| CVE |
Vendors |
Products |
Updated |
CVSS v3.1 |
| vLLM through 0.29.0 fails to validate the tp_size parameter in kv_transfer_params on OpenAI-compatible completion endpoints, allowing attackers to allocate unbounded memory. Attackers can supply arbitrary tp_size values in prefill/decode disaggregated deployments to exhaust memory and trigger kernel OOM-kill of the decode worker process. |
| The CompressionFilter class uses ZLib to deflate and inflate data sent and received. When we inflate incoming data, the filter does not control the resulting size, and create a buffer no matter what.
Some compressed data may have a compression ration greater than 1 thousand, leading to an exhaustion of the application memory, as we don't control the deflated size.
The fix adds such a control by allowing the application developer to provide a fixed size limit, which when reached throws an exception. It also allows the user to provide a compression ratio that should not be exceeded, protected the application from small inflated files that inflate in gigantic files, but with a grace limit for the resulting size (1Mb) to avoid false positive (like a very small file inflating with a high ratio, but resulting with a acceptable size, like a few thousands bytes)
For application using this feature, it is highly recommended to create the CompressionFilter and to pass the maximum limit as a forth constructor parameter, maxDecompressedSize:
public CompressionFilter(final boolean compressInbound, final boolean compressOutbound, final int compressionLevel, final int maxDecompressedSize)Optionally one can also provide a maxDecompressRatio fifth parameter, and a decompressRatioMinSize sixth parameter to allow small inflated files with a high compression ratio to still be accepted.
Here are the additional constructor:
public CompressionFilter(final boolean compressInbound, final boolean compressOutbound,
final int compressionLevel, final int maxDecompressedSize,
final long maxDecompressRatio, final long decompressRatioMinSize)
Also note that a fluent API has been added to spare the users the pain to call a constructor with that many parameters:
CompressionFilter compressionFilter = new CompressionFilter()
.setCompressionLevel(Zlib.COMPRESSION_MAX)
.setMaxDecompressedSize(1_000_000)
.setMaxDecompressRatio(100).
.setDecompressRatioMinSize(100_000);
Applications using Apache MINA are advised to upgrade and configure their CompressionFilter instance. |
| KubeEdge is an open source system for extending native containerized application orchestration capabilities to hosts at Edge. From 1.0.0 until 1.21.2, 1.22.2, and 1.23.1, Reader.Read in pkg/viaduct/pkg/packer trusts the 32-bit PackageHeader.PayloadLen received through the CloudHub viaduct message-processing path and allocates that amount of memory before validating an upper bound. An authenticated malicious or compromised edge peer can repeatedly send crafted headers with excessive declared lengths, causing memory exhaustion, CloudHub process termination or restart loops, and temporary disruption of cloud-edge communication. This issue does not provide unauthenticated access or direct code execution. This issue is fixed in versions 1.21.2, 1.22.2, and 1.23.1. |
| adm-zip is a JavaScript library for creating and extracting ZIP archives in Node.js. Prior to 0.6.1, getData() in zipEntry.js trusts an entry's central-directory uncompressed size and allocates output memory before validating that value against the actual compressed data and decompression result. A small crafted ZIP can declare a multi-gigabyte uncompressed size, causing Buffer.alloc and decompression handling to commit excessive resident memory before CRC validation reports an error. Applications that read entries from untrusted archives can therefore be terminated by the operating system or suffer service-wide memory exhaustion. This issue is fixed in version 0.6.1. |
| In the Linux kernel, the following vulnerability has been resolved:
blk-cgroup: fix race between policy activation and blkg destruction
When switching an IO scheduler on a block device, blkcg_activate_policy()
allocates blkg_policy_data (pd) for all blkgs attached to the queue.
However, blkcg_activate_policy() may race with concurrent blkcg deletion,
leading to use-after-free and memory leak issues.
The use-after-free occurs in the following race:
T1 (blkcg_activate_policy):
- Successfully allocates pd for blkg1 (loop0->queue, blkcgA)
- Fails to allocate pd for blkg2 (loop0->queue, blkcgB)
- Enters the enomem rollback path to release blkg1 resources
T2 (blkcg deletion):
- blkcgA is deleted concurrently
- blkg1 is freed via blkg_free_workfn()
- blkg1->pd is freed
T1 (continued):
- Rollback path accesses blkg1->pd->online after pd is freed
- Triggers use-after-free
In addition, blkg_free_workfn() frees pd before removing the blkg from
q->blkg_list. This allows blkcg_activate_policy() to allocate a new pd
for a blkg that is being destroyed, leaving the newly allocated pd
unreachable when the blkg is finally freed.
Fix these races by extending blkcg_mutex coverage to serialize
blkcg_activate_policy() rollback and blkg destruction, ensuring pd
lifecycle is synchronized with blkg list visibility. |
| A vulnerability in the sftunnel inter-device communication protocol of Cisco Secure FMC Software and Cisco Secure FTD Software could allow an unauthenticated, remote attacker to exhaust the available memory of an affected device.
This vulnerability is due to improper management of memory resources during sftunnel TLS connection setup. An attacker could exploit this vulnerability by sending crafted sftunnel TLS frames to an affected device during the connection setup. A successful exploit could allow the attacker to exhaust the available memory on the affected device, which could result in a DoS condition. |
| In the Linux kernel, the following vulnerability has been resolved:
nvme-fc: unmap cmd_iu DMA on rsp_iu mapping failure in init_request
__nvme_fc_init_request() maps cmd_iu and then rsp_iu for DMA. If the
rsp_iu mapping fails, the original code only recorded the error and fell
through: it left the already-mapped cmd_iu unmapped and still marked the
op as FCPOP_STATE_IDLE before returning. Since blk-mq does not call
.exit_request() when .init_request() fails, the cmd_iu mapping is leaked
for every op whose rsp_iu mapping fails.
Jump to an error path on rsp_iu mapping failure that unmaps cmd_iu and
returns the error without marking the op idle, so it stays in the
FCPOP_STATE_UNINIT state set by the initial memset(). |
| In the Linux kernel, the following vulnerability has been resolved:
media: rtl2832_sdr: use vb2_video_unregister_device() on remove to fix DMA leak
rtl2832_sdr_remove() runs on USB disconnect and clears dev->udev to
NULL before any pending streaming teardown has run. When user space
later closes its file descriptor, vb2 calls rtl2832_sdr_stop_streaming()
which in turn calls rtl2832_sdr_free_stream_bufs(). That helper releases
each coherent buffer with:
usb_free_coherent(dev->udev, dev->buf_size,
dev->buf_list[dev->buf_num],
dev->dma_addr[dev->buf_num]);
usb_free_coherent() returns immediately when its dev argument is NULL,
so every DMA stream buffer that was live at disconnect is silently
leaked. The URBs allocated in rtl2832_sdr_alloc_urbs() outlive the
device for the same reason.
The rtl2832_sdr driver uses vb2_fop_release() in its file_operations,
so replace video_unregister_device(&dev->vdev) with
vb2_video_unregister_device(&dev->vdev) and move it before clearing
dev->udev. vb2_video_unregister_device() releases the vb2 queue, which
synchronously runs rtl2832_sdr_stop_streaming() if streaming is active,
so URBs and coherent DMA stream buffers are freed while dev->udev is
still valid.
vb2_video_unregister_device() locks vdev->queue->lock (vb_queue_lock)
internally, and stop_streaming() locks v4l2_lock, so the previous outer
mutex_lock(&dev->vb_queue_lock) / mutex_lock(&dev->v4l2_lock) pair
around the unregister sequence would self-deadlock and has been removed.
A short v4l2_lock critical section around dev->udev = NULL remains so
any ioctl path that still holds the file descriptor sees coherent state.
Issue identified by automated review of the INV-003 series at
https://sashiko.dev/ |
| Imager versions before 1.036 for Perl exit the process reading a TGA with a colour map length of 32768 or more in tga_palette_read.
The reader unpacks the two-byte colour map length into a signed short, so a length of 32768 or more becomes negative. tga_palette_read() casts that value to size_t and asks mymalloc() for a size near SIZE_MAX. The allocation fails and Imager's allocator calls exit(3).
Reading an attacker-supplied file through Imager->read() triggers an uncatchable exit. |
| A flaw was found in Wildfly. A remote unauthenticated attacker can trigger OutOfMemoryError as CSIv2Util's GSS token decoder reads an attacker-controlled length field without bounds checking and attempts to allocate a byte array of that size. |
| A vulnerability has been found in O-RAN-SC SMO OAM 2025-06-10. Affected is an unknown function of the component VES Collector. Such manipulation of the argument additionalFields.padding leads to uncontrolled memory allocation. The attack can be launched remotely. The exploit has been disclosed to the public and may be used. The project was informed of the problem early through a bug report but has not responded yet. |
| GNU libextractor before 1.15 contains a stack-based buffer overflow vulnerability in the process_star_office function that sizes a variable-length stack array from attacker-controlled OLE2 stream data. Attackers can craft malicious StarOffice documents that allocate up to 4 MB on the stack, causing stack overflow and crashing any application extracting metadata from the document. |
| Vouch Proxy is an SSO and OAuth/OIDC login solution for Nginx using the auth_request module. Prior to 0.48.0, Cookie in pkg/cookie/cookie.go parses the total part count from an attacker-controlled multipart cookie name and passes the value to make([]string, numParts) without checking that the value is positive or reasonably bounded. Requests to /validate and /_external-auth-:id reach JWTCacheHandler in pkg/jwtmanager/jwtcache.go, FindJWT in pkg/jwtmanager/jwtmanager.go, and the vulnerable cookie reassembly before JWT validation, so no account or valid session is required. A cookie name such as VouchCookie_1of10000000000 causes an attempted slice allocation of roughly 160 GB and a fatal Go runtime out-of-memory condition, allowing one request to crash the authentication proxy and repeated requests to sustain unavailability. This vulnerability is fixed in 0.48.0. |
| ExifReader is a JavaScript Exif information parser. Prior to 4.41.1, ExifReader parses attacker-controlled HEIC or AVIF ISO-BMFF files in getItems() within src/image-header-iso-bmff-iloc.js and trusts iloc itemCount and extentCount values while allocating an extent object for every nested-loop iteration. When offsetSize, lengthSize, baseOffsetSize, and indexSize are zero, the extent fields consume no input bytes and the buffer offset does not advance, but the parser can still allocate up to itemCount multiplied by extentCount objects without an allocation budget. A small malicious iloc box can therefore cause hundreds of megabytes of heap growth or exhaust system memory, terminating a Node.js process and denying service to web, desktop, or mobile applications that parse untrusted images. The zero field widths are valid ISO-BMFF values indicating absent fields, so the vulnerable parser must bound work rather than relying on offset advancement. The issue is fixed in version 4.41.1. |
| Update for September 16, 2026: The original 1.0 version of this advisory was specific to the Cisco Adaptive Security Virtual Appliance (ASAv) and Cisco Secure Firewall Threat Defense Virtual (FTDv) models. However, it was later found that this vulnerability affects all Cisco Secure Firewall Adaptive Security Appliance (ASA) Software and Cisco Secure Firewall Threat Defense (FTD) Software platforms.
A vulnerability in the VPN and management web servers of the Cisco Secure Firewall ASA Software and Cisco Secure FTD Software platforms could allow an unauthenticated, remote attacker to cause an affected device to run out of system memory or buffer blocks, which in turn could cause SSL VPN connection processing to slow down and eventually cease altogether.
This vulnerability is due to a lack of proper memory management for new incoming SSL/TLS connections. An attacker could exploit this vulnerability by sending a large number of new incoming SSL/TLS connections to the targeted device. A successful exploit could allow the attacker to deplete system memory or buffers, resulting in a denial of service (DoS) condition. The memory or buffers could be reclaimed slowly if the attack traffic is stopped, but a manual reload may be required to restore operations quickly. |
| Envoy Gateway is an open source project for managing Envoy Proxy as a standalone or Kubernetes-based application gateway. Prior to 1.7.4 and 1.8.1, getFileFromGZ in internal/wasm/httpfetcher.go calls io.ReadAll on a gzip.Reader without limiting decompressed output when a tenant-controlled EnvoyExtensionPolicy.spec.wasm[].code.http.url points to a reachable compressed Wasm payload. The 256 MiB compressed-input cap does not constrain the expanded size, no operator Wasm URL allowlist exists, and the optional sha256 check occurs only after decompression, so a comparatively small gzip stream can force a multi-gigabyte allocation in the shared controller. The resulting out-of-memory termination restarts the controller, re-reconciles the persistent custom resource, and can create a persistent cross-tenant control-plane outage. This issue is fixed in versions 1.7.4 and 1.8.1. |
| RabbitMQ amqp091-go is a Go AMQP 0.9.1 client. Prior to 1.13.0, Channel.recvContent in channel.go preallocates the message body slice with the uint64 ch.header.Size value supplied by an AMQP content header without capping the allocation to the negotiated Connection.Config.FrameSize value. A malicious or compromised broker can send an extreme declared body size and cause the Go runtime to attempt a correspondingly large allocation before body data is received. The allocation can exhaust memory and terminate the client process. This issue is fixed in version 1.13.0. |
| OOM Denial of Service via Unbounded Map Pre-Sizing in Apache OpenNLP SymSpellModelSerializer
Versions Affected:
- 3.0.0-M4
- 3.0.0-M5
(The opennlp-spellcheck extension was introduced in 3.0.0-M4. Releases 1.x and 2.x do not contain the affected code.)
Description:
The SymSpellModelSerializer.create() method reads two 32-bit signed integer count fields (unigramCount and bigramCount) from a binary SymSpell model stream and passes each value directly to LinkedHashMap.newLinkedHashMap() after validating only that it is non-negative. No upper bound is applied, so the count is fully attacker-controlled when the model file originates from an untrusted source.
A crafted .bin model file in which either count field is set to Integer.MAX_VALUE (or any value large enough to exhaust the available heap) causes the map to be pre-sized to a capacity of 2^30 entries. The oversized backing array is allocated on the first put() into that map, requesting 4–8 GB depending on whether compressed oops are in effect, and the load fails with an OutOfMemoryError. Because the count fields sit immediately after a fixed-size header (magic, format version, three UTF strings, the configuration fields, and the edit-distance identifier) the attacker pays no meaningful size cost to weaponize a payload: a file of well under 100 bytes plus a single real entry is sufficient to crash a JVM that loads it.
Any code path that deserializes a SymSpell model is affected, including SymSpellModels.deserialize(InputStream), SymSpellModels.fromBytes(byte[]), classpath model loading via SymSpellModelResolver.resolveByLanguage(String), the CorrectTextTool command-line tool, and model-archive loading through the registered ArtifactSerializer. The opennlp-spellcheck extension ships in the official OpenNLP binary distribution.
The practical impact is denial of service against processes that load SymSpell model files from untrusted or semi-trusted origins.
Mitigation:
- 3.x users should upgrade to 3.0.0-M6.
Note: The fix applies an upper bound to both count fields, checked before the map is pre-sized; counts that are negative or exceed the bound cause an IOException to be thrown and the read to fail fast with no large allocation. The bound is the existing AbstractModelReader.MAX_ENTRIES limit introduced earlie, which the current change promotes to public visibility so that serializers implementing their own binary format can share it. The default bound is 10,000,000, which is well above the entry counts of legitimate SymSpell dictionaries but far below any value that would threaten heap exhaustion. Deployments that legitimately need to load larger dictionaries can raise the limit at JVM startup by setting the OPENNLP_MAX_ENTRIES system property to the desired positive integer (e.g. -DOPENNLP_MAX_ENTRIES=50000000); invalid or non-positive values fall back to the default. Note that this property is shared with the model-reader limit and raising it relaxes both.
Users who cannot upgrade immediately should treat all SymSpell .bin model files as untrusted input unless their provenance is verified, and should avoid loading models supplied by end users or fetched from third-party repositories without integrity checks. |
| Envoy Gateway is an open source project for managing Envoy Proxy as a standalone or Kubernetes-based application gateway. Prior to 1.7.4 and 1.8.1, internal/wasm/imagefetcher.go follows tenant-controlled EnvoyExtensionPolicy spec.wasm[].code.image.url values to Docker or OCI Wasm layers, and extractWasmPluginBinary uses the untrusted tar-header h.Size value to allocate memory before validating the entry name or declared size. A small PAX or GNU tar header can therefore claim a multi-terabyte entry even though the surrounding LimitReader restricts only the bytes read from the stream, and no registry allowlist prevents a permitted tenant from selecting an attacker-controlled registry that the controller can reach. The allocation is attempted for every tar entry and can cause an unrecoverable Go runtime out-of-memory failure; because the custom resource persists, reconciliation repeatedly crash-loops the shared controller and causes a single-request, non-volumetric, cluster-wide control-plane denial of service. This issue is fixed in versions 1.7.4 and 1.8.1. |
| Description
The worker's Netty message decoder is installed ahead of the SASL authentication handlers in the pipeline
and acts on frames before any authentication has taken place. It allocated buffers sized from a
length field carried in the frame, so a single frame from an unauthenticated peer able to reach a worker
slot port could drive a large allocation.
`storm.messaging.netty.authentication` defaults to false, and the decoder runs before the handler that
enforces it in any case, so no credentials are required. The attacker needs only TCP reachability to a
worker port.
The effect of a single frame at the default 768 MB worker heap has not been measured to distinguish
sustained worker loss from transient garbage-collection pressure. The severity assigned to this advisory
reflects the more conservative reading; consumers who require a precise figure should test against their own
worker heap configuration.
Mitigation
Upgrade to 3.1.0, where frames are decoded only after the handshake completes.
Users who cannot upgrade immediately should ensure that worker slot ports are reachable only from within the
cluster, as the security model already recommends, and should enable
`storm.messaging.netty.authentication` where the deployment permits it.
Credit
The ASF -- found using Claude agents to study the security of open-source projects, validated and reported by Apache Storm. |
