On Tue 30-01-18 11:53:58, Eric Dumazet wrote: [...] > How is __GFP_NORETRY working exactly ?
this is what the documentation says. * __GFP_NORETRY: The VM implementation will try only very lightweight * memory direct reclaim to get some memory under memory pressure (thus * it can sleep). It will avoid disruptive actions like OOM killer. The * caller must handle the failure which is quite likely to happen under * heavy memory pressure. The flag is suitable when failure can easily be * handled at small cost, such as reduced throughput > Surely, if some firewall tools attempt to load a new iptables rules, we > do not want to abort them if the request can be satisfied after few > pages moved on swap or written back to disk. I am not sure this really goes along with "namespace admin can request arbitrary amount of memory" very well. > We want to avoid huge allocations, but leave reasonable ones succeed. Yes, that would be the best way forward. From the previous discussion with Florian [1] it seems that "reasonable" is not that easy to figure out. Anyway, this patch merely gets us back to pre eacd86ca3b03 times where __GFP_NORETRY has been used for both kmalloc and vmalloc paths. So it is more a quick band aid than a longterm solution. [1] http://lkml.kernel.org/r/20180129165722.gf5...@breakpoint.cc -- Michal Hocko SUSE Labs
