
CVE-2026-77818 - Yordam Kütüphane Otomasyon Sistemi - Üç ayrı noktada yansıtılmış HTML enjeksiyonu, form action ele geçirme ve kimlik bilgisi hırsızlığı (CWE-79)
CVE-2026-77818 · CVSS 3.1 6.1 (Medium) · Cybersecurity Presidency · Published 2026-09-04 · TR-26-1011
Status: The vulnerabilities have been fixed in version v22.2. Affected installations must be upgraded to v22.2 or later.
Yordam Library Automation System is a commercial library automation and online catalog (OPAC) software widely used in university, public, and institutional libraries in Turkey. The installation is on-premise; a separate copy runs at each customer organization.
In version v22.1 of the product, there are reflected HTML injections at three separate points that are independent of each other. All three require no authentication, and all three are triggered by a single link.
| # | Point | Root cause |
|---|---|---|
| 1 | Login page, devam parameter | No escaping applied at all |
| 2 | Hidden form field value attribute | URL decode a second time after escaping |
| 3 | Hidden form field name attribute | Escaping applied only to the value, not to the name |
Since all three are in the same version of the same product and in the same vulnerability class, they have been grouped under a single advisory and published under a single CVE ID. In terms of impact, the most severe is point number 1.
devam ParameterThis is the most critical one. The injection point is directly the HTML tag of the authentication form itself.
The devam parameter carries the address the user will return to after logging in and comes in hex-encoded — the value 2f796f7264616d2f means /yordam/. The application decodes this value from hex and writes it into the opening tag of the login form. There is no escaping operation in between:
<form class='girisForm collapse show ikiAdimliGiris' method='post'
action='inc/islem.fm.inc.php'
data-url='<HEX ÇÖZÜLMÜŞ KULLANICI GİRDİSİ>'
autocomplete="off">
The <, > and quote characters appear raw in the output. The only thing holding the payload is the data-url attribute being wrapped in single quotes. When a single quote is placed inside the input, it terminates too: the attribute closes, the <form> tag closes, and the HTML written by the attacker replaces the page's authentication form.
The form action is hijacked. No fake form is drawn here — the application's own form is left empty and closed, and immediately after it a new <form> carrying the same CSS classes is opened. The username, password and verification code fields on the page all remain inside this new form because they are the application's original HTML. The user sees the real form, fills in the real form; the information they enter goes to the attacker's server. There is no difference that can be distinguished visually.
The critical point: the injection occurs not on a random page, but on the page where the user is already expected to enter their password. In an ordinary reflected injection, the attacker must convince the victim; here, the application's own interface does the convincing.
value Attribute — Double URL DecodingOn the search page, the values of GET parameters are written into hidden form fields. At this point escaping is applied — but in the wrong order.
The same q value is used in three separate contexts within a single response, and each has a different decoding depth:
| Context | Decoding | Status |
|---|---|---|
JS string in <script> block | 1 time | Safe |
Main search box <input value="…"> | 1 time | Safe |
Hidden form fields <input type='hidden' value="…"> | 2 times | Vulnerable |
The order of operations is as follows:
İstemci girdisi : %2522
↓ $_GET çözümlemesi
PHP değişkeni : %22
↓ giriş filtresi → zararlı içerik görmüyor, ortada tırnak yok
↓ htmlspecialchars → kaçıracak karakter yok, değişiklik olmuyor
↓ urldecode → %22 çözülüyor
Sayfaya basılan : " ← ham tırnak, attribute'tan çıkış
The input filter and escape run in the first decoding layer, while the output is fed from the second layer. When the single- and double-encoded versions of the same payload are compared, the difference is clear:
| Sent | Response | Hidden field output |
|---|---|---|
q=foo%22… (single encoding) | 302 Found | value="foo"…" — the filter catches it |
q=foo%2522… (double encoding) | 200 OK | value="foo"><…>" — raw HTML |
The vulnerability is not specific to the q parameter. The block that generates the hidden field iterates over all GET parameters in the request; it has also been verified separately on tip and alan.
name Attribute — Injection in the Parameter NameThe same block produces the following structure for each GET parameter:
<input type='hidden' name="<PARAMETRE ADI>" value="<PARAMETRE DEĞERİ>"/>
Escape is applied only to the value side. It is never applied to the name side. At this point there is no need for double encoding either — single encoding is sufficient, because there is no escape to bypass in the first place.
A fabricated parameter name is written directly into the name attribute as raw text and can break out of the attribute. Since the parameter name is under attacker control, it does not even need to be a parameter the application recognizes.
Items 2 and 3 of these three points originate from the same code block, and this block repeats in six separate forms: dilForm, adetForm, siralaForm, tkForm, ekForm, tmForm. That is, in a single request, injection occurs six times.
That the block is dynamic has been verified by comparing the output of two requests:
İstek A: ?p=1&dil=0&alan=&tip=basit&gorunum=liste&q=…
Çıktı A: name="p" · name="alan" · name="tip" · name="gorunum" · name="q"
İstek B: ?p=2&dil=0&devam=…
Çıktı B: name="p" · name="devam"
The generated fields are not derived from a fixed list, but directly from the parameters in the request. Therefore, both the content written to the name and value attributes is under attacker control.
All the attacker needs is a link for the victim to open. No login, no account required.
action target is redirected to the attacker. Confirmed.Since the affected platform hosts library members' credentials and personal data, access to member records becomes possible through compromised accounts.
The application uses CSP, and script-src and object-src are nonce-based; that is, classic script-based XSS does not run on these pages. At first glance, this seems to reduce the finding to the level of "content tampering only."
The full policy is as follows:
Content-Security-Policy: script-src 'nonce-...'; object-src 'nonce-...'; frame-ancestors 'self'
No form-action. No default-src either — therefore there is no default to fall back on for undefined directives. Result: the form POSTing to the attacker's server is not blocked by the browser in any way.
Running JavaScript is not required to steal credentials. Plain HTML is enough, and CSP does not stop plain HTML.
Primary
Related