Category Archives: AI and ML

AI and ML, Potential Risk of CVE

Regarding CVE-2024-0099 and CVE-2024-0084: Is this a renewed focus on vulnerabilities discovered in 2021? 10-June-2024

Leave a comment

Original posted 06/06/2024

Preface: Oracle and Citrix have large customer bases and use Xen as their primary hypervisor. Red Hat, SUSE, and Canonical support KVM as a virtualization option in their Linux versions. When it comes to cloud computing, administrators face a similar decision: Citrix and Oracle offer Xen-based offerings rather than Google’s KVM.

Background: In a hypervisor command shell, such as the Citrix Hypervisor dom0 shell or the VMware ESXi host shell. You can do the following command to verify your NVIDIA virtual GPU software version.

[root@vgpu ~]# nvidia-smi

|NVIDIA-SMI 550[.]90[.]05                Driver Version: 550[.]90[.]05

NVIDIA vGPU software can be used in a variety of ways. The method we mentioned here is related to this vulnerability. In GPU pass-through mode, an entire physical GPU is directly assigned to one VM, bypassing the NVIDIA Virtual GPU Manager. In this mode of operation, the GPU is accessed exclusively by the NVIDIA driver running in the VM to which it is assigned. The GPU is not shared among VMs.

Exploiting a buffer overflow vulnerability often involves manipulating pointers to redirect program execution or inject malicious code. By overwriting the return address of a function, an attacker can divert the control flow to a different section of the program where their code is placed.

Vulnerability details:

CVE‑2024‑0099 NVIDIA vGPU software for Linux contains a vulnerability in the Virtual GPU Manager, where the guest OS could cause buffer overrun in the host. A successful exploit of this vulnerability might lead to information disclosure, data tampering, escalation of privileges, and denial of service.

CVE‑2024‑0089 NVIDIA vGPU software for Linux contains a vulnerability in the Virtual GPU Manager, where the guest OS could execute privileged operations. A successful exploit of this vulnerability might lead to information disclosure, data tampering, escalation of privileges, and denial of service.

Official announcement: For detail, please refer to link – https://nvidia.custhelp.com/app/answers/detail/a_id/5551

AI and ML, Application Development, IoT, Potential Risk of CVE

CVE-2024-31335 GPU – PowerVR: Wrong order of operations in DevmemIntUnmapPMR2 may lead to temporarily dangling PTEs.AI accelerators called Neural Network Accelerator (NNA) staying alert! (7 June 2024)

Leave a comment

Official Posted: 31st May 2024

Preface: PowerVR not limited 2D and 3D rendering, and for video encoding, decoding, associated image processing. It also develops AI accelerators called Neural Network Accelerator (NNA). The IMG Series4 is a revolutionary neural network accelerator (NNA) for the automotive industry that enables ADAS and autonomous driving.

PowerVR accelerators are not manufactured by PowerVR, but instead their IP blocks of integrated circuit designs and patents are licensed to other companies.

Remark: An IP block is a reusable unit of logic, cell, or chip layout design and can be used as building block for various chip- and logic designs. By making this technology available NXP is opening up the opportunity for chip designers to leverage our building blocks in a wide assortment of on-chip solutions.

Background: What is DDK? To build the Android kernel and other kernel artifacts (modules, boot images, etc.), they provide a framework called “Kleaf”. For Android 14+, Kleaf is strongly recommended. One part of Kleaf is the Driver Development Kit (DDK) which is used to build external modules.

Vulnerability details: CVE-2024-31335 – GPU – PowerVR: Wrong order of operations in DevmemIntUnmapPMR2 may lead to temporarily dangling PTEs.

Software installed and run as a non-privileged user may conduct improper GPU system calls to trigger use-after-free kernel exceptions.

Official announcement: For detail, please refer to link –

https://www.imaginationtech.com/gpu-driver-vulnerabilities/#may24

AI and ML, Cell Phone (iPhone, Android, windows mobile), Potential Risk of CVE

CVE-2024-26926: Kernel – The vulnerability in this section could lead to local escalation of privilege in the kernel with no additional execution privileges needed (6 Jun 2024)

Leave a comment

Preface: In linux distributions the term ‘upstream’ (also applied to kernel) refers to the original version (as is released by software developers) of a program/software (kernel in your case) while ‘downstream’ refers to the software provided by linux distribution.

Background: There are many ways to communicate with IPC, such as: Shared Memory, Message Queue, PIPE, FIFO, Unix Socket, etc. A process cannot access another process’s memory. However, the kernel has control over all processes and therefore can expose an interface that enables IPC. In Binder, this interface is the /dev/binder device, which is implemented by the Binder kernel driver.

Ref: A Mutex is a Mutually exclusive flag. It acts as a gate keeper to a section of code allowing one thread in and blocking access to all others.

Vulnerability details:

Kernel -The vulnerability in this section could lead to local escalation of privilege in the kernel with no additional execution privileges needed.

Official announcement: For detail, please refer to link –

https://source.android.com/docs/security/bulletin/2024-06-01

AI and ML, Potential Risk of CVE

CVE-2024-22476: Improper input validation in some Intel® Neural Compressor software (5 June 2024)

Leave a comment

Original article published on 14-05-2024

Preface: Ancient humans hunted for survival. As times goes by, the evolution make them become intelligence biology. This pursuit of progress divided into different level of human. Human want is never ending. When Artificial Intelligence has born. It is the creator’s final blessing to human.

Background: Intel Neural Compressor performs model optimization to reduce the model size and increase the speed of deep learning inference for deployment on CPUs or GPUs.

Intel Neural Compressor aims to provide popular model compression techniques such as quantization, pruning (sparsity), distillation, and neural architecture search on mainstream frameworks such as TensorFlow, PyTorch, ONNX Runtime, and MXNet, as well as Intel extensions such as Intel Extension for TensorFlow and Intel Extension for PyTorch.

Vulnerability details:

CVEID:  CVE-2024-22476

Description: Improper input validation in some Intel® Neural Compressor software before version 2.5.0 may allow an unauthenticated user to potentially enable escalation of privilege via remote access.

CVSS Vector:  CVSS:3.1/AV:L/AC:H/PR:L/UI:N/S:U/C:H/I:N/A:N

Affected Products:

Intel® Neural Compressor software before version 2.5.0.

Official announcement: For detail, please refer to link –

https://www.intel.com/content/www/us/en/security-center/advisory/intel-sa-01109.html

AI and ML, Application Development, Network (Protocol, Topology & Standard), Potential Risk of CVE

CVE-2024-0103 – NVIDIA Triton Inference Server for Linux hit Incorrect Initialization of Resource vulnerability (31-05-2024)

Leave a comment

Preface: AI-powered systems analyse the severity of the vulnerability, potential impact, and exploitability and prioritise patches based on the criticality of the vulnerability. Perhaps AI contains self diagostic and do remedy by himself!

Background: An open-source software that helps standardize model deployment and delivers fast and scalable AI in production.

Vulnerability details:

CVE-2024-0103 Information disclosure

NVIDIA Triton Inference Server for Linux contains a vulnerability where a user may cause an incorrect Initialization of resource by network issue. A successful exploit of this vulnerability may lead to information disclosure.

Ref: For example, the minimum packet size is 60 bytes (the card typically adds a frame checksum to this, making the minimum packet size on the line 64 bytes). If you only have 40 bytes, then it will still transmit 60 bytes.

Because 40 bytes you send plus the next 20 bytes that happen to be sitting in the buffer beyond the 40 you intended to send.

If you haven’t explicitly initialized that area, those 20 bytes might well be data leftover from a previously sent packet, which may have belonged to some other connection. Or that memory could have previously been a data page for some program that was recently running (and hence could contain a password, or an encryption key or just about any kind of sensitive information).

Official details: For detail, please refer to link – https://nvidia.custhelp.com/app/answers/detail/a_id/5546

AI and ML, Potential Risk of CVE, System

CVE-2024-38016: tty: n_gsm: fix possible out-of-bounds in gsm0_receive() (29-05-2024)

Leave a comment

Preface: In the Linux Kernels n_gsm serial line discipline, which can be exploited by local attackers to gain kernel level root access. It original published by other Linux brand on 8th May 2024.

Background: In Unix systems, a tty (which is short for “teletypewriter”) is the standard representation of a terminal device, with at least input and output capabilities and usually much more. These were originally connected to serial ports, but most today are virtual terminals, connected to either a text-mode console (DOS-like) or a graphical terminal program (like xterm or gnome-terminal).

Vulnerability details: tty: n_gsm: fix possible out-of-bounds in gsm0_receive()

Assuming the following:

– side A configures the n_gsm in basic option mode

– side B sends the header of a basic option mode frame with data length 1

– side A switches to advanced option mode

– side B sends 2 data bytes which exceeds gsm->len

Reason: gsm->len is not used in advanced option mode.

– side A switches to basic option mode

– side B keeps sending until gsm0_receive() writes past gsm->buf Reason: Neither gsm->state nor gsm->len have been reset after reconfiguration.

Official details: For detail, please refer to link – https://www.tenable.com/cve/CVE-2024-36016

AI and ML, Potential Risk of CVE, System

Red Hat security advisory: Important – glibc security update (29-05-2024)

Leave a comment

Preface: You can clear the cache of nscd by performing the following actions:

Execute the following command: sudo /etc/init[.]d/nscd restart.

Background:

Nscd is a daemon that provides a cache for the most common name service requests. The default configuration file, /etc/nscd. conf, determines the behavior of the cache daemon.

The glibc packages provide the standard C libraries (libc), POSIX thread libraries (libpthread), standard math libraries (libm), and the name service cache daemon (nscd) used by multiple programs on the system. Without these libraries, the Linux system cannot function correctly.

The iconv() function shall convert the sequence of characters from one codeset, in the array specified by inbuf, into a sequence of corresponding characters in another codeset, in the array specified by outbuf. The codesets are those specified in the iconv_open() call that returned the conversion descriptor, cd.

Vulnerability details:

glibc: Out of bounds write in iconv may lead to remote code execution (CVE-2024-2961)

glibc: stack-based buffer overflow in netgroup cache (CVE-2024-33599)

 glibc: null pointer dereferences after failed netgroup cache insertion (CVE-2024-33600)

 glibc: netgroup cache may terminate daemon on memory allocation failure (CVE-2024-33601)

 glibc: netgroup cache assumes NSS callback uses in-buffer strings (CVE-2024-33602)

Official announcement: For detail, please refer to link – https://access.redhat.com/errata/RHSA-2024:3464

AI and ML, Potential Risk of CVE

CVE-2024-5274: Google Chrome fixed remote code execution vulnerability (28-05-2024)

Leave a comment

Preface: Every time I start learning CVE. It helps me enrich my knowledge.  Even though it was released months ago.

Background: V8 is a JavaScript and WebAssembly engine developed by Google for its Chrome browser. Each WebAssembly module executes within a sandboxed environment separated from the host runtime using fault isolation techniques.

Ref: wasmtime is a runtime for WebAssembly. The 19.0.0 release of Wasmtime contains a regression introduced during its development which can lead to a guest WebAssembly module causing a panic in the host runtime. A valid WebAssembly module, when executed at runtime, may cause this panic. This vulnerability has been patched in version 19.0.1.

Vulnerability details: This update includes 1 security fix. Below, we highlight fixes that were contributed by external researchers. Please see the Chrome Security Page for more information.

[N/A][341663589] High CVE-2024-5274: Type Confusion in V8. Reported by Clément Lecigne of Google’s Threat Analysis Group and Brendon Tiszka of Chrome Security on 2024-05-20

Google is aware that an exploit for CVE-2024-5274 exists in the wild.

Official announcement: For detail, please refer to link – https://chromereleases.googleblog.com/2024/05/stable-channel-update-for-desktop_23.html?m=1

AI and ML, Cell Phone (iPhone, Android, windows mobile), Potential Risk of CVE

CVE-2024-23354 Memory corruption when the IOCTL call is interrupted by a signal. (24May 2024)

Leave a comment

Originally published on May 6, 2024

Preface: The Snapdragon 8 Gen 2 Mobile Platform defines a new premium standard for connected computing. Intelligently engineered with groundbreaking AI across the board, this AI marvel enables truly extraordinary experiences.

Background: A vertex buffer object (VBO) is an OpenGL feature that provides methods for uploading vertex data (position, normal vector, color, etc.) to the video device for non-immediate-mode rendering.

KGSL allocates GPU-shared memory from its own page pool. A VBO is a buffer of memory which the gpu can access. That’s all it is. A VAO is an object that stores vertex bindings. This means that when you call glVertexAttribPointer and friends to describe your vertex format that format information gets stored into the currently bound VAO.

Vulnerability details: Memory corruption when the IOCTL call is interrupted by a signal.

Remedy: The VBO bind operation is often synchronous, and needs to be waited on by the ioctl thread. Allocate the completion struct used to synchronize between the ioctl and bind operation on the heap for simplicity.

Official announcement: Please refer to the link for details – https://nvd.nist.gov/vuln/detail/CVE-2024-23354

AI and ML, Network (Protocol, Topology & Standard), Potential Risk of CVE

CVE-2024-36008: The impact may be widespread but has been resolved. Linux, you did a great job. (21May 2024)

Leave a comment

Preface: Syzbot has begun to report kernel findings to LKML in 2017. Syzbot is a continuous kernel build / fuzz / report aggregation system.

Background: Linux has two mechanisms for setting routes, one is fib, and the other is routing generated by dynamic. fib uses route (man 8 route) to specify a static route table. What net/ipv4/route[.]c does is dynamic generate routing hash to speed up route decision.

Vulnerability details: In the Linux kernel, the following vulnerability has been resolved: ipv4: check for NULL idev in ip_route_use_hint() . syzbot was able to trigger a NULL deref in fib_validate_source() in an old tree .

It appears the bug exists in latest trees. All calls to __in_dev_get_rcu() must be checked for a NULL result.

Official announcement: Please refer to the link for details – https://nvd.nist.gov/vuln/detail/CVE-2024-36008