Submit

Back to the channels

#java

  1. silla123 hello
  2. jbosmans Java RCE if the server cpu supports avx512 https://calif.io/java4shell
  3. nevet Java4Shell: Write once, RCE everywhere
  4. enoq looking at the latest java4shell vulnerability, there are no eclipse temurin docker images available yet, correct?
  5. enoq what version does 25.0.5_8 correspond to?
  6. enoq I suppose 25.0.5+8 which is not 25.0.5.1
  7. jbosmans it's pretty fresh so i think the images and temurin itself still needs to be updated. For vuln systems I'd look into the mitigations to switch from avx512 to avx2 support
  8. Para Add -XX:-UseBASE64Intrinsics to your JVM opts by whatever mechanism works for your case, redeploy, onwards to next CVE.
  9. Para Specifically this requires the AVX-512 VBMI instruction which is quite rare in reality, rough estimate is that only about 15-20% of CPUs out there even have it, and it's entirely possible cloud service providers haven't enabled it in hypervisor settings anyway, so the attack surface to mitigate is very confusing.
  10. [twisti] 'quite rare' ... '20%'
  11. [twisti] well i certainly relaxed, probably wont even bother to look into it
  12. Para That 20% includes AMD Zen4 and Zen5 processors, which are mostly home user CPUs.
  13. enoq thank you
  14. sonOfRa I've switched a bunch of systems to - XX:UseAVX=2 instead of wholly disabling intrinsics. With java -XX:+PrintFlagsFinal -version, you can quickly find out if the system you're on supports avx512- there will be a UseAVX value of 3 in the output somewhere. If it says 2, or nothing at all, at least that particular jvm on your system is unaffected
  15. enoq sonOfRa: so you expect there to be more avx issues?
  16. sonOfRa No, it's a tradroff. My workload in this case is keycloak and due to the whole jwt deal it does a lot of base64 - so disabling base64 acceleration probably loses me more than falling back to avx2 for everything
  17. sonOfRa I'll see in my cpu metrics after a day or two. Performance might even improve, because avx512 is not always actually a speedup due to silly cpu architecture reasons
  18. deebo yeah was just about to ask what minor version does temurins 25.0.5+8 correspond to, is it still vulnerable?
  19. enoq deebo: still vulnerable
  20. deebo https://github.com/adoptium/adoptium-support/issues/1575 yep
  21. nevet ETA for Temurin 25 and 21 releases fixing CVE-2026-114938 (Base64 intrinsic memory corruption, "Java4Shell") · Issue #1575 · adoptium/adoptium-support
  22. sonOfRa enoq: first 2 hours of production workload show no statistically significant performance deviation from our 30 day mean after setting useavx=2
  23. sonOfRa YMMV
  24. Para It's also somewhat framework dependent if you even hit the native base64 path. Run something old/backwards compatible enough and it uses e.g. Commons Codec instead.
  25. deebo also requires 4 cpu features to be available, those are reserved for slop agents, not to run real software!
  26. enoq had a look at some claude generated java code recently btw
  27. enoq loads of while and for loops with break and continue statements
  28. enoq refactoring took as long as implementing the feature from scratch
  29. dreamreal enoq: that sounds like whoever was driving did a poor job
  30. deebo yeah needs more handholding for style, or a checker tool that says no
  31. deebo should finally setup spotless and some static analyzer
  32. deebo sonarqube was a lifesaver until they decided they wont support java>8 in the "free" version
  33. cheeser deebo: spotless is great. i have it autowired in to my build cycles.
  34. dreamreal I do too
  35. dreamreal cheeser: what happened to javabot? Should I unmute nevet?
  36. sonOfRa pmd+spotless as a pre-commit hook works wonders to rein in claude
  37. Optic i just checked all our physical servers against the cpu capabilities required for java4shell, and we don't own a single one with avx512 :P
  38. Optic we have a couple with bmi2
  39. Optic sometimes shit old hardware is a win??
  40. dreamreal all of our deployed servers are technically vulnerable but don't expose endpoints that would provide easy ingress. So naturally we're sure we're invulnerable! *sigh* (We're deploying with the flag even so. "invulnerable" is a descriptor that only means "we don't know the ingress yet.")
  41. Optic i'm going to wait for the November patch cycle to release and calmly deploy that everywhere
  42. dreamreal We just deployed new images with the flags in place
  43. dreamreal coretto images are already patched but they're 120+ MB larger than temurin and we don't rely on the base64 decoding enough for it to impact us negatively
  44. Optic all our stuff is in spring-boot containers, so when liberica pubs the new versions it will just be a CI trigger + deploy to update
  45. Para Just in time for the next CVE, nested KVMs.
  46. Zapek where did oracle hide the graalvm with the base64 fix...
  47. dreamreal right there behind the paywall!
  48. Para Hm, I should probably slap pmd+spotless+whatever else into my template repo.
  49. cheeser dreamreal: restarting it now
  50. dreamreal cheeser: I tried to start it yesterday but it was apparently unsuccessful
  51. deebo proper enterprise java is on java8 and thus invulnerable
  52. * dreamreal snrks
  53. Para And definitely not on x86 derivatives.
  54. Para (that's too new)
  55. Para ((my Hetzner VM is protected because its CPU is ancient and doesn't support any AVX))
  56. dreamreal cheeser: what happened when I tried to start it?
  57. nimaje deebo: but at least some java8 versions are vulnerable, oracle says oracle java se 8u511b33-perf is vulnerable. No idea if any other as well, or why they just call out that specific version, is that just their currently supported java8 version? Does anyone else still support some java8 version?
  58. cheeser dreamreal: i didn't see anything about that.
  59. dreamreal *nod* I started it and waiting for a few minutes and it never showed up here, didn't do any triage
  60. sbalmos dreamreal: maybe it's like Python where if you're missing something, it just silently does nothing on startup. :D

Search posts