urllib3 known bugs

urllib3 doesn't treat the `Cookie` HTTP header special or provide any helpers for managing cookies over HTTP, that is the responsibility of the user. However, it is possible for a user to specify a `Cookie` header and unknowingly leak information via HTTP redirects to a different origin if that user doesn't disable redirects explicitly. Users **must** handle redirects themselves instead of relying on urllib3's automatic redirects to achieve safe processing of the `Cookie` header, thus we decided to strip the header by default in order to further protect users who aren't using the correct approach. ## Affected usages We believe the number of usages affected by this advisory is low. It requires all of the following to be true to be exploited: * Using an affected version of urllib3 (patched in v1.26.17 and v2.0.6) * Using the `Cookie` header on requests, which is mostly typical for impersonating a browser. * Not disabling HTTP redirects * Either not using HTTPS or for the origin server to redirect to a malicious origin. ## Remediation * Upgrading to at least urllib3 v1.26.17 or v2.0.6 * Disabling HTTP redirects using `redirects=False` when sending requests. * Not using the `Cookie` header.

### Impact When provided with a URL containing many `@` characters in the authority component the authority regular expression exhibits catastrophic backtracking causing a denial of service if a URL were passed as a parameter or redirected to via an HTTP redirect. ### Patches The issue has been fixed in urllib3 v1.26.5. ### References - [CVE-2021-33503](https://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2021-33503) - [JVNVU#92413403 (English)](https://jvn.jp/en/vu/JVNVU92413403/) - [JVNVU#92413403 (Japanese)](https://jvn.jp/vu/JVNVU92413403/) - [urllib3 v1.26.5](https://github.com/urllib3/urllib3/releases/tag/1.26.5) ### For more information If you have any questions or comments about this advisory: * Ask in our [community Discord](https://discord.gg/urllib3) * Email [[email protected]](mailto:[email protected])

The urllib3 library before 1.24.2 for Python mishandles certain cases where the desired set of CA certificates is different from the OS store of CA certificates, which results in SSL connections succeeding in situations where a verification failure is the correct outcome. This is related to use of the `ssl_context`, `ca_certs`, or `ca_certs_dir` argument.

The _encode_invalid_chars function in util/url.py in the urllib3 library 1.25.2 through 1.25.7 for Python allows a denial of service (CPU consumption) because of an inefficient algorithm. The percent_encodings array contains all matches of percent encodings. It is not deduplicated. For a URL of length N, the size of percent_encodings may be up to O(N). The next step (normalize existing percent-encoded bytes) also takes up to O(N) for each step, so the total time is O(N^2). If percent_encodings were deduplicated, the time to compute _encode_invalid_chars would be O(kN), where k is at most 484 ((10+6*2)^2).

## Impact urllib3 supports chained HTTP encoding algorithms for response content according to RFC 9110 (e.g., `Content-Encoding: gzip, zstd`). However, the number of links in the decompression chain was unbounded allowing a malicious server to insert a virtually unlimited number of compression steps leading to high CPU usage and massive memory allocation for the decompressed data. ## Affected usages Applications and libraries using urllib3 version 2.5.0 and earlier for HTTP requests to untrusted sources unless they disable content decoding explicitly. ## Remediation Upgrade to at least urllib3 v2.6.0 in which the library limits the number of links to 5. If upgrading is not immediately possible, use [`preload_content=False`](https://urllib3.readthedocs.io/en/2.5.0/advanced-usage.html#streaming-and-i-o) and ensure that `resp.headers["content-encoding"]` contains a safe number of encodings before reading the response content.

### Impact urllib3's [streaming API](https://urllib3.readthedocs.io/en/2.6.2/advanced-usage.html#streaming-and-i-o) is designed for the efficient handling of large HTTP responses by reading the content in chunks, rather than loading the entire response body into memory at once. urllib3 can perform decoding or decompression based on the HTTP `Content-Encoding` header (e.g., `gzip`, `deflate`, `br`, or `zstd`). When using the streaming API, the library decompresses only the necessary bytes, enabling partial content consumption. However, for HTTP redirect responses, the library would read the entire response body to drain the connection and decompress the content unnecessarily. This decompression occurred even before any read methods were called, and configured read limits did not restrict the amount of decompressed data. As a result, there was no safeguard against decompression bombs. A malicious server could exploit this to trigger excessive resource consumption on the client (high CPU usage and large memory allocations for decompressed data; CWE-409). ### Affected usages Applications and libraries using urllib3 version 2.6.2 and earlier to stream content from untrusted sources by setting `preload_content=False` when they do not disable redirects. ### Remediation Upgrade to at least urllib3 v2.6.3 in which the library does not decode content of redirect responses when `preload_content=False`. If upgrading is not immediately possible, disable [redirects](https://urllib3.readthedocs.io/en/2.6.2/user-guide.html#retrying-requests) by setting `redirect=False` for requests to untrusted source.

### Impact urllib3's [streaming API](https://urllib3.readthedocs.io/en/2.5.0/advanced-usage.html#streaming-and-i-o) is designed for the efficient handling of large HTTP responses by reading the content in chunks, rather than loading the entire response body into memory at once. When streaming a compressed response, urllib3 can perform decoding or decompression based on the HTTP `Content-Encoding` header (e.g., `gzip`, `deflate`, `br`, or `zstd`). The library must read compressed data from the network and decompress it until the requested chunk size is met. Any resulting decompressed data that exceeds the requested amount is held in an internal buffer for the next read operation. The decompression logic could cause urllib3 to fully decode a small amount of highly compressed data in a single operation. This can result in excessive resource consumption (high CPU usage and massive memory allocation for the decompressed data; CWE-409) on the client side, even if the application only requested a small chunk of data. ### Affected usages Applications and libraries using urllib3 version 2.5.0 and earlier to stream large compressed responses or content from untrusted sources. `stream()`, `read(amt=256)`, `read1(amt=256)`, `read_chunked(amt=256)`, `readinto(b)` are examples of `urllib3.HTTPResponse` method calls using the affected logic unless decoding is disabled explicitly. ### Remediation Upgrade to at least urllib3 v2.6.0 in which the library avoids decompressing data that exceeds the requested amount. If your environment contains a package facilitating the Brotli encoding, upgrade to at least Brotli 1.2.0 or brotlicffi 1.2.0.0 too. These versions are enforced by the `urllib3[brotli]` extra in the patched versions of urllib3. ### Credits The issue was reported by @Cycloctane. Supplemental information was provided by @stamparm during a security audit performed by [7ASecurity](https://7asecurity.com/) and facilitated by [OSTIF](https://ostif.org/).

urllib3 is a user-friendly HTTP client library for Python. urllib3 previously wouldn't remove the HTTP request body when an HTTP redirect response using status 301, 302, or 303 after the request had its method changed from one that could accept a request body (like `POST`) to `GET` as is required by HTTP RFCs. Although this behavior is not specified in the section for redirects, it can be inferred by piecing together information from different sections and we have observed the behavior in other major HTTP client implementations like curl and web browsers. Because the vulnerability requires a previously trusted service to become compromised in order to have an impact on confidentiality we believe the exploitability of this vulnerability is low. Additionally, many users aren't putting sensitive data in HTTP request bodies, if this is the case then this vulnerability isn't exploitable. Both of the following conditions must be true to be affected by this vulnerability: 1. Using urllib3 and submitting sensitive information in the HTTP request body (such as form data or JSON) and 2. The origin service is compromised and starts redirecting using 301, 302, or 303 to a malicious peer or the redirected-to service becomes compromised. This issue has been addressed in versions 1.26.18 and 2.0.7 and users are advised to update to resolve this issue. Users unable to update should disable redirects for services that aren't expecting to respond with redirects with `redirects=False` and disable automatic redirects with `redirects=False` and handle 301, 302, and 303 redirects manually by stripping the HTTP request body.

urllib3 before 1.24.2 does not remove the authorization HTTP header when following a cross-origin redirect (i.e., a redirect that differs in host, port, or scheme). This can allow for credentials in the authorization header to be exposed to unintended hosts or transmitted in cleartext. NOTE: this issue exists because of an incomplete fix for CVE-2018-20060 (which was case-sensitive).

urllib3 is a user-friendly HTTP client library for Python. urllib3 doesn't treat the `Cookie` HTTP header special or provide any helpers for managing cookies over HTTP, that is the responsibility of the user. However, it is possible for a user to specify a `Cookie` header and unknowingly leak information via HTTP redirects to a different origin if that user doesn't disable redirects explicitly. This issue has been patched in urllib3 version 1.26.17 or 2.0.5.

The urllib3 library 1.26.x before 1.26.4 for Python omits SSL certificate validation in some cases involving HTTPS to HTTPS proxies. The initial connection to the HTTPS proxy (if an SSLContext isn't given via proxy_config) doesn't verify the hostname of the certificate. This means certificates for different servers that still validate properly with the default urllib3 SSLContext will be silently accepted.

An issue was discovered in urllib3 before 1.26.5. When provided with a URL containing many @ characters in the authority component, the authority regular expression exhibits catastrophic backtracking, causing a denial of service if a URL were passed as a parameter or redirected to via an HTTP redirect.

The _encode_invalid_chars function in util/url.py in the urllib3 library 1.25.2 through 1.25.7 for Python allows a denial of service (CPU consumption) because of an inefficient algorithm. The percent_encodings array contains all matches of percent encodings. It is not deduplicated. For a URL of length N, the size of percent_encodings may be up to O(N). The next step (normalize existing percent-encoded bytes) also takes up to O(N) for each step, so the total time is O(N^2). If percent_encodings were deduplicated, the time to compute _encode_invalid_chars would be O(kN), where k is at most 484 ((10+6*2)^2).

urllib3 before 1.25.9 allows CRLF injection if the attacker controls the HTTP request method, as demonstrated by inserting CR and LF control characters in the first argument of putrequest(). NOTE: this is similar to CVE-2020-26116.

The urllib3 library before 1.24.2 for Python mishandles certain cases where the desired set of CA certificates is different from the OS store of CA certificates, which results in SSL connections succeeding in situations where a verification failure is the correct outcome. This is related to use of the ssl_context, ca_certs, or ca_certs_dir argument.

In the urllib3 library through 1.24.1 for Python, CRLF injection is possible if the attacker controls the request parameter.

urllib3 before version 1.23 does not remove the Authorization HTTP header when following a cross-origin redirect (i.e., a redirect that differs in host, port, or scheme). This can allow for credentials in the Authorization header to be exposed to unintended hosts or transmitted in cleartext.

Versions 1.17 and 1.18 of the Python urllib3 library suffer from a vulnerability that can cause them, in certain configurations, to not correctly validate TLS certificates. This places users of the library with those configurations at risk of man-in-the-middle and information leakage attacks. This vulnerability affects users using versions 1.17 and 1.18 of the urllib3 library, who are using the optional PyOpenSSL support for TLS instead of the regular standard library TLS backend, and who are using OpenSSL 1.1.0 via PyOpenSSL. This is an extremely uncommon configuration, so the security impact of this vulnerability is low.

urllib3 before 1.25.9 allows CRLF injection if the attacker controls the HTTP request method, as demonstrated by inserting CR and LF control characters in the first argument of `putrequest()`. NOTE: this is similar to CVE-2020-26116.

Versions 1.17 and 1.18 of the Python urllib3 library suffer from a vulnerability that can cause them, in certain configurations, to not correctly validate TLS certificates. This places users of the library with those configurations at risk of man-in-the-middle and information leakage attacks. This vulnerability affects users using versions 1.17 and 1.18 of the urllib3 library, who are using the optional PyOpenSSL support for TLS instead of the regular standard library TLS backend, and who are using OpenSSL 1.1.0 via PyOpenSSL. This is an extremely uncommon configuration, so the security impact of this vulnerability is low.

In the urllib3 library through 1.24.2 for Python, CRLF injection is possible if the attacker controls the request parameter.

urllib3 handles redirects and retries using the same mechanism, which is controlled by the `Retry` object. The most common way to disable redirects is at the request level, as follows: ```python resp = urllib3.request("GET", "https://httpbin.org/redirect/1", redirect=False) print(resp.status) # 302 ``` However, it is also possible to disable redirects, for all requests, by instantiating a `PoolManager` and specifying `retries` in a way that disable redirects: ```python import urllib3 http = urllib3.PoolManager(retries=0) # should raise MaxRetryError on redirect http = urllib3.PoolManager(retries=urllib3.Retry(redirect=0)) # equivalent to the above http = urllib3.PoolManager(retries=False) # should return the first response resp = http.request("GET", "https://httpbin.org/redirect/1") ``` However, the `retries` parameter is currently ignored, which means all the above examples don't disable redirects. ## Affected usages Passing `retries` on `PoolManager` instantiation to disable redirects or restrict their number. By default, requests and botocore users are not affected. ## Impact Redirects are often used to exploit SSRF vulnerabilities. An application attempting to mitigate SSRF or open redirect vulnerabilities by disabling redirects at the PoolManager level will remain vulnerable. ## Remediation You can remediate this vulnerability with the following steps: * Upgrade to a patched version of urllib3. If your organization would benefit from the continued support of urllib3 1.x, please contact [[email protected]](mailto:[email protected]) to discuss sponsorship or contribution opportunities. * Disable redirects at the `request()` level instead of the `PoolManager()` level.

urllib3 before 1.24.2 does not remove the authorization HTTP header when following a cross-origin redirect (i.e., a redirect that differs in host, port, or scheme). This can allow for credentials in the authorization header to be exposed to unintended hosts or transmitted in cleartext. NOTE: this issue exists because of an incomplete fix for CVE-2018-20060 (which was case-sensitive).

urllib3 previously wouldn't remove the HTTP request body when an HTTP redirect response using status 303 "See Other" after the request had its method changed from one that could accept a request body (like `POST`) to `GET` as is required by HTTP RFCs. Although the behavior of removing the request body is not specified in the section for redirects, it can be inferred by piecing together information from different sections and we have observed the behavior in other major HTTP client implementations like curl and web browsers. From [RFC 9110 Section 9.3.1](https://www.rfc-editor.org/rfc/rfc9110.html#name-get): > A client SHOULD NOT generate content in a GET request unless it is made directly to an origin server that has previously indicated, in or out of band, that such a request has a purpose and will be adequately supported. ## Affected usages Because the vulnerability requires a previously trusted service to become compromised in order to have an impact on confidentiality we believe the exploitability of this vulnerability is low. Additionally, many users aren't putting sensitive data in HTTP request bodies, if this is the case then this vulnerability isn't exploitable. Both of the following conditions must be true to be affected by this vulnerability: * If you're using urllib3 and submitting sensitive information in the HTTP request body (such as form data or JSON) * The origin service is compromised and starts redirecting using 303 to a malicious peer or the redirected-to service becomes compromised. ## Remediation You can remediate this vulnerability with any of the following steps: * Upgrade to a patched version of urllib3 (v1.26.18 or v2.0.7) * Disable redirects for services that you aren't expecting to respond with redirects with `redirects=False`. * Disable automatic redirects with `redirects=False` and handle 303 redirects manually by stripping the HTTP request body.

### Impact Users who are using an HTTPS proxy to issue HTTPS requests and haven't configured their own SSLContext via `proxy_config`. Only the default SSLContext is impacted. ### Patches [urllib3 >=1.26.4 has the issue resolved](https://github.com/urllib3/urllib3/releases/tag/1.26.4). urllib3<1.26 is not impacted due to not supporting HTTPS requests via HTTPS proxies. ### Workarounds Upgrading is recommended as this is a minor release and not likely to break current usage. Configuring an `SSLContext` with `check_hostname=True` and passing via `proxy_config` instead of relying on the default `SSLContext` ### For more information If you have any questions or comments about this advisory: * Email us at [[email protected]](mailto:[email protected])

urllib3 [supports](https://urllib3.readthedocs.io/en/2.4.0/reference/contrib/emscripten.html) being used in a Pyodide runtime utilizing the [JavaScript Fetch API](https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API) or falling back on [XMLHttpRequest](https://developer.mozilla.org/en-US/docs/Web/API/XMLHttpRequest). This means you can use Python libraries to make HTTP requests from your browser or Node.js. Additionally, urllib3 provides [a mechanism](https://urllib3.readthedocs.io/en/2.4.0/user-guide.html#retrying-requests) to control redirects. However, the `retries` and `redirect` parameters are ignored with Pyodide; the runtime itself determines redirect behavior. ## Affected usages Any code which relies on urllib3 to control the number of redirects for an HTTP request in a Pyodide runtime. ## Impact Redirects are often used to exploit SSRF vulnerabilities. An application attempting to mitigate SSRF or open redirect vulnerabilities by disabling redirects may remain vulnerable if a Pyodide runtime redirect mechanism is unsuitable. ## Remediation If you use urllib3 in Node.js, upgrade to a patched version of urllib3. Unfortunately, browsers provide no suitable way which urllib3 can use: `XMLHttpRequest` provides no control over redirects, the Fetch API returns `opaqueredirect` responses lacking data when redirects are controlled manually. Expect default browser behavior for redirects.

When using urllib3's proxy support with `ProxyManager`, the `Proxy-Authorization` header is only sent to the configured proxy, as expected. However, when sending HTTP requests *without* using urllib3's proxy support, it's possible to accidentally configure the `Proxy-Authorization` header even though it won't have any effect as the request is not using a forwarding proxy or a tunneling proxy. In those cases, urllib3 doesn't treat the `Proxy-Authorization` HTTP header as one carrying authentication material and thus doesn't strip the header on cross-origin redirects. Because this is a highly unlikely scenario, we believe the severity of this vulnerability is low for almost all users. Out of an abundance of caution urllib3 will automatically strip the `Proxy-Authorization` header during cross-origin redirects to avoid the small chance that users are doing this on accident. Users should use urllib3's proxy support or disable automatic redirects to achieve safe processing of the `Proxy-Authorization` header, but we still decided to strip the header by default in order to further protect users who aren't using the correct approach. ## Affected usages We believe the number of usages affected by this advisory is low. It requires all of the following to be true to be exploited: * Setting the `Proxy-Authorization` header without using urllib3's built-in proxy support. * Not disabling HTTP redirects. * Either not using an HTTPS origin server or for the proxy or target origin to redirect to a malicious origin. ## Remediation * Using the `Proxy-Authorization` header with urllib3's `ProxyManager`. * Disabling HTTP redirects using `redirects=False` when sending requests. * Not using the `Proxy-Authorization` header.

urllib3 before version 1.23 does not remove the Authorization HTTP header when following a cross-origin redirect (i.e., a redirect that differs in host, port, or scheme). This can allow for credentials in the Authorization header to be exposed to unintended hosts or transmitted in cleartext.