Jiadai Wang

A denial of service issue exists where API clients can abort in-flight generation requests belonging to other users. This occurs when an attacker submits streaming requests to the '/generate' endpoint using a short, client-controlled `rid` (request identifier). By using single hex character `rid` values and then disconnecting, the attacker triggers prefix-matching abort logic that terminates all other active client requests.

In prefill/decode disaggregation mode, a server-side request forgery issue exists that allows unauthenticated attackers to trigger outbound requests. By sending crafted requests to the '/generate' endpoint of the Decode worker, attackers can use the `bootstrap host` and `bootstrap port` fields to probe internal hosts and ports. This action can block the shared scheduler loop, resulting in delays for other inference requests.

An authorization bypass exists where holders of the data-plane API key can access mesh High Availability (HA) management routes. This occurs because the `/ha/*` endpoints skip control-plane Admin role checks. An attacker can read cluster topology, write gossiped configuration, modify global rate limits, and use the `POST /ha/shutdown` endpoint to silently dissolve HA coordination.

SGLang with the Mooncake disaggregation backend allows denial of service when it accepts unauthenticated decode registration messages on the Prefill rank ZMQ PULL socket. Network-adjacent attackers can exhaust Prefill memory, leading to the scheduler being OOM-killed (Out-Of-Memory killed, a process where the operating system terminates a process to free up memory), by sending numerous registrations containing large nested-buffer frames and unique `mooncake session id` values.

An uncontrolled memory allocation issue exists in the Mooncake disaggregation backend within the `handle staging rsp()` function. Unauthenticated network-adjacent attackers can exhaust Prefill memory by obtaining the rank address from the '/route' endpoint and sending STAGING RSP messages containing excessively large `chunk idx` values under new room IDs. This action can lead to the scheduler being OOM-killed, which refers to an Out-Of-Memory state where the operating system terminates the process to reclaim memory.

A code injection issue exists in the ` evaluate filter()` function. Remote attackers can execute Python code by escaping the `eval()` sandbox using `re.search globals `. This is possible by sending a crafted filter expression to the unauthenticated '/dumper/configure' endpoint when the `DUMPER SERVER PORT` variable is set, which results in code execution within the scheduler process during the subsequent forward pass.

The software trusts caller-supplied `mm hashes` values as the multimodal prefix-cache key. This allows remote attackers to poison cached KV (Key-Value) entries across requests. By sending a different image under a known or predicted hash via the '/generate' endpoint, an attacker can disclose the image content of another request or corrupt the responses received by victims.

An unbounded memory allocation issue exists that allows unauthenticated remote attackers to cause a denial of service. By uploading oversized files to the '/v1/audio/transcriptions' endpoint, attackers can send concurrent large multipart uploads that are read entirely into memory before validation, resulting in a crash of the HTTP/tokenizer worker.

A decompression amplification issue exists in the `load audio()` function. Remote attackers can cause a denial of service by submitting highly compressed audio files, such as small FLAC-encoded silence files, which are decoded without duration or sample limits. This process expands the files into massive PCM (Pulse Code Modulation) buffers, exhausting worker memory and CPU resources. The attack can be executed via the `audio url` or `audio data` inputs.

An unbounded memory allocation issue exists in the `preprocess video` function within `qwen vl.py`. The system decodes video frames at their full source resolution before applying `max pixels` limits. Remote attackers can exploit this by submitting high-resolution videos to the '/v1/chat/completions' endpoint on Qwen-VL deployments. Concurrent requests of this nature can exhaust worker memory, leading to a denial of service.

A resource budget bypass allows remote attackers to override GLM-V video processing limits. By providing specific values in the `video url` object, attackers can manipulate the `fps`, `max frames`, `max tokens per frame`, and `max image tokens` variables. This causes the `glm4v.py` function to decode every frame of high-resolution video at full size, leading to worker memory exhaustion and a denial of service.

A denial of service issue exists where the `NanoNemotronVLImageProcessor` disables Pillow decompression bomb protection process-wide by setting `Image.MAX IMAGE PIXELS` to `None`. This allows remote attackers to submit small, highly compressed images with massive dimensions to the generic image loader, leading to worker memory exhaustion during the decoding process. A decompression bomb is a malicious archive or image file designed to crash a system by expanding to an enormous size when decompressed.

The iApp message dispatcher contains a reachable assertion. The dispatcher uses the `assert()` function to validate incoming E2AP messages against a 9-entry whitelist. A remote unauthenticated attacker can send a decodable E2AP PDU with a message type not present in the whitelist, triggering a SIGABRT (a signal sent to a process to force it to abort) and crashing the iApp process on port 36422. Because iApp and the near-RT RIC share a single process, this action terminates the entire RIC service, resulting in the disconnection of all xApps and E2 Nodes.

The iApp process on port 36421 crashes when it receives a duplicate E2 SETUP REQUEST from the same or a spoofed E2 Node. This occurs because the iApp registry uses an `assert()` function to enforce node ID uniqueness instead of implementing a controlled rejection. A remote unauthenticated attacker can trigger a SIGABRT (a signal sent to a process to force it to abort) by sending two E2 SETUP REQUESTs with identical E2 node configurations.

A validation mismatch between layers allows a remote unauthenticated attacker to crash the iApp process on port 36422 via SIGABRT. The issue occurs when the iApp receives an 'E42 RIC SUBSCRIPTION REQUEST' containing an empty `ricEventTriggerDefinition` field; while the E42 layer decoder accepts the empty field as valid, the E2AP encoder enforces a non-empty constraint when forwarding the request, leading to the crash.

The near-RT RIC contains reachable `assert(0)` calls within stub message handlers for E2AP message types that are whitelisted but not yet implemented. A remote unauthenticated attacker can cause the near-RT RIC process on port 36421 to crash via SIGABRT by sending a decodable E2AP PDU of an unimplemented type, such as `E2nodeConfigurationUpdate`. The message successfully passes whitelist validation but triggers an unconditional assertion in the handler.

A reachable assertion in the `e2ap create pdu()` function is triggered when ASN.1 PER (Packed Encoding Rules, a binary encoding format for ASN.1 data) decoding fails. A remote unauthenticated attacker can send a non-PER byte sequence over SCTP to the near-RT RIC on port 36421 or the iApp on port 36422, causing the process to crash via SIGABRT. This occurs before any protocol-level validation takes place and affects E2AP protocol versions v1.01, v2.03, and v3.01.

An issue exists where a `uint16 t` counter is used for `xapp id` assignment but is stored in `uint32 t` message fields. After more than 65,530 E42 SETUP REQUESTs, the 16-bit counter wraps around, resulting in duplicate `xapp ids`. This causes the iApp (port 36422) to crash when it attempts to register a duplicate ID within its internal data structure. A remote attacker can trigger this crash by repeatedly connecting and requesting new xApp registrations.

An issue exists where the software trusts the `xapp id` field from E42 message payloads without binding it to the sender's SCTP association. The validation function `valid xapp id()` only verifies that the value is within the assigned range. A remote unauthenticated attacker can impersonate any xApp by specifying a target `xapp id` in requests sent to the iApp endpoint on port 36422. This leads to responses being misrouted to the victim xApp, which can cause the victim xApp, the RIC, or the iApp to crash due to state inconsistencies in the red-black tree data structure, a self-balancing binary search tree used for efficient data storage and retrieval.