@david_chisnall@infosec.exchange
Post #1833336
2026-04-22 07:33 UTC
@bms48 @downey
Two security advisories today were from Claude. One allowed a process to become root due to a kernel use-after-free. If you detach from the controlling terminal, the process structure now has a dangling pointer in it. This is pretty important, though I’m curious whether it can be a jail escape, since the jail is stored elsewhere in the process structure (I presume not). I think this code has been around for a while and it’s very much a ‘that’s what happens if you write C’ kind of bug. Newer languages give you tools for avoiding this kind of thing by construction.
The other bug is in page-table handling for MPK. This is a pretty useless CPU feature. You can use it (in combination with the landing-pads feature and some binary patching) to build a fairly efficient sandbox, but it’s Intel making a bad copy of an AArch32 feature. A bit embarrassing that whoever did this didn’t understand how x86 page tables work. Again, this is the kind of thing that happens when every part of a system needs to understand the data structures it’s modifying. Apple rewrote their pmap in C++ a few years ago and this would not have been possible in their model because the page-table manipulation code would have gone through something that understands superpages.
Replies (1)
-
@bms48@mastodon.social 2026-04-22 07:43
@david_chisnall @downey I don't have any Intell SGX capable hardware at the moment which irks me as enforcing capabilities at hardware level is potentially interesting to me in future. However there are workarounds for threaded applications, they just suck from a raw throughput p.o.v.