On Thu, Sep 28, 2023 at 09:00:44PM +0200, Florian Westphal wrote: > Phil Sutter <phil@xxxxxx> wrote: > > Eliminate the direct use of netlink_callback::args when dumping rules by > > casting nft_rule_dump_ctx over netlink_callback::ctx as suggested in > > the struct's comment. > > > > The value for 's_idx' has to be stored inside nft_rule_dump_ctx now and > > make it hold the 'reset' boolean as well. > > > > Note how this patch removes the zeroing of netlink_callback::args[1-5] - > > none of the rule dump callbacks seem to make use of them. > > Do you think we can fix the reset race in -next instead of -nf? > > If yes, you could detach preparation patches like this one and > split the series in several batches. Yes, I noticed this series is no longer the "add some spinlock to prevent races" it was in the beginning. TBH, I chose nf mostly because nf-next lacked a commit I needed. But it's there now, so v3 will address nf-next. Thanks, Phil
