Fabian Keil found the reason for a SIGBUS:
In the "gpg-agent --daemon" case, main() calls pth_kill()
after the client has been forked, so when es_deinit() is
called on exit, acquiring the estream_list_lock seems to
cause pth to dereference a pointer located in a memory
region that has previously been free()'d.
My approach to fix it is different than his suggestion. It should
allow to continue all estream operations after a pth_kill except for
restarting pth.
From Grant Olsons's mail to gnupg-devel@:
This is three-quarters NIIBE and one-quarter me, but I though it would
be easier to have a single complete patch that applies to
STABLE-BRANCH-2-0 than two patches floating around.
This is basically NIIBE Yatuka's patch here:
http://lists.gnupg.org/pipermail/gnupg-devel/2010-November/025828.html
Plus another check that closes down the reader explicitly if we get a
SW_HOST_NO_READER error, so that an unplugged reader doesn't hold onto
the first slot, preventing us from getting to the active plugged reader.
overflow when picking an algorithm (not a security issue since we
can't pick something not present in all preference lists, but we might
pick something that isn't scored first choice).
* pkclist.c (select_algo_from_prefs): Slightly improve the handling of
MD5 in preference lists. Instead of replacing MD5 with SHA-1, just
remove MD5 from the list altogether, and let the next-highest ranked
algorithm be chosen.
