Do not ignore arc_adjust() return value

[mav]

Summary

Looking through the ARC back pressure code I see a scenario when I think ARC may not shrink as fast as it could:

1. arc_size < arc_c and arc_adjust() does not evict anything, returning zero to arc_reclaim_thread();
2. arc_available_memory() reports memory pressure, which can not be satisfied by arc_kmem_reap_now();
3. arc_shrink() reduces arc_c and calls arc_adjust(), return of which is ignored;
4. even if the last arc_adjust() could not satisfy arc_size < arc_c, arc_reclaim_thread() will still go to sleep, since the first one returned zero().

This patch supposed to cover that case.

Can anybody see any flaw in my logic?

[markj]

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/arc.c

4921 ↗

(On Diff #50220)

Should this be +=? If the first call returned a non-zero value, we may clobber it here (though I suspect that is quite unlikely in practice).

[allanjude]

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/arc.c

4921 ↗

(On Diff #50220)

I agree

[mav]

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/arc.c

4921 ↗

(On Diff #50220)

I thought about this for some time too. Specific value here is not important, only zero or non-zero. Zero here can be only in two cases: either there is nothing to evict, which is unlikely, or we couldn't evict anything at the last call. In either case I see no real reason why we would care what the first call returned. But I see no problem from += there, it will just do another try, not a dead loop, so I can add it if you like.