Bugzilla Bug 116822

from pinentry-0.7.2.ebuild:
58 for x in curses qt
59 do
60  [ -f ${D}/usr/bin/pinentry-${x} ] && fperms +s /usr/bin/pinentry-${x}
61 done

I think the developer meant u+s, this needlessly gives users the ability to
read and overwrite files with gid 0 (also, is giving a qt application suid root
a great idea? an strace suggests it reads numerous configuration files, i'm
sure a local root could be found in this package).

# cat secret-file
secret
$ echo > secret-file
-bash: secret-file: Permission denied
$ ls -l secret-file
-rw-rw----  1 root root 7 Dec 26 23:57 secret-file
$ pinentry-curses 
OK Your orders please
OPTION ttyname=secret-file
OK
OPTION ttytype=dumb
OK
GETPIN
^C
# cat secret-file
...

swegener please comment/fix.

I added a -r2 with the fperms call fixed. For the general suid problem, I'd
like
to mention that pinentry drops the privileges very early, directly after
locking
the memory.

Thx Sven.
Should we call for stableization on the -r2 ?

Bah, let's do it.
Target KEYWORDS=KEYWORDS="alpha amd64 ia64 ~mips ppc sparc x86"

amd64 stable

sparc stable.

x86 stable

ppc stable

Stable on alpha now.

Cheers,
Ferdy

Thx everyone.
GLSA 200601-01
ia64 should mark stable to benefit from GLSA

Folks,

it seems that you did not understand what pinentry is doing.  It is designed to
make sure that passpharses are entered as securely as possible.  For Gtk+ we
even wrote a new secure entry widget to make use of non-swappable memory (in
gpg parlance "secure memory").  When using Linux the required mlock() will only
succeed for euid==0.  Thus pinentry allocates the memory right at startup and
then drops the privileges.

So for extra security it makes actually sense to install it suid(root). However
this is not my descision and the same rationale used to install gpg suid(root)
should be applied here.  If gentoo installs Linux with an encrypted swap, there
is no need to install it suid(root) but in other cases one might want to do it.

Note, that the whole secure memory business is limited by the IPC mechanism
used:  The passphrases is eventually passed trough pipe buffers; we expect that
they are soon reused and zeroed out.

Hey werner, the bug was that the ebuild incorrectly set the sgid bit, we still
install pinentry with the suid bit set. This was a bug in our ebuild, not with
pinentry btw :)