conf -var drupalExclusions string +var drupalDirectives string -// mediawikiExclusions are the eleventh change: the fields of MediaWiki's -// own requests that skip the rules that refused them there, as Coraza -// directives. -// //go:embed mediawiki.conf -var mediawikiExclusions string +var mediawikiDirectives string -// cookiesNotRead are the cookies the Core Rule Set reads a request -// without, the rest of the fifth change. +// apps are the apps whose fixes sww has, by the name SWW_APPS gives each. // -//nolint:gochecknoglobals // a constant cannot be a list -var cookiesNotRead = []string{"gitea_flash", "redirect_to"} +//nolint:gochecknoglobals // a constant cannot be a map +var apps = map[string]app{ + "drupal": {directives: drupalDirectives}, + "gitea": { + directives: giteaDirectives, + cookies: []string{"gitea_flash", "redirect_to"}, + }, + "mediawiki": {directives: mediawikiDirectives}, + "wordpress": {directives: wordpressDirectives}, +} + +// AppNames returns the names of the apps whose fixes sww has, in +// alphabetical order. +func AppNames() []string { + return slices.Sorted(maps.Keys(apps)) +} // Params are what New needs. type Params struct { @@ -226,6 +212,10 @@ type Params struct { // Core Rule Set's rules (SWW_CORE_RULE_SET_EXCLUSIONS_FILE), "" while // there is none. ExclusionsFile string + // Apps are the names of the apps each hostname serves, by hostname in + // lowercase (SWW_APPS). A request gets the fixes of its hostname's + // apps only, or of every app while Apps is nil. + Apps map[string][]string } // The ids of sww's own rules, which an exclusions file may not @@ -317,13 +307,15 @@ func checkExclusions(exclusions string) error { type CoreRuleSet struct { engine coraza.WAF bodyLimit int64 + apps map[string][]string } -// New returns the Core Rule Set with the eleven changes, at params' -// paranoia level, without the rules it switches off, reading request -// bodies up to params' limit, and with params' exclusions file loaded -// before the Core Rule Set's rules. An exclusions file is an error if it -// cannot be read or does not parse, or if checkExclusions refuses it. +// New returns the Core Rule Set with the six changes and the apps' fixes, +// at params' paranoia level, without the rules it switches off, reading +// request bodies up to params' limit, and with params' exclusions file +// loaded before the Core Rule Set's rules. An exclusions file is an error +// if it cannot be read or does not parse, or if checkExclusions refuses +// it. func New(params Params) (*CoreRuleSet, error) { body := "" if params.BodyLimit > 0 { @@ -346,9 +338,13 @@ func New(params Params) (*CoreRuleSet, error) { } } + var fixes strings.Builder + for _, name := range AppNames() { + fixes.WriteString(apps[name].directives + "\n") + } + text := fmt.Sprintf(directives, params.ParanoiaLevel, body, exclusions, - giteaExclusions, wordpressExclusions, drupalExclusions, - mediawikiExclusions) + fixes.String()) if len(params.DisabledRules) > 0 { ids := make([]string, len(params.DisabledRules)) @@ -366,7 +362,9 @@ func New(params Params) (*CoreRuleSet, error) { return nil, fmt.Errorf("load the Core Rule Set: %w", err) } - return &CoreRuleSet{engine: engine, bodyLimit: params.BodyLimit}, nil + return &CoreRuleSet{ + engine: engine, bodyLimit: params.BodyLimit, apps: params.Apps, + }, nil } // Result is what the Core Rule Set found in a request. @@ -378,11 +376,12 @@ type Result struct { Score int } -// Inspect runs the Core Rule Set on r, a request from client: on its -// method, its URL with the query, and its headers, the Cookie header -// without the cookies in cookiesNotRead, and on body, r's body as the -// caller has it, as readBody reads it. It returns what it found, what it -// read of body, and the error that ended the reading early, if one did. +// Inspect runs the Core Rule Set on r, a request from client, with the +// fixes of the apps appsFor names for r's hostname: on its method, its URL +// with the query, and its headers, the Cookie header without those apps' +// cookies, and on body, r's body as the caller has it, as readBody reads +// it. It returns what it found, what it read of body, and the error that +// ended the reading early, if one did. func (c *CoreRuleSet) Inspect( r *http.Request, client netip.Addr, body io.Reader, ) (Result, []byte, error) { @@ -390,13 +389,24 @@ func (c *CoreRuleSet) Inspect( // Closing would remove the files Coraza wrote, and it writes none. defer func() { _ = tx.Close() }() + // The transaction's variables are set only through Coraza's interface + // for plugins. + state := tx.(plugintypes.TransactionState) //nolint:forcetypeassert // every one is + + var cookiesNotRead []string + + for _, name := range c.appsFor(r.Host) { + state.Variables().TX().Set("app_"+name, []string{"1"}) + cookiesNotRead = append(cookiesNotRead, apps[name].cookies...) + } + tx.ProcessConnection(client.String(), 0, "", 0) tx.ProcessURI(r.URL.String(), r.Method, r.Proto) for name, values := range r.Header { for _, value := range values { if name == "Cookie" { - value = withoutCookiesNotRead(value) + value = withoutCookies(value, cookiesNotRead) if value == "" { continue // it held those cookies alone } @@ -489,9 +499,26 @@ func score(tx types.Transaction) int { return n } -// withoutCookiesNotRead returns value, a Cookie header's, without the -// cookies in cookiesNotRead. -func withoutCookiesNotRead(value string) string { +// appsFor returns the names of the apps whose fixes a request for host, +// its Host header, gets: every app while SWW_APPS is unset, and otherwise +// those SWW_APPS lists for host's name, without the port and in +// lowercase. +func (c *CoreRuleSet) appsFor(host string) []string { + if c.apps == nil { + return AppNames() + } + + hostname, _, err := net.SplitHostPort(host) + if err != nil { + hostname = host // it has no port + } + + return c.apps[strings.ToLower(hostname)] +} + +// withoutCookies returns value, a Cookie header's, without the cookies +// named in cookiesNotRead. +func withoutCookies(value string, cookiesNotRead []string) string { var kept []string for cookie := range strings.SplitSeq(value, ";") { diff --git a/internal/coreruleset/coreruleset_test.go b/internal/coreruleset/coreruleset_test.go index 241a80d..95b0464 100644 --- a/internal/coreruleset/coreruleset_test.go +++ b/internal/coreruleset/coreruleset_test.go @@ -36,12 +36,14 @@ var loaded = struct { }{loaders: map[string]func() (*coreruleset.CoreRuleSet, error){}} // loaderFor returns the function that loads the Core Rule Set with params -// once, making it the first time params are asked for. +// once, making it the first time params are asked for. %#v writes a nil +// Apps, which gives every app's fixes, apart from an empty one, which +// gives none. func loaderFor(params coreruleset.Params) func() (*coreruleset.CoreRuleSet, error) { loaded.Lock() defer loaded.Unlock() - settings := fmt.Sprintf("%+v", params) + settings := fmt.Sprintf("%#v", params) load, found := loaded.loaders[settings] if !found { @@ -202,8 +204,8 @@ func atDefaults(t *testing.T) *coreruleset.CoreRuleSet { } // wantChange checks that crs lets through passes, a request one of the -// eleven changes is for, and still finds result in refused, a request like -// it that the change is not for. +// six changes or an app's fixes are for, and still finds result in +// refused, a request like it that they are not for. func wantChange( t *testing.T, crs *coreruleset.CoreRuleSet, passes, refused request, result coreruleset.Result, @@ -464,6 +466,123 @@ func TestImagePushThatReusesALayerIsNotCheckedForCommandNames(t *testing.T) { matched(930120, 932160)) } +// giteaFixed are gitea requests that only gitea's fixes let through, one +// for each kind of fix, each with what the Core Rule Set finds in it +// without them. +// +//nolint:gochecknoglobals // a constant cannot be a list +var giteaFixed = []struct { + request request + result coreruleset.Result +}{ + // A search of code, on the path of gitea's search. + {get("/explore/code?q=SELECT%20*%20FROM%20users%20WHERE"), matched(942100)}, + // A file name, in a parameter gitea sends names in, on any path. + {get("/?path=.gitignore"), matched(930120)}, + // A file name, in a cookie of gitea's. + { + get("/owner/repo", "Cookie: gitea_flash=success%3DFile%2Bpackage.json%2Bdeleted"), + matched(930120), + }, +} + +// The hostnames of the tests of SWW_APPS: the one that serves gitea, and +// one that serves another app. +const ( + giteaHost = "git.example" + otherHost = "blog.example" +) + +// gitea is gitea's name in SWW_APPS. +const gitea = "gitea" + +// wordpress is WordPress's name in SWW_APPS. +const wordpress = "wordpress" + +// withHost returns r sent to host. +func withHost(r request, host string) request { + return request{r.method, r.target, append([]string{"Host: " + host}, r.headers...)} +} + +// servingApps returns the Core Rule Set at the defaults, with SWW_APPS +// giving apps. +func servingApps(t *testing.T, apps map[string][]string) *coreruleset.CoreRuleSet { + t.Helper() + + return shared(t, coreruleset.Params{ + ParanoiaLevel: 1, DisabledRules: defaultDisabledRules, Apps: apps, + }) +} + +func TestAppsUnsetGivesEveryHostnameEveryAppsFixes(t *testing.T) { + t.Parallel() + + crs := servingApps(t, nil) + + for _, fixed := range giteaFixed { + for _, host := range []string{giteaHost, otherHost} { + wantResult(t, crs, withHost(fixed.request, host), coreruleset.Result{}) + } + } +} + +func TestListedHostnameGetsItsAppsFixes(t *testing.T) { + t.Parallel() + + crs := servingApps(t, map[string][]string{giteaHost: {gitea}}) + + // The hostname is matched without the port and without regard to case. + for _, host := range []string{giteaHost, "GIT.Example:8443"} { + for _, fixed := range giteaFixed { + wantResult(t, crs, withHost(fixed.request, host), coreruleset.Result{}) + } + } +} + +func TestUnlistedHostnameGetsTheFullCheck(t *testing.T) { + t.Parallel() + + for _, apps := range []map[string][]string{ + {giteaHost: {gitea}}, + // SWW_APPS set empty lists no hostname. + {}, + } { + crs := servingApps(t, apps) + + for _, host := range []string{otherHost, otherHost + ":8443", "git.example.org"} { + for _, fixed := range giteaFixed { + wantResult(t, crs, withHost(fixed.request, host), fixed.result) + } + } + } +} + +func TestWordPressFixesApplyOnlyToAHostnameThatServesWordPress(t *testing.T) { + t.Parallel() + + crs := servingApps(t, map[string][]string{ + giteaHost: {gitea}, otherHost: {wordpress}, + }) + + // A search, which only WordPress's fixes let through. + search := get("/?s=python3") + + wantResult(t, crs, withHost(search, otherHost), coreruleset.Result{}) + wantResult(t, crs, withHost(search, giteaHost), matched(932260)) +} + +func TestHostnameListingTwoAppsGetsTheFixesOfBoth(t *testing.T) { + t.Parallel() + + crs := servingApps(t, map[string][]string{giteaHost: {gitea, wordpress}}) + + // A file name that only gitea's fixes let through, and a search that only + // WordPress's fixes let through. + for _, target := range []string{"/?path=.gitignore", "/?s=python3"} { + wantResult(t, crs, withHost(get(target), giteaHost), coreruleset.Result{}) + } +} + // withExclusions returns the Core Rule Set at the defaults, with the // exclusions file at path. func withExclusions(path string) (*coreruleset.CoreRuleSet, error) { @@ -619,15 +738,16 @@ const ( var hardValues = []string{"python3", "powershell", "ls%20-la", ".htaccess"} // wantRecordedRequestsPass checks that none of the recorded ordinary -// requests in the file name in testdata is refused, at the defaults and -// while bodies are read. Each request runs as its own parallel subtest. -func wantRecordedRequestsPass(t *testing.T, name string) { +// requests in the file name in testdata is refused by defaults, which +// reads no body, or by bodies, which reads it. Each request runs as its +// own parallel subtest. +func wantRecordedRequestsPass( + t *testing.T, name string, defaults, bodies *coreruleset.CoreRuleSet, +) { t.Helper() const threshold = 5 // SWW_WAF_ANOMALY_THRESHOLD's default - defaults, bodies := atDefaults(t), readingBodies(t) - for _, recorded := range recordedRequests(t, name) { t.Run(recorded.Method+" "+recorded.Target, func(t *testing.T) { t.Parallel() @@ -810,13 +930,16 @@ func wordpressAttackTargets(attack string) []string { } } -func TestOrdinaryWordPressRequestsPassAndAttacksAreRefused(t *testing.T) { - t.Parallel() +// wantWordPressRequestsPass checks that the recorded ordinary WordPress +// requests pass defaults, which reads no body, and that those with a body +// pass bodies, which reads it. +func wantWordPressRequestsPass( + t *testing.T, defaults, bodies *coreruleset.CoreRuleSet, +) { + t.Helper() const threshold = 5 // SWW_WAF_ANOMALY_THRESHOLD's default - defaults, bodies := atDefaults(t), readingBodiesUpTo64K(t) - for _, recorded := range recordedRequests(t, "wordpress-requests.txt") { t.Run(recorded.Method+" "+recorded.Target, func(t *testing.T) { t.Parallel() @@ -836,6 +959,16 @@ func TestOrdinaryWordPressRequestsPassAndAttacksAreRefused(t *testing.T) { } }) } +} + +func TestOrdinaryWordPressRequestsPassAndAttacksAreRefused(t *testing.T) { + t.Parallel() + + const threshold = 5 // SWW_WAF_ANOMALY_THRESHOLD's default + + defaults, bodies := atDefaults(t), readingBodiesUpTo64K(t) + + wantWordPressRequestsPass(t, defaults, bodies) for _, attack := range []string{ sqlInjection, @@ -912,6 +1045,20 @@ func wordpressWords() []struct { } } +func TestHostnameListingWordPressLetsItsRecordedRequestsThrough(t *testing.T) { + t.Parallel() + + // The requests were recorded on otherHost. + apps := map[string][]string{otherHost: {wordpress}} + + bodies := shared(t, coreruleset.Params{ + ParanoiaLevel: 1, DisabledRules: defaultDisabledRules, BodyLimit: 64<<10 - 1, + Apps: apps, + }) + + wantWordPressRequestsPass(t, servingApps(t, apps), bodies) +} + // wordpressBlock is the comment that starts every block of WordPress's // HTML, which rule 941180 matches. const wordpressBlock = "" diff --git a/internal/coreruleset/drupal.conf b/internal/coreruleset/drupal.conf index 3ee559c..46eabe8 100644 --- a/internal/coreruleset/drupal.conf +++ b/internal/coreruleset/drupal.conf @@ -1,4 +1,4 @@ -# The tenth change: on Drupal's paths, some fields skip the rules that refused +# Drupal's fixes: on Drupal's paths, some fields skip the rules that refused # ordinary Drupal requests there. They are the fields where Drupal sends a # search, a name, a file name or an address. Each exclusion runs with the # request, before the rules it switches off, so this file is loaded before the @@ -25,6 +25,13 @@ # 932230: ls -la # 932235: python3 +# Unless SWW_APPS lists drupal for the request's hostname, every rule below is +# skipped. Inspect sets tx.app_drupal when it does. +# The skip reaches only phase 1, so a rule of a later phase must check +# tx.app_drupal itself. +SecRule &TX:app_drupal "@eq 0" \ + "id:900790,phase:1,pass,nolog,skipAfter:END_DRUPAL" + # Every admin page's filters and autocompletes, and the page to return to: # title: /admin/content?title=python3 # name: /admin/content/media?name=python3 @@ -237,3 +244,5 @@ SecRule REQUEST_FILENAME "@rx ^(?:/[a-z]{2,3}(?:-[a-z]+)?)?/user/password$" \ SecRule &ARGS_GET:json.mail "@eq 0" \ "ctl:ruleRemoveTargetById=932120;ARGS:json.mail,\ ctl:ruleRemoveTargetById=932260;ARGS:json.mail" + +SecMarker END_DRUPAL diff --git a/internal/coreruleset/drupal_test.go b/internal/coreruleset/drupal_test.go index 35aaa7e..8ec9f1b 100644 --- a/internal/coreruleset/drupal_test.go +++ b/internal/coreruleset/drupal_test.go @@ -5,6 +5,8 @@ import ( "net/url" "slices" "testing" + + "sneak.berlin/go/smallwebwaf/internal/coreruleset" ) // drupalThreshold is SWW_WAF_ANOMALY_THRESHOLD's default. @@ -34,7 +36,35 @@ const drupalName = "name=" func TestOrdinaryDrupalRequestsPass(t *testing.T) { t.Parallel() - wantRecordedRequestsPass(t, "drupal-requests.txt") + wantRecordedRequestsPass(t, "drupal-requests.txt", atDefaults(t), readingBodies(t)) +} + +func TestDrupalFixesApplyOnlyToAHostnameThatServesDrupal(t *testing.T) { + t.Parallel() + + crs := servingApps(t, map[string][]string{ + giteaHost: {gitea}, otherHost: {"drupal"}, + }) + + // A search, which only Drupal's fixes let through. + search := get("/search?keys=python3") + + wantResult(t, crs, withHost(search, otherHost), coreruleset.Result{}) + wantResult(t, crs, withHost(search, giteaHost), matched(932260)) +} + +func TestHostnameListingDrupalLetsItsRecordedRequestsThrough(t *testing.T) { + t.Parallel() + + // The requests were recorded on example.com. + apps := map[string][]string{"example.com": {"drupal"}} + + bodies := shared(t, coreruleset.Params{ + ParanoiaLevel: 1, DisabledRules: defaultDisabledRules, BodyLimit: bodyLimit, + Apps: apps, + }) + + wantRecordedRequestsPass(t, "drupal-requests.txt", servingApps(t, apps), bodies) } // drupalAttacks are an SQL injection, a path traversal and a script diff --git a/internal/coreruleset/gitea.conf b/internal/coreruleset/gitea.conf index 2736fbe..8eeca7c 100644 --- a/internal/coreruleset/gitea.conf +++ b/internal/coreruleset/gitea.conf @@ -1,8 +1,53 @@ -# The eighth change: on gitea's paths, some fields skip the rules that refused -# ordinary gitea requests there. They are the fields where gitea sends a search, -# a file name or a branch name. Each exclusion runs with the request, before the -# rules it switches off, those of phase 1 included, so this file is loaded before -# the Core Rule Set's rules. Each comment names a request the rule refused. +# gitea's fixes: some fields skip the rules that refused ordinary gitea requests. +# They are the fields where gitea sends a search, a file name or a branch name. +# Each exclusion runs with the request, before the rules it switches off, those +# of phase 1 included, so this file is loaded before the Core Rule Set's rules. +# Each comment names a request the rule refused. + +# Unless SWW_APPS lists gitea for the request's hostname, every rule below is +# skipped. Inspect sets tx.app_gitea when it does. +# The skip reaches only phase 1, so a rule of a later phase must check +# tx.app_gitea itself. +SecRule &TX:app_gitea "@eq 0" \ + "id:900590,phase:1,pass,nolog,skipAfter:END_GITEA" + +# The query parameters in which gitea sends names within a repository or its +# own records, or a page of its own site, are not checked against the lists of +# system files, shell paths and command names, on any path: +# 930120: .gitignore +# 932160: bin/docker-entrypoint +# 932260: docker-build +SecAction "id:900591,phase:1,pass,nolog,\ + ctl:ruleRemoveTargetById=930120;ARGS:path,\ + ctl:ruleRemoveTargetById=930120;ARGS:files,\ + ctl:ruleRemoveTargetById=930120;ARGS:skip-to,\ + ctl:ruleRemoveTargetById=930120;ARGS:sub_path,\ + ctl:ruleRemoveTargetById=930120;ARGS:ref,\ + ctl:ruleRemoveTargetById=930120;ARGS:sha,\ + ctl:ruleRemoveTargetById=930120;ARGS:branch,\ + ctl:ruleRemoveTargetById=930120;ARGS:workflow,\ + ctl:ruleRemoveTargetById=930120;ARGS:artifactName,\ + ctl:ruleRemoveTargetById=930120;ARGS:redirect_to,\ + ctl:ruleRemoveTargetById=932160;ARGS:path,\ + ctl:ruleRemoveTargetById=932160;ARGS:files,\ + ctl:ruleRemoveTargetById=932160;ARGS:skip-to,\ + ctl:ruleRemoveTargetById=932160;ARGS:sub_path,\ + ctl:ruleRemoveTargetById=932160;ARGS:ref,\ + ctl:ruleRemoveTargetById=932160;ARGS:sha,\ + ctl:ruleRemoveTargetById=932160;ARGS:branch,\ + ctl:ruleRemoveTargetById=932160;ARGS:workflow,\ + ctl:ruleRemoveTargetById=932160;ARGS:artifactName,\ + ctl:ruleRemoveTargetById=932160;ARGS:redirect_to,\ + ctl:ruleRemoveTargetById=932260;ARGS:path,\ + ctl:ruleRemoveTargetById=932260;ARGS:files,\ + ctl:ruleRemoveTargetById=932260;ARGS:skip-to,\ + ctl:ruleRemoveTargetById=932260;ARGS:sub_path,\ + ctl:ruleRemoveTargetById=932260;ARGS:ref,\ + ctl:ruleRemoveTargetById=932260;ARGS:sha,\ + ctl:ruleRemoveTargetById=932260;ARGS:branch,\ + ctl:ruleRemoveTargetById=932260;ARGS:workflow,\ + ctl:ruleRemoveTargetById=932260;ARGS:artifactName,\ + ctl:ruleRemoveTargetById=932260;ARGS:redirect_to" # gitea's searches of code, of issues and pull requests, and of commits send the # text searched for in q, which skips these rules there: @@ -160,8 +205,8 @@ SecRule REQUEST_FILENAME \ # 942140: information_schema.sql # 944110: java/lang/Runtime.java # 944130: java.lang.String.md -# Each request skips the rules that refused it. The fourth change already has -# path and redirect_to skip 930120, 932160 and 932260 on every path. +# Each request skips the rules that refused it. Rule 900591 already has path and +# redirect_to skip 930120, 932160 and 932260 on every path. # A file's commit graph, in file. SecRule REQUEST_FILENAME "@rx ^/[^/]+/[^/]+/graph$" \ @@ -286,3 +331,5 @@ SecRule REQUEST_FILENAME \ "@rx ^/api/actions_pipeline/_apis/pipelines/workflows/[0-9]+/artifacts/[^/]+/upload$" \ "id:900640,phase:1,pass,nolog,\ ctl:ruleRemoveTargetById=920450;REQUEST_HEADERS_NAMES:Content-Range" + +SecMarker END_GITEA diff --git a/internal/coreruleset/mediawiki.conf b/internal/coreruleset/mediawiki.conf index 590a63a..9dc7f57 100644 --- a/internal/coreruleset/mediawiki.conf +++ b/internal/coreruleset/mediawiki.conf @@ -1,4 +1,4 @@ -# The eleventh change: on MediaWiki's paths, some fields skip the rules that +# MediaWiki's fixes: on MediaWiki's paths, some fields skip the rules that # refused ordinary MediaWiki requests there. They are the fields where MediaWiki # sends a search, a title, a name, a file name, an address, a summary or a # reason, and the fields that hold wikitext. Each exclusion runs with the @@ -21,6 +21,13 @@ # 932250: ls -la # 932260: python3 +# Unless SWW_APPS lists mediawiki for the request's hostname, every rule below +# is skipped. Inspect sets tx.app_mediawiki when it does. +# The skip reaches only phase 1, so a rule of a later phase must check +# tx.app_mediawiki itself. +SecRule &TX:app_mediawiki "@eq 0" \ + "id:900890,phase:1,pass,nolog,skipAfter:END_MEDIAWIKI" + # The pages and their titles, the searches, and the special pages' fields # for a page, a user, a file or an address: # title: /index.php?title=Python3_notes&action=history @@ -198,10 +205,12 @@ SecRule REQUEST_FILENAME "@rx ^/(?:w/)?index\.php(?:$|/.)|^/wiki/" \ # This rule is in phase 2, because a phase 1 rule runs before the body is read # and cannot see what wpTitles holds. It checks that the body sends wpTitles # once, since it would otherwise skip these rules for a second wpTitles -# holding a command. +# holding a command. It checks tx.app_mediawiki, since the skip above does not +# reach phase 2. SecRule REQUEST_FILENAME "@rx ^/(?:w/)?index\.php(?:$|/.)|^/wiki/" \ "id:900916,phase:2,pass,nolog,chain" SecRule REQUEST_FILENAME|ARGS_GET_NAMES "@rx ^/(?:w/)?index\.php/.|^/wiki/|^title$" "chain" + SecRule &TX:app_mediawiki "@eq 1" "chain" SecRule &ARGS_GET:wpTitles "@eq 0" "chain" SecRule &ARGS_POST:wpTitles "@eq 1" "chain" SecRule ARGS_POST:wpTitles "!@rx [;|&`{]" \ @@ -210,3 +219,5 @@ SecRule REQUEST_FILENAME "@rx ^/(?:w/)?index\.php(?:$|/.)|^/wiki/" \ ctl:ruleRemoveTargetById=932235;ARGS:wpTitles,\ ctl:ruleRemoveTargetById=932260;ARGS:wpTitles,\ ctl:ruleRemoveTargetById=932380;ARGS:wpTitles" + +SecMarker END_MEDIAWIKI diff --git a/internal/coreruleset/mediawiki_test.go b/internal/coreruleset/mediawiki_test.go index d1d2d20..ea249a2 100644 --- a/internal/coreruleset/mediawiki_test.go +++ b/internal/coreruleset/mediawiki_test.go @@ -4,11 +4,16 @@ import ( "net/http" "slices" "testing" + + "sneak.berlin/go/smallwebwaf/internal/coreruleset" ) // mediawikiThreshold is SWW_WAF_ANOMALY_THRESHOLD's default. const mediawikiThreshold = 5 +// mediawiki is MediaWiki's name in SWW_APPS. +const mediawiki = "mediawiki" + // mediawikiEdit is the address MediaWiki's edit form posts to. const mediawikiEdit = "/index.php?title=Example&action=submit" @@ -45,7 +50,58 @@ const ( func TestOrdinaryMediaWikiRequestsPass(t *testing.T) { t.Parallel() - wantRecordedRequestsPass(t, "mediawiki-requests.txt") + wantRecordedRequestsPass(t, "mediawiki-requests.txt", atDefaults(t), readingBodies(t)) +} + +func TestMediaWikiFixesApplyOnlyToAHostnameThatServesMediaWiki(t *testing.T) { + t.Parallel() + + crs := servingApps(t, map[string][]string{ + giteaHost: {gitea}, otherHost: {mediawiki}, + }) + + // A search, which only MediaWiki's fixes let through. + search := get("/index.php?title=Special:Search&search=python3") + + wantResult(t, crs, withHost(search, otherHost), coreruleset.Result{}) + wantResult(t, crs, withHost(search, giteaHost), matched(932260)) +} + +func TestMediaWikiWatchlistFixAppliesOnlyToAHostnameThatServesMediaWiki(t *testing.T) { + t.Parallel() + + crs := shared(t, coreruleset.Params{ + ParanoiaLevel: 1, DisabledRules: defaultDisabledRules, BodyLimit: bodyLimit, + Apps: map[string][]string{giteaHost: {gitea}, otherHost: {mediawiki}}, + }) + + // A watchlist that only the phase 2 rule 900916 lets through. + r := request{http.MethodPost, mediawikiWatchlist.target, []string{formData}} + body := mediawikiWatchlist.field + "Lighthouse%20notes%0D%0APython3%20notes" + + got, _ := inspectBody(t, crs, withHost(r, otherHost), body) + if got.Score >= mediawikiThreshold { + t.Errorf("POST %s %s on %s refused: %+v", r.target, body, otherHost, got) + } + + got, _ = inspectBody(t, crs, withHost(r, giteaHost), body) + if got.Score < mediawikiThreshold { + t.Errorf("POST %s %s on %s let through: %+v", r.target, body, giteaHost, got) + } +} + +func TestHostnameListingMediaWikiLetsItsRecordedRequestsThrough(t *testing.T) { + t.Parallel() + + // The requests were recorded on example.com. + apps := map[string][]string{"example.com": {mediawiki}} + + bodies := shared(t, coreruleset.Params{ + ParanoiaLevel: 1, DisabledRules: defaultDisabledRules, BodyLimit: bodyLimit, + Apps: apps, + }) + + wantRecordedRequestsPass(t, "mediawiki-requests.txt", servingApps(t, apps), bodies) } // mediawikiQueryFields and mediawikiBodyFields together hold every field that diff --git a/internal/coreruleset/wordpress.conf b/internal/coreruleset/wordpress.conf index da98bab..f97c9dc 100644 --- a/internal/coreruleset/wordpress.conf +++ b/internal/coreruleset/wordpress.conf @@ -1,10 +1,17 @@ -# The ninth change: on WordPress's paths, some fields skip the rules that refused +# WordPress's fixes: on WordPress's paths, some fields skip the rules that refused # ordinary WordPress requests there. Each exclusion runs with the request, before # the rules it switches off, so this file is loaded before the Core Rule Set's # rules. Each comment names a request the rule refused. An exclusion on a field # of a body applies only when the query has no parameter of that name, so it # matters only while SWW_WAF_BODY_LIMIT is set. +# Unless SWW_APPS lists wordpress for the request's hostname, every rule below is +# skipped. Inspect sets tx.app_wordpress when it does. +# The skip reaches only phase 1, so a rule of a later phase must check +# tx.app_wordpress itself. +SecRule &TX:app_wordpress "@eq 0" \ + "id:900690,phase:1,pass,nolog,skipAfter:END_WORDPRESS" + # WordPress takes a call to its REST API at any of four addresses: # /wp-json/wp/v2/posts/1 # /index.php/wp-json/wp/v2/posts/1 @@ -195,3 +202,5 @@ SecRule REQUEST_FILENAME "@rx ^/(?:(?:index\.php/)?page/[0-9]+/?|index\.php)?$" # so skips 932340 there. SecRule REQUEST_FILENAME "@streq /wp-admin/upload.php" \ "id:900711,phase:1,pass,nolog,ctl:ruleRemoveTargetById=932340;ARGS:_wp_http_referer" + +SecMarker END_WORDPRESS diff --git a/internal/discriminator/coreruleset_test.go b/internal/discriminator/coreruleset_test.go index b2196be..df1e062 100644 --- a/internal/discriminator/coreruleset_test.go +++ b/internal/discriminator/coreruleset_test.go @@ -10,6 +10,7 @@ import ( "testing" "sneak.berlin/go/smallwebwaf/internal/alerts" + "sneak.berlin/go/smallwebwaf/internal/discriminator" "sneak.berlin/go/smallwebwaf/internal/ratelimit" "sneak.berlin/go/smallwebwaf/internal/requestlog" ) @@ -21,6 +22,7 @@ const ( wafDisabledRules = "SWW_WAF_DISABLED_RULES" wafExemptPaths = "SWW_WAF_EXEMPT_PATHS" wafBodyLimit = "SWW_WAF_BODY_LIMIT" + apps = "SWW_APPS" block = "block" detect = "detect" ) @@ -290,6 +292,27 @@ func TestDisabledRulesSwitchOffWhatGiteaWouldBeRefused(t *testing.T) { } } +func TestGiteasFixesApplyOnlyToAHostnameThatServesGitea(t *testing.T) { + t.Parallel() + + // Only gitea's fixes let a file name through in path, a parameter + // gitea sends file names in. + const fileName = "/?path=.gitignore" + + s := startDiscriminator(t, "", map[string]string{ + wafMode: block, apps: siteA + ":gitea", + }) + + wantDisposition(t, do(t, s.question(siteA, client, fileName)), + http.StatusOK, discriminator.DispositionAllow) + wantDisposition(t, do(t, s.question(siteB, client, fileName)), + http.StatusForbidden, discriminator.DispositionDeny) + + lines := s.out.requestLines(t, 2) + wantLine(t, lines[0], http.StatusOK, requestlog.ActionForward) + wantLine(t, lines[1], http.StatusForbidden, requestlog.ActionBlockedByCoreRuleSet) +} + func TestAttackInAFormBodyIsRefusedOnlyWhileBodiesAreRead(t *testing.T) { t.Parallel() diff --git a/internal/discriminator/proxy.go b/internal/discriminator/proxy.go index 3e6215c..e269221 100644 --- a/internal/discriminator/proxy.go +++ b/internal/discriminator/proxy.go @@ -221,7 +221,8 @@ func newReputation( // NewCoreRuleSet returns the Core Rule Set at SWW_WAF_PARANOIA_LEVEL, // without the rules SWW_WAF_DISABLED_RULES switches off, reading bodies // up to SWW_WAF_BODY_LIMIT, with the exclusions in -// SWW_CORE_RULE_SET_EXCLUSIONS_FILE, or nil while SWW_WAF_MODE is off. It +// SWW_CORE_RULE_SET_EXCLUSIONS_FILE and the fixes of the apps in SWW_APPS, +// or nil while SWW_WAF_MODE is off. It // reads at most QuestionBodyMaxBytes less one byte: a question's body of // QuestionBodyMaxBytes may be cut short, and the Core Rule Set does not read a // JSON or XML body over its limit. An exclusions file that coreruleset.New refuses is @@ -236,6 +237,7 @@ func NewCoreRuleSet(cfg *config.Config) (*coreruleset.CoreRuleSet, error) { ParanoiaLevel: cfg.WAFParanoiaLevel, DisabledRules: cfg.WAFDisabledRules, BodyLimit: min(cfg.WAFBodyLimit, QuestionBodyMaxBytes-1), ExclusionsFile: cfg.CoreRuleSetExclusionsFile, + Apps: cfg.Apps, }) if err != nil { return nil, fmt.Errorf("SWW_CORE_RULE_SET_EXCLUSIONS_FILE: %w", err)