Razkrito je bilo, da obstaja kritična ranljivost (CVE-2026-114938) v OpenJDK, ki omogoča oddaljeno izvajanje kode ali izpust vsebine pomnilnika procesa pri napadih na projekte, napisane v Javi, ki uporabljajo privzeti dekoder podatkov v formatu Base64. Base64 se uporablja v protokolih za overjanje, pri nalaganju datotek na spletni strežnik, v elektronski pošti in v mnogih drugih situacijah, kjer je potrebna prenos binarnih podatkov v tekstovni obliki. Ranljivost vodi do zapisovanja podatkov, ki jih nadzoruje napadalec, v pomnilnik izven dodeljenega območja in je posledica napake v optimizacijah v asemblerju, ki uporabljajo AVX-512 VBMI (Vector Byte Manipulation Instructions).
Za sprožitev nujnega zaključka aplikacije je dovolj, da predamo Base64 zaporedje, sestavljeno iz 64 simbolov 'A', ki se konča s simbolom '='. Uspešni napadi za oddaljeno izvajanje kode, ki delujejo v fazi pred overjanjem, so bili dokazani za projekte Keycloak, Atlassian Jira, Atlassian Confluence, JFrog Artifactory, Jenkins in PingFederate. V fazi po overjanju pa je bilo mogoče izkoristiti Apache Kafka Broker, Apache Solr, Elasticsearch, Gerrit in Spinnaker.
A vendar so to le projekti, zajeti v začetnem testiranju, in glede na široko uporabo formata Base64 lahko težava vpliva na mnoge druge sisteme. Med drugim so bili razviti prototipi exploitov, ki lahko napadajo aplikacije, ki temeljijo na okvirjih Apache Struts in Spring Boot. Prav tako je bila prikazana možnost napada na odjemalske Java aplikacije, ki vzpostavljajo TLS povezavo s serverom, ki jo nadzorujejo napadalci. Tehnika izkoriščanja ni univerzalna in za vsako uporabo Base64 je potrebno ustvariti poseben exploit, vendar so raziskovalcem uspelo avtomatizirati generacijo podobnih exploitov s pomočjo AI.
Vulnerability manifests starting from OpenJDK 17.0.6 and affects derivative Java distributions, including Oracle Java SE, Oracle GraalVM, Amazon Corretto, Eclipse Temurin, Microsoft OpenJDK, Azul Zulu, BellSoft Liberica, SAP SapMachine, and Red Hat OpenJDK. Oracle has addressed the vulnerability in yesterday's patch update for its products, which also fixed 10 other vulnerabilities in addition to the one discussed. Vulnerabilities have been resolved in Java SE releases 27.0.1.1, 25.0.5.1, 21.0.13.1, 17.0.21.1, 11.0.33.1, and 8u513. The status of vulnerability fixes in the distributions can be evaluated on these pages: Debian, Ubuntu, SUSE/openSUSE, RHEL, Gentoo, Arch, Fedora, FreeBSD.
The attack is only possible on systems that support AVX-512 VBMI instructions. For Intel systems, the vulnerability has been observed on Intel Xeon CPUs Ice Lake, Sapphire Rapids, Emerald Rapids, and Granite Rapids, while for AMD systems, it affects EPYC and Ryzen CPUs starting from the Zen 4 microarchitecture. ARM64 architecture systems are not vulnerable to the issue. As a workaround to mitigate the vulnerability, OpenJDK can be launched with the flag "-XX:-UseBASE64Intrinsics" to disable assembly optimizations or "-XX:UseAVX=2" to turn off AVX-512.
To check for support of AVX-512 VBMI instructions in Linux, you can use the command (it should display 4 flags: avx512f, avx512bw, avx512vbmi, and bmi2): grep -m1 '^flags' /proc/cpuinfo | tr ' ' ' ' | grep -E '^(avx512f|avx512bw|avx512vbmi|bmi2)$'
The vulnerability exists in the assembly implementation of the Base64 decoder accelerated using AVX-512 VBMI vector instructions, which allow processing large data blocks in a single cycle using bit masks. The issue arises from incorrect calculation of the size of strings that end with the character '=', used as a padding character for incomplete blocks, when the length of the input data is not a multiple of 3 bytes.
During the processing of '=' characters, the decoder reduces the size of the resulting string since the final block contains 1 or 2 bytes instead of 3, but due to an error in the calculation, an integer underflow occurred when subtracting, and the presence of the '=' character at the end of a string of a specific size led to calculating a negative value of -1 or -2. These values were then treated as unsigned 64-bit numbers, resulting in the creation of a bit mask in which all bits are equal to 1 (-1u64).
Za ohranjanje rezultata dekodiranja pakiranega niza velikosti 64 bajta je bil dodeljen pomnilniški prostor velikosti 50 bajtov (v Base64 so vsakih 3 izvorni bajti kodirani v 4 znake), vendar je pri določanju bitne maske -1u64 v navodilu evmovdqub pri obdelavi končnih znakov prišlo do zapisa 64 bajtov namesto 2 na pozicijo s premikom 48, kar pomeni, da je bilo 62 bajtov zapisanih preko meje pomnilnika. Za izkoriščanje ranljivosti se izvede prepisovanje polja z velikostjo naslednjega niza, kar omogoča napadalcu priti do primitvov za poljuben dostop do branja in pisanja v pomnilniku.
Izvor: opennet.ru
