[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <ZpbAlnAeCv-eiNzF@google.com>
Date: Tue, 16 Jul 2024 18:48:54 +0000
From: Carlos Llamas <cmllamas@...gle.com>
To: Todd Kjos <tkjos@...gle.com>
Cc: Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
Arve Hjønnevåg <arve@...roid.com>,
Todd Kjos <tkjos@...roid.com>, Martijn Coenen <maco@...roid.com>,
Joel Fernandes <joel@...lfernandes.org>,
Christian Brauner <brauner@...nel.org>,
Suren Baghdasaryan <surenb@...gle.com>,
Alice Ryhl <aliceryhl@...gle.com>, linux-kernel@...r.kernel.org,
kernel-team@...roid.com, syzkaller-bugs@...glegroups.com,
stable@...r.kernel.org,
syzbot+3dae065ca76952a67257@...kaller.appspotmail.com
Subject: Re: [PATCH] binder: fix descriptor lookup for context manager
On Tue, Jul 16, 2024 at 10:40:20AM -0700, Todd Kjos wrote:
> If context manager doesn't need to be bit 0 anymore, then why do we
> bother to prefer bit 0? Does it matter?
>
> It would simplify the code below if the offset is always 0 since you
> wouldn't need an offset at all.
Yes, it would make things simplier if references to the context manager
could get any descriptor id. However, there seems to be an expectation
from libbinder that this descriptor would be zero. At least according to
some folks more familiar with userspace binder than myself.
I think we can revisit this expectation though and also look closer at
the scenario of a context manager "swap". The procs can still reach the
new context manager using descriptor 0. However, this may cause some
issues with operations with refs such as BC_INCREFS/BC_DECREFS.
AFAICT, the context manager doesn't even need a reference. But while we
dig furhter into this I think the best option is to keep the behavior
the same for now: reserve descriptor zero for the context manager node
unless it's already taken. Changing this is non-trivial IMO.
--
Carlos Llamas
Powered by blists - more mailing lists