Security Vulnerabilities
vLLM before 0.29.0 contains a resource-limit bypass vulnerability in PyNvVideoCodec decoder allocation where sampler subclass shadowing allows independent counter increments. Unauthenticated attackers can select different sampler subclasses in video requests to exceed configured decoder limits and exhaust unaccounted GPU memory.
CVSS Score
6.3
EPSS Score
0.003
Published
2026-09-26
vLLM through 0.29.0 fetches and fully materializes remote or inline media before enforcing its documented media controls (the VLLM_MAX_AUDIO_CLIP_FILESIZE_MB compressed-audio size cap, default 25 MB, and the per-modality --limit-mm-per-prompt item limits). Across four ingress paths — the shared media-acquisition layer (HTTPConnection.get_bytes()/async_get_bytes()), the chat completions audio_url/base64 path, the batch speech runner, and the Rust frontend POST /tokenize route — the server reads the entire HTTP response body, base64-decodes the inline payload, or spawns one fetch/decode task per media part, and only then applies the limit (or, on some paths, never applies it). A remote attacker can therefore cause the API server or batch-runner process to allocate memory and consume outbound bandwidth proportional to an attacker-chosen body size or media item count before the request is rejected, resulting in pre-inference memory and bandwidth exhaustion (denial of service). The chat and batch surfaces require an API key when one is configured; the Rust frontend /tokenize route is unauthenticated by design. There is no code execution or data disclosure impact.
CVSS Score
7.1
EPSS Score
0.006
Published
2026-09-26
vLLM before 0.29.0 fails to enforce decoder prompt-length validation on the disaggregated serving endpoint /inference/v1/generate. When the request contains a 'features' (multimodal) payload, vllm/entrypoints/serve/disagg/serving.py builds a multimodal EngineInput directly from the caller-supplied token_ids, and GenerateRequest.token_ids (vllm/entrypoints/serve/disagg/protocol.py) is not checked against model_config.max_model_len. For multimodal processors that report skip_prompt_length_check=True (for example Nemotron Parse, Whisper, and FireRedLID), InputProcessor._validate_prompt_len() returns immediately for both encoder and decoder prompts, so an overlong prompt becomes an EngineCoreRequest and reaches the worker input-batch copy into a fixed max_model_len-wide NumPy row. A client able to reach the endpoint on an affected model configuration can therefore submit an overlong token_ids list to trigger a worker failure and denial of service. Fixed in 0.29.0.
CVSS Score
7.1
EPSS Score
0.003
Published
2026-09-26
vLLM versions 0.22.0 through 0.23.0 fail to validate stop_token_ids against vocabulary bounds in Rust HTTP and gRPC frontends, allowing out-of-vocabulary token IDs to reach MinTokensLogitsProcessor. Attackers can submit requests with min_tokens greater than zero and out-of-vocabulary stop_token_ids to trigger CUDA tensor indexing failures that leave EngineCore in a fatal state requiring service restart.
CVSS Score
8.2
EPSS Score
0.003
Published
2026-09-26
In the Linux kernel, the following vulnerability has been resolved:
cgroup: Avoid iteration of dying tasks with zero refcount
The commit 260fbcb92bbea ("cgroup: Move dying_tasks cleanup from
cgroup_task_release() to cgroup_task_free()") extended the lifetime of
tasks on the dying_tasks list.
The iterators have provision to go through dying_tasks because of
dying threadgroup leaders or explicit CSS_TASK_ITER_WITH_DEAD, however,
it was expected that such tasks can obtain a new reference (that is
possible before cgroup_task_release()/put_task_struct_rcu_user()).
The tasks after cgroup_task_release() and before cgroup_task_free()
are subject to race when they may or may not have ->usage count > 0.
The race window is between css_task_iter_next() invocations
when css_set_lock is released and we may arrive at a new ->task_pos.
The iterator should not attempt to resurrect tasks whose ->usage count
dropped to zero. (When that happens, __put_task_struct_rcu_cb() is
already imminent and the returned task_struct would could be used
after free.)
As for the fix, we cannot simply check the signal->live count of a task
on the dying list because that won't distinguish regular zombies waiting
to be reaped from RCU remnant tasks that are going to be free'd.
Therefore add an extra check to rule out ->usage==0 tasks from any
iteration.
The repeat: loop in css_task_iter_advance() doesn't consider ->usage
count, so add a new loop to css_task_iter_next() to skip de-used tasks
on the dying_list.
Rough illustration of the possible race
R (reader of cgroup.procs) T (thread) L (group leader)
--------------------------------- -------------------------------- --------------------------------
L exits, signal->live > 0
cgroup_task_dead(L)
css_set_skip_task_iters() // skips only cset->tasks
list_add_tail(&L->cg_list, &cset->dying_tasks)
css_task_iter_next()
take css_set_lock
css_task_iter_advance()
leader && signal->live != 0
=> it->task_pos = &L->cg_list
release css_set_lock
T exits
--signal->live == 0
cgroup_task_dead(T) // css_set_lock
release_task(T)
cgroup_task_release(T)
release_task(L) // zap_leader
cgroup_task_release(L)
put_task_struct_rcu_user(L)
...RCU...
put_task_struct(L)
L->usage = 0
/* L still on dying_tasks */
...RCU...
__put_task_struct(L)
css_task_iter_next() // another iteration
take css_set_lock
it->task_pos = &L->cg_list
get_task_struct(L)
=> addition on 0
drop css_set_lock
cgroup_task_free(L)
css_set_skip_task_iters() // dying skip comes too late
free_task(L)
cgroup_procs_show()
task_pid_vnr(L)
CVSS Score
7.0
EPSS Score
0.001
Published
2026-09-26
RabbitMQ is a messaging and streaming broker. From 3.13.0 until 3.13.19, 4.0.24, 4.1.15, 4.2.10, and 4.3.5, RabbitMQ Management rendered an AMQP authorization-error reason containing an attacker-controlled queue name as HTML when the OAuth management UI was enabled. Exploitation requires an attacker with queue configure permission, a management administrator who can see but cannot read that queue, and the administrator clicking Get Message(s). A queue name containing a base element can then retarget the automatic relative refresh because the Content Security Policy omits base-uri and connect-src, and an attacker endpoint that permits the management origin through CORS can receive the victim's Authorization header. This issue is fixed in versions 3.13.19, 4.0.24, 4.1.15, 4.2.10, and 4.3.5.
CVSS Score
4.5
EPSS Score
0.003
Published
2026-09-25
IBM Server Firmware FW1120.00 through FW1120.01, FW1110.00 through FW1110.31, FW1060.00 through FW1060.81, and FW950.00 through FW950.H3 is affected by a vulnerability in the ASMI web interface. An unauthenticated attacker on the management network can send a malformed HTTPS request to ASMI, causing the web server to crash with possible memory corruption and generate an error log. The ASMI web interface will restart automatically; however, repeated exploitation could result in a sustained loss of access to the ASMI management interface, resulting in an integrity and availability impact.
CVSS Score
7.1
EPSS Score
0.002
Published
2026-09-25
In the Linux kernel, the following vulnerability has been resolved:
staging: rtl8723bs: fix mismatched free of HalData in rtw_sdio_if1_init()
padapter->HalData is allocated via vzalloc(), but incorrectly freed
using kfree() in the rtw_sdio_if1_init() error path. Using kfree() to
release this vmalloc-backed buffer can lead to memory corruption.
Use rtw_hal_data_deinit() to pair the free correctly and free
HalData with vfree().
The bug was first flagged by an experimental static analysis tool we
are developing for kernel memory-management bugs. Manual inspection
confirms that the issue is still present in current mainline.
An x86_64 allyesconfig build showed no new warnings. As we do not have
suitable RTL8723BS SDIO hardware to test with, no runtime testing was
able to be performed.
CVSS Score
5.5
EPSS Score
0.001
Published
2026-09-25
In the Linux kernel, the following vulnerability has been resolved:
smb/server: fix tree connection leak in smb2_tree_connect()
See the procedure below:
smb2_tree_connect
ksmbd_tree_conn_connect
xa_store(&sess->tree_conns, tree_conn->id, tree_conn)
ksmbd_counter_inc(KSMBD_COUNTER_TREE_CONNS)
ksmbd_share_tree_conn_inc(sc)
ksmbd_iov_pin_rsp // fail
status.ret = KSMBD_TREE_CONN_STATUS_NOMEM
// do not disconnect tree_conn
Disconnect the new tree connection if ksmbd_iov_pin_rsp() fails.
CVSS Score
5.5
EPSS Score
0.001
Published
2026-09-25
In the Linux kernel, the following vulnerability has been resolved:
nvdimm: pmem: keep PREFLUSH before data writes
pmem_submit_bio() records a REQ_PREFLUSH error, but continues to copy the
bio data and can later overwrite the error with a successful REQ_FUA flush.
That lets data writes run after a failed preflush and can complete the bio
successfully despite the failed ordering barrier.
Run the REQ_PREFLUSH flush synchronously before touching the bio data and
complete the bio with the flush error if it fails. Keep asynchronous flush
chaining for REQ_FUA. At that point, data copy has completed and the parent
bio can wait for the chained flush bio.
CVSS Score
5.5
EPSS Score
0.001
Published
2026-09-25