RE: [PATCH 14/15] kernel: convert futex_pi_state.refcount from atomic_t to refcount_t

From: Thomas Gleixner
Date: Mon Jul 17 2017 - 13:57:48 EST


On Mon, 17 Jul 2017, Reshetova, Elena wrote:
> > On Mon, 17 Jul 2017, Elena Reshetova wrote:
> > > refcount_t type and corresponding API should be
> > > used instead of atomic_t when the variable is used as
> > > a reference counter. This allows to avoid accidental
> > > refcounter overflows that might lead to use-after-free
> > > situations.
> >
> > Copying the same sentence over and over avoids thinking about a proper
> > changelog, right? You fail to explain, how you come to the conclusion that
> > futex_pi_state.refcount is a pure reference counter (aside of the name) and
> > therefor can be safely converted to refcount_t.

> OK, this is not very useful for many cases. Yes, I am using automated log
> on these patches, because I used to have 240 of them and writing manual
> logs for them would be fun.

Been there, done that.

> Moreover, in many cases, writing manual logs don't bring any value since
> I would have to repeat the same things all over again: xyz conversions
> was found by using *.cocci pattern first, then looked at manually and it
> looked like a standard reference counter that frees the things after
> calling refcount_dec_and_test() (or its variation with lock which is
> rare). Other things also looked correct, like I didn't see increments
> from zero, counter starts at 1 etc. I would really have to repeat the
> same thing in each changelog. Does it really bring value?

You don't have to go into that level of detail, but you can provide enough
information with a template as well, e.g.:

atomic_t variables are often used to implement pure reference counters:
- starting at 1
- freeing a resource after reaching zero
- only using basic atomic operations (init, inc, dec_and_test)

These variables should be converted to refcount_t because the refcount_t
operations can catch and prevent accidental underflows and overflows.

The variable FOO is used as a pure reference counter. Convert it to
refcount_t and fix up the operations.

That gives enough context for someone who looks at a patch because then the
reviewer can look for:

starts at 1, frees at 0, does not use any fancy operations

and has not to use Gurgle to figure out what your understanding of
reference counters is.

Replacing FOO with the real variable name can be done with a script easy
enough.

Thanks,

tglx