These attacks exploit the fact that browsers allow scripts to send requests without doing any CORS preflight check when they:

* don't have a `Content-Type` header (e.g. using `fetch()` with a `Blob` body)
* and don't send any authentication credentials.

This type of attack is mainly relevant when:

    # Return some error
    ...
```

But by using the `secrets.compare_digest()` it will be secure against a type of attacks called "timing attacks".

### Timing Attacks { #timing-attacks }

But what's a "timing attack"?

Let's imagine some attackers are trying to guess the username and password.

And they send a request with a username `johndoe` and a password `love123`.

* plug-in: plug-in (do not translate to "complemento" or "extensión")
* plug-ins: plug-ins (do not translate to "complementos" or "extensiones")
* full stack: full stack (do not translate to "pila completa")
* full-stack: full-stack (do not translate to "de pila completa")
* stack: stack (do not translate to "pila")
* loop (as in async loop): loop (do not translate to "bucle" or "ciclo")

```Python
if not (credentials.username == "stanleyjobson") or not (credentials.password == "swordfish"):
    # 어떤 오류를 반환
    ...
```

하지만 `secrets.compare_digest()`를 사용하면 "timing attacks"라고 불리는 한 유형의 공격에 대해 안전해집니다.

### 타이밍 공격 { #timing-attacks }

그렇다면 "timing attack"이란 무엇일까요?

공격자들이 사용자명과 비밀번호를 추측하려고 한다고 가정해봅시다.

그리고 사용자명 `johndoe`, 비밀번호 `love123`으로 요청을 보냅니다.

그러면 애플리케이션의 Python 코드는 대략 다음과 같을 것입니다:

```Python

mitigate the risk in this scenario is to sandbox the model execution.

### Hardware attacks

Physical GPUs or TPUs can also be the target of attacks. [Published
research](https://scholar.google.com/scholar?q=gpu+side+channel) shows that it
might be possible to use side channel attacks on the GPU to leak data from other
running models or processes in the same system. GPUs can also have

 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 */
package jcifs.smb;

import jcifs.CIFSException;

/**
 * Exception thrown when an SMB protocol downgrade attack is detected.
 * Indicates that the negotiated protocol version is lower than expected or required.
 *
 * @author mbechler
 *
 */
public class SMBProtocolDowngradeException extends CIFSException {

    /**

* mount (verb): монтировать
* mount (noun): точка монтирования / mount (keep in English if it's a FastAPI keyword)
* plugin: плагин
* plug-in: плагин
* full stack: full stack (do not translate)
* full-stack: full-stack (do not translate)
* loop (as in async loop): цикл событий
* Machine Learning: Машинное обучение
* Deep Learning: Глубокое обучение
* callback hell: callback hell (clarify as `ад обратных вызовов`)

- 没有 `Content-Type` 头（例如使用 `fetch()` 携带 `Blob` 作为 body）
- 且不发送任何认证凭据。

这种攻击主要在以下情况下相关：

- 应用在本地（如 `localhost`）或内网中运行
- 且应用没有任何认证，假定来自同一网络的请求都可信。

## 攻击示例 { #example-attack }

假设你构建了一个本地运行的 AI 代理。

它提供了一个 API，地址为

```
http://localhost:8000/v1/agents/multivac
```

另有一个前端，地址为

```
http://localhost:8000
```

/// tip | 提示

注意它们的主机相同。

///

#### 反应时间对攻击者的帮助 { #the-time-to-answer-helps-the-attackers }

通过服务器花费了更多微秒才发送**错误的用户或密码**响应，攻击者会知道猜对了一些内容，起码开头字母是正确的。

然后，他们就可以放弃 `johndoe`，再用类似 `stanleyjobsox` 的内容进行尝试。

#### **专业**攻击 { #a-professional-attack }

当然，攻击者不用手动操作，而是编写每秒能执行成千上万次测试的攻击程序，每次都会找到更多正确字符。

但是，在您的应用的**帮助**下，攻击者利用时间差，就能在几分钟或几小时内，以这种方式猜出正确的用户名和密码。

#### 使用 `secrets.compare_digest()` 修补 { #fix-it-with-secrets-compare-digest }

#### 回應時間幫了攻擊者 { #the-time-to-answer-helps-the-attackers }

此時，透過觀察伺服器回覆「Incorrect username or password」多花了幾個微秒，攻擊者就知道他們有某些地方猜對了，前幾個字母是正確的。

接著他們會再嘗試，知道它更可能接近 `stanleyjobsox` 而不是 `johndoe`。

#### 「專業」的攻擊 { #a-professional-attack }

當然，攻擊者不會手動嘗試這一切，他們會寫程式來做，可能每秒進行上千或上百萬次測試，一次只多猜中一個正確字母。

但這樣做，幾分鐘或幾小時內，他們就能在我們應用程式「協助」下，僅靠回應時間就猜出正確的使用者名稱與密碼。

#### 用 `secrets.compare_digest()` 修正 { #fix-it-with-secrets-compare-digest }