Red Hat® Enterprise Linux® AI is a foundation model platform to seamlessly develop, test, and run Granite family large language models (LLMs) for enterprise applications.
This update provides the latest Red Hat Enterprise Linux AI 3.4.1 container images.
For a full list of changes in this release, see the Red Hat Enterprise Linux AI Release Notes linked in the References section.
Net::BitTorrent versions through 2.1.0 for Perl allow remote memory exhaustion via deeply nested bencoded input.
bdecode recurses once per nested list or dictionary level with no depth cap, and each recursive call receives the remaining buffer by value while the list and dictionary branches capture the whole remainder, so every live recursion frame keeps its own copy of the shrinking buffer (O(N^2) bytes for an N-deep input). The decoder runs on every untrusted bencode source: .torrent files, BEP09 metadata fetched from peers, DHT messages, and tracker responses.
A bencoded input of roughly 150,000 nested lists (about 150 KB on the wire) drives multi-gigabyte peak memory, so one short message from any peer, or one crafted .torrent file or magnet link, terminates the client.
Net::BitTorrent versions through 2.1.0 for Perl allow remote memory exhaustion via an uncapped peer-wire message-length prefix.
The peer-wire framing in _process_messages trusts the 4-byte length prefix sent by a connected peer with no upper bound, while receive_data appends every inbound byte to the input buffer. A peer announces a length prefix of up to about 4 GiB and then streams bytes; the decoder waits until the buffer holds the full message before processing it, so the buffer grows without limit.
Peer connections are unauthenticated, so any peer in the swarm exhausts the downloading process's memory. The largest legitimate message is a 16 KiB piece block, so any announced length far above that is anomalous.
brace-expansion through 5.0.6 is vulnerable to denial of service. The expand() function exhibits exponential-time complexity in the number of consecutive non-expanding '{}' brace groups. An attacker who passes a crafted string to expand(), directly or transitively, can cause significant CPU consumption and event-loop blocking. The max option does not mitigate this, as it bounds the output size rather than the recursion work.
decode-uri-component through 0.4.1 is vulnerable to denial of service. The decode() function splits input on '%' producing N tokens and calls decodeComponents(), exhibiting super-linear parsing time: 200 '%ab' tokens takes approximately 0.7s, 700 tokens approximately 6s, and 1400 tokens approximately 33s. An attacker can cause significant CPU consumption and event-loop blocking via crafted input.
JavaScript::Minifier::XS versions before 0.16 for Perl leak memory on every call to minify(), allowing unbounded memory growth.
In JsMinify (XS.xs) the cleanup frees only the NodeSet structures and never the per-token contents buffers allocated in JsSetNodeContents; JsDiscardNode unlinks nodes without freeing their contents. Each token's contents buffer is therefore leaked on every call, and the two early returns taken when the node list is empty leak the whole NodeSet.
A long-lived process that minifies repeatedly, such as an asset pipeline or a server-side minifier endpoint, grows in memory without bound until it exhausts available memory and is killed, causing denial of service.
### Impact
The HTML backend did not perform sufficient validation during resource handling:
- Accepted `file://` URIs enabling local file system access when `enable_local_fetch=True`
- Path resolution allowed traversal outside intended directories via `../` sequences and absolute paths
- Did not block internal network resources under `enable_remote_fetch=True`
- HTTP redirects were not validated, potentially redirecting to unintended schemes
- No resource limits for remote image downloads and `data:` URIs
### Patches
Fixed in versions 2.91.0 (initial fixes) and 2.94.0 (additional improvements). The fixes implement:
- Updated local path treatment: absolute files always blocked, relative paths require `enable_local_fetch=True` (default: False) and containment within configured `base_path` for path traversal protection
- `file://` scheme stripped & treated as local path (above)
- IP address validation to prevent SSRF
- HTTP redirect validation, connection and read timeouts
- Size limit for both remote images (with streaming download) and base64-decoded data URIs
### Workarounds
Keep both `enable_local_fetch=False` and `enable_remote_fetch=False` (defaults) when processing untrusted HTML documents.
### References
- Initial fixes: [v2.91.0](https://github.com/docling-project/docling/releases/tag/v2.91.0)
- Additional improvements: [v2.94.0](https://github.com/docling-project/docling/releases/tag/v2.94.0)
By sending a deeply nested ASN1 structure to a Apache Kerby client or service, it's possible to trigger a StackOverFlow Exception which can lead to denial of service issues. Users are recommended to upgrade to version 2.1.2, which fixes this issue.
A flaw in Node.js HTTP/2 client allows a server to send an unlimited number of ORIGIN frames, which could lead to an Out of Memory error on the client.
This vulnerability affects all supported release lines: **Node.js 22**, **Node.js 24**, and **Node.js 26**.
AutoGPT is a workflow automation platform for creating, deploying, and managing continuous artificial intelligence agents. In versions prior to 0.6.52, the Fill Text Template block is vulnerable to a Denial of Service (DoS) attack. While the backend implements a SandboxedEnvironment to prevent unauthorized attribute access (e.g., blocking __class__), it fails to limit the computational complexity or execution time of the expressions. An attacker can input computationally expensive Python/Jinja2 expressions that consume the server's CPU and memory, leading to a complete system hang or crash. In multi-tenant or self-hosted environments, this results in a complete service outage and "noisy neighbor" effects that require manual administrative intervention to recover. This issue has been fixed in version 0.6.52.
Mistune is a Python Markdown parser with renderers and plugins. Prior to 3.3.0, Mistune is vulnerable to a CPU exhaustion DoS due to superlinear (approximately O(n²)) behavior in parse_link_text. When parsing Markdown containing many consecutive [ characters, parse_link_text repeatedly scans the input using a regex search inside a loop. Each iteration re-scans a large portion of the remaining string, resulting in quadratic-time behavior. An attacker-controlled Markdown input can therefore trigger excessive CPU usage with a very small payload. This vulnerability is fixed in 3.3.0.
jackson-databind contains the general-purpose data-binding functionality and tree-model for Jackson Data Processor. From 2.13.0 until 2.14.0, a potential Denial-of-Service exists when attacker sends deeply nested JSON if (and only if) the service reads deeply nested (1000s of levels) JSON as JsonNode (ObjectMapper.readTree()) and writes out same (or modifided) node using JsonNode.toString(). This can consume significant amount of resources with concurrent relatively small requests (1000 nested arrays is 2kB). This vulnerability is fixed in 2.14.0.
Langflow is a tool for building and deploying AI-powered agents and workflows. Prior to 1.0.19, an attacker can send a /api/v1/files/upload/ request without any authentication token/cookies and abuse a very long multipart form boundary to make the langflow app unusable for all users for an indefinite amount of time. This vulnerability is fixed in 1.0.19.
pypdf is a free and open-source pure-python PDF library. Prior to 6.12.2, an attacker who uses this vulnerability can craft a PDF which leads to large memory usage. This requires extracting the text of a page which contains a form XObject with self-references. This vulnerability is fixed in 6.12.2.
### Summary
When parsing `application/x-www-form-urlencoded` bodies, `QuerystringParser` located the field separator with a two step lookup: it first scanned the entire remaining buffer for `&`, and only when no `&` existed anywhere ahead did it fall back to scanning for `;`. For a body that uses `;` as the separator and contains no `&`, every field iteration performed a full failed `&` scan over the entire remaining buffer before locating the nearby `;`. With N semicolon separated fields in a chunk of size B, this yields O(B^2) byte comparisons per chunk.
An attacker can submit a small crafted body of the form `a;a;a;...` and cause the parser to spend seconds of CPU per request. A handful of concurrent requests can exhaust worker processes.
### Details
In `python_multipart/multipart.py`, both the `FIELD_NAME` and `FIELD_DATA` states located the next separator like this:
```python
sep_pos = data.find(b"&", i)
if sep_pos == -1:
sep_pos = data.find(b";", i)
```
`data.find(b"&", i)` scans from `i` to the end of the buffer and returns `-1` only when there is no `&` anywhere in the remainder. For a `;` separated body with no `&`, this failed full buffer scan repeats once per field, making parsing quadratic in the body length.
For example, a 1 MiB url encoded body consisting of `a;` repeated ~500,000 times, submitted with `Content-Type: application/x-www-form-urlencoded`, causes the parser to perform on the order of 10^11 byte comparisons, consuming several seconds of CPU for a single request. Cost scales quadratically with chunk size.
The parser is reachable through the public `QuerystringParser` class and through the high level `FormParser`, `create_form_parser`, and `parse_form` APIs for url encoded bodies. It is also the parser Starlette and FastAPI use for `application/x-www-form-urlencoded` request bodies via `request.form()`.
### Impact
Uncontrolled CPU consumption (denial of service). Parsing is synchronous, so a single small crafted form body occupies the handling worker for seconds, blocking any other work on that worker until parsing finishes. Sustained concurrent requests keep workers continuously busy, degrading or denying service.
### Mitigation
Upgrade to `python-multipart` `0.0.30` or later, which treats only `&` as a field separator (per the [WHATWG URL standard](https://url.spec.whatwg.org/#urlencoded-parsing)) using a single bounded scan, making parsing linear in the body length.
The public dashboard query endpoint does not limit request body size before processing, allowing unauthenticated attackers to trigger excessive memory allocation by sending arbitrarily large JSON payloads. This can lead to denial of service through memory exhaustion. No valid dashboard access token or authentication is required to exploit this vulnerability.
IBM WebSphere Application Server 9.0, and 8.5 and IBM WebSphere Application Server - Liberty 17.0.0.3 through 26.0.0.6 are vulnerable to a denial of service, caused by sending a specially-crafted request. A remote attacker could exploit this vulnerability to cause the server to consume memory resources.
IBM WebSphere Application Server 9.0, and 8.5 and IBM WebSphere Application Server - Liberty 17.0.0.3 through 26.0.0.6 are vulnerable to a denial of service, caused by sending a specially-crafted request. A remote attacker could exploit this vulnerability to cause the server to consume memory resources.
A TraceQL query in Grafana Tempo with a large exemplars hint value can cause the Tempo instance to allocate an excessive amount of memory, resulting in an out-of-memory crash. This could allow an authenticated user to trigger a denial of service against the Tempo service.
js-toml is a TOML parser for JavaScript, fully compliant with the TOML 1.0.0 Spec. Versions up to and including 1.1.1 parse hexadecimal / octal / binary integer literals via a hand-written `parseBigInt` loop that multiplies a `BigInt` accumulator by the radix once per input digit. Each iteration performs a `BigInt * BigInt` operation on an accumulator that grows linearly with the number of digits already consumed, so the whole loop is O(n²) in the literal length. The lexer regex places no upper bound on the literal length, so a single TOML document containing one ~500 kB hex literal pins one CPU core for ~40 seconds on a modern laptop (Apple M-series, Node v22). Memory amplification is bounded but CPU amplification is severe and grows quadratically: doubling the literal length quadruples the work. A caller that invokes `load()` on attacker-controlled TOML (configuration upload endpoints, CI/CD systems ingesting third-party `*.toml`, IDE plugins, build tools) is exposed to a single-request CPU exhaustion DoS. Version 1.1.2 fixes the issue.