On Wed, May 09, 2012 at 05:12:03PM +0100, Al Viro wrote: > On Wed, May 09, 2012 at 05:25:14PM +0200, Sasha Levin wrote: > > Hi all, > > > > I've started seeing the following warning while fuzzing inside a KVM guest with the latest -next: > > [->read() may grab ->cred_guard_mutex, but it may itself be called by > prepare_binprm() after having ->cred_guard_mutex grabbed] > > Nasty, that... What's more, it's not just prepare_binprm() itself - > ->load_binary() might end up calling read(); it doesn't have to > limit itself to mmap(), so essentially anything that can be grabbed > by ->read() of a regular file might nest under ->cred_guard_mutex. > > AFAICS, /proc/*/stack, /proc/*/syscall, /proc/*/personality, > /proc/*/io_accounting, /proc/*/auxv, /proc/*/environ, /proc/*/*maps > and /proc/*/pagemap have ->cred_guard_mutex grabbed on read. seq_file > is a red herring here - io_accounting has the same issue and it does > things directly, without seq_read(). > > It's not a realistic attack, fortunately, since you need root > to get past open_exec() on any of those... Wait. How _did_ you get > past open_exec(), anyway? MAY_EXEC is not supposed to be granted on > anything that has no exec bits at all and AFAICS none of those files > have them. FWIW, that's _probably_ a false positive, but I really wonder what has triggered it. It would take seq_file-based file somewhere with _some_ exec bits set (otherwise it shouldn't have been seen by prepare_binprm()). The file itself isn't one of those that grab ->cred_guard_mutex anywhere in their ->read(), but since lockdep can't tell one seq_file from another, we get the warning. The interesting part is who the hell had managed to do executable seq_file-based anything - false positive or not, it's almost certainly a bug... -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
