On Wed, Apr 30, 2025 at 05:23:19PM +0200, Michel Casabona wrote:
Le 30/04/2025 à 14:48, Colin Watson a écrit :
Could you drop --leak-check=full from the valgrind call, and instead add
--main-stacksize=67108864 (i.e. eight times the current value)? Then
provoke the bug again and send me the new valgrind output. Let's see if
that tells us something different.
Same output :-( Log attached.
So mysterious!
Could I also get your /etc/ssh/sshd_config and /etc/ssh/sshd_config.d/*
files (of course you can edit out anything secret, but if you do then
please at least keep the structure)?
The (unedited) config files are attached too
I'm trying to get my test system closer to yours, but no luck so far.
The best I've been able to come up with is an overlap between source and
destination in a strlcpy call, which should probably be fixed, but
there's no sign of it in your output. I'm trying to provoke a
similar-ish segfault manually to experiment with coredumps, but haven't
quite had enough time so far.
What's the username you're trying to ssh to? (This is surely very
unlikely to matter, but you never know.)
What does "ls -l ~TARGET_USERNAME/.ssh/" for the appropriate target
username _on the server_ say?
Can I confirm that you're on amd64, just as the original reporter was?
Is there anything else at all unusual about your system? Antivirus, any
other system-wide LD_PRELOADs, any locally-modified packages that aren't
from Debian testing, ...?
--
Colin Watson (he/him) [cjwat...@debian.org]
