Hi,
here's version two of the patch. The important change is that now
throttle_group_restart_tgm() is protected by tg->lock and there's new
logic there.
Note that in case (2) it is theoretically possible for two different
tgms to run throttle_group_restart_queue() in parallel, but I think
this is safe because the actual restart of the queue is once more
protected by the lock.
The alternative that I initially considered (doing nothing if another
tgm is waiting for a timer) is probably not a good idea because
throttle_group_restart_tgm() is used by callers who want to drain a
tgm so I think that the requests should be processed immediately:
if (qatomic_fetch_inc(&tgm->io_limits_disabled) == 0) {
throttle_group_restart_tgm(tgm);
}
Once more, please test this carefully to see if it solves issue #3194.
I have only run the basic qemu iotests and I haven't touched this code
in a long time :)
Regards,
Berto
Alberto Garcia (1):
throttle-group: Fix race condition in throttle_group_restart_queue()
block/throttle-groups.c | 70 +++++++++++++++++++++++++++++++----------
1 file changed, 54 insertions(+), 16 deletions(-)
--
2.47.3
