open
https://gitlab.synchro.net/main/sbbs/-/work_items/1355
The web server's test that a request stays inside the web root is a
prefix comparison with no path-separator check, so a directory beside
the root whose name begins with the root's name is served.
**Reproduced** on Linux (current master, scratch server on loopback):
with `RootDirectory` = `<dir>/root` and a sibling `<dir>/root2`
containing `sib.txt`, each of these returns 200 and the file:
```
GET /../root2/sib.txt
GET /%2e%2e/root2/sib.txt
GET /plain/../../root2/sib.txt
```
A request for a directory outside the root with an unrelated name (`/../apps/secret.txt`) is correctly refused with 400 and logged as
"outside of web root".
So the exposure needs a sibling such as `html2`, `html.bak` or
`html-old` next to a root named `html`. Nothing else was reachable: `webctrl.ini` access rules still applied to every path tried, and
`webctrl.ini` itself was not served.
**Cause** (`src/sbbs3/websrvr.cpp`): the containment test is
`strnicmp(path, root_dir, strlen(root_dir))`. It should also require
that the path ends there or continues with a path separator. The coming
`[dir]` section of `web_alias.ini` has a helper that does this for its directories.
**Related hardening, not reproduced:**
1. `is_legal_path()` refuses nothing on non-Windows builds, so `..` is
accepted in request paths there and the root test above is the only
containment.
2. The root test is case-insensitive on every platform; on a
case-sensitive file system a sibling differing only in case would
pass.
3. The "never serve webctrl.ini" test is a case-sensitive `strcmp`.
On Linux a differently-cased name is simply another file; on
Windows `/WEBCTRL.INI` may be served. Not tested on Windows.
4. The CGI `PATH_TRANSLATED` value is built with an unbounded
`sprintf` (CGI is disabled by default with `NO_CGI`). From reading
the source only.
-- *Authored by Claude (Claude Code), on behalf of @rswindell*
--- SBBSecho 3.38-Linux
* Origin: Vertrauen - [vert/cvs/bbs].synchro.net (1:103/705)