Re: [nft PATCH v4 1/7] src: Fix cache_flush() in cache_needs_more() logic

[Eric Garver <eric@garver.life>]

On Tue, May 28, 2019 at 11:03:17PM +0200, Phil Sutter wrote:
> Commit 34a20645d54fa enabled cache updates depending on command causing

Actual hash is eeda228c2d17. 

> it. As a side-effect, this disabled measures in cache_flush() preventing
> a later cache update. Re-establish this by setting cache->cmd in
> addition to cache->genid after dropping cache entries.
> 
> While being at it, set cache->cmd in cache_release() as well. This
> shouldn't be necessary since zeroing cache->genid should suffice for
> cache_update(), but better be consistent (and future-proof) here.
> 
> Fixes: 34a20645d54fa ("src: update cache if cmd is more specific")

Actual hash is eeda228c2d17. 

> Signed-off-by: Phil Sutter <phil@nwl.cc>
> ---
> Changes since v1:
> - Adjust cache_release() as well, just to make sure.
> ---
>  src/rule.c | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/src/rule.c b/src/rule.c
> index 326edb5dd459a..966948cd7ae90 100644
> --- a/src/rule.c
> +++ b/src/rule.c
> @@ -299,12 +299,15 @@ void cache_flush(struct nft_ctx *nft, enum cmd_ops cmd, struct list_head *msgs)
>  
>  	__cache_flush(&cache->list);
>  	cache->genid = mnl_genid_get(&ctx);
> +	cache->cmd = CMD_LIST;
>  }
>  
>  void cache_release(struct nft_cache *cache)
>  {
>  	__cache_flush(&cache->list);
>  	cache->genid = 0;
> +	cache->cmd = CMD_INVALID;
> +
>  }
>  
>  /* internal ID to uniquely identify a set in the batch */
> -- 
> 2.21.0

Otherwise looks good.