FS#73820 - [openssh] enable spectre mitigation flags

Attached to Project: Arch Linux
Opened by T.J. Townsend (blakkheim) - Wednesday, 16 February 2022, 22:23 GMT
Last edited by T.J. Townsend (blakkheim) - Tuesday, 06 June 2023, 14:21 GMT
Task Type Feature Request
Category Packages: Core
Status Closed
Assigned To Levente Polyak (anthraxx)
Giancarlo Razzolini (grazzolini)
Architecture All
Severity Low
Priority Normal
Reported Version
Due in Version Undecided
Due Date Undecided
Percent Complete 100%
Votes 0
Private No

Details

Description:
The OpenSSH configure script will automatically append -mretpoline to CFLAGS (for spectre v2 userland mitigation) if clang is used as the compiler. This diff adds clang to the makedepends line and exports it as the CC/CXX used for the build.

Additional info:
One could argue that -mretpoline should be enabled by default on Arch's clang package, and that everything should be built with clang instead of GCC for this (and other) security benefits, but that's beyond the scope of this proposal.
   openssh.diff (0.9 KiB)
This task depends upon

Closed by  T.J. Townsend (blakkheim)
Tuesday, 06 June 2023, 14:21 GMT
Reason for closing:  Won't implement
Comment by Allan McRae (Allan) - Wednesday, 16 February 2022, 23:36 GMT
There are equivalent flags for GCC - something like -mindirect-branch=thunk.
Comment by Evangelos Foutras (foutrelis) - Thursday, 17 February 2022, 04:23 GMT
Upstream originally included flags for GCC too but later dropped them. [1] [2] [3]

The GCC flags are also incompatible with -fcf-protection which we use by default. [4]

I'm not sure if this warrants switching compilers. Is Spectre even a concern for SSH?

[1] https://github.com/openssh/openssh-portable/commit/3377df00ea3f
[2] https://github.com/openssh/openssh-portable/commit/1673274aee67
[3] https://lists.archive.carbon60.com/openssh/dev/72261
[4] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87412
Comment by T.J. Townsend (blakkheim) - Thursday, 17 February 2022, 04:30 GMT
I would say Soectre is still a concern for anything that puts secrets into memory, so SSH is one of the best candidates for the added protection. This is probably why upstream defaults to always enabling it if it's available. Just my 2 cents.
Comment by T.J. Townsend (blakkheim) - Tuesday, 06 June 2023, 14:21 GMT
Florian at RedHat sent me the following reply about the subject, so I'm gonna close this task.

"As far as I understand it, recommendations for the code sequences to use
occasionally change and depend on CPU and microcode versions, so this
seems feasible only with run-time code generation. The original
retpolines are incompatible with SHSTK. Furthermore, avoiding indirect
jumps in the PLT needs run-time code generation with additional ASLR
constraints (or prelink-style ahead-of-time ASLR).

So I don't think the situation has changed."

Loading...