Re: [PATCH 1/1] migration/multifd: fix channel count TOCTOU race on cancel and retry

[Fabiano Rosas <farosas@suse.de>]

Peter Xu <peterx@redhat.com> writes:

> On Thu, Apr 23, 2026 at 03:13:42PM -0300, Fabiano Rosas wrote:
>> Looking again at this argument I put (too many variables), I notice we
>> also have multifd_send_state->channels_ready and
>
> channels_ready seems to be special?  In busy systems I think it should
> normally always be less than the number of threads on sender side, because
> some of them will be busy.
>

Argh, sorry, I meant channels_created!

>> multifd_recv_state->count, both of which should contain the right number
>
> This one is used so far to just count how many channels are established on
> dest side.  Logically after setup phase it can be gone.  It's just not hurt
> to be available until cleanups.
>
>> of channels after multifd_send_setup() and migration_start_incoming(),
>> respectively. Maybe we could unify all of this into a single semaphore
>> used in both sides and take the semaphore count as number of
>> channels. @Peter, do you think it's worth it?
>
> My memory is reading counter from a sem isn't always reliable.  I always
> only use sem as thread-safety purpose primitives.
>
> So if I was correct on channels_ready above, then it may not apply on the
> idea to merge.  We're essentially trying to save one 8B for ->count.. and
> it can only be gone after it's finished its use (IOW, we still need it to
> work before recv side setup).
>
> Just leave them as-is?