Re: [PATCH net] ipv4: include NLM_F_APPEND flag in append route notifications

[Alexander Duyck <alexander.duyck@gmail.com>]

On 06/17/2015 07:50 AM, roopa wrote:
> On 6/17/15, 12:50 AM, Scott Feldman wrote:
>> On Tue, Jun 16, 2015 at 9:11 AM, Roopa Prabhu
>> <roopa@cumulusnetworks.com> wrote:
>>
>> [snip]
>>
>>> @@ -1203,6 +1204,8 @@ int fib_table_insert(struct fib_table *tb,
>>> struct fib_config *cfg)
>>>
>>>                  if (!(cfg->fc_nlflags & NLM_F_APPEND))
>>>                          fa = fa_first;
>>> +               else
>>> +                       nlflags |= NLM_F_APPEND;
>>>          }
>> The if and else parts above don't seem logically related.
> I have it at this place because here is where the decision to append
> really happens.
>>   Maybe you
>> could initialize nlflags as:
>>
>>      unsigned int nlflags = cfg->fc_nlflags &
>> (NLM_F_REPLACE|NLM_F_APPEND);
>>
>> And then pass rtmsg_fib(..., nlflags) to avoid the flag test/set?
> nlflags should only contain NLM_F_REPLACE and NLM_F_APPEND if a replace or
> append really took place. Hence the check and setting of nlflags is at
> the place where that
> decision is made.
>
> I had tried this patch a couple of other ways.... Do you think the below
> would be less confusing ?
>
> thanks.
>
>
>
> diff --git a/net/ipv4/fib_trie.c b/net/ipv4/fib_trie.c
> index 3c699c4..9bfa3d8 100644
> --- a/net/ipv4/fib_trie.c
> +++ b/net/ipv4/fib_trie.c
> @@ -1082,6 +1082,7 @@ int fib_table_insert(struct fib_table *tb, struct
> fib_config *cfg)
>          struct trie *t = (struct trie *)tb->tb_data;
>          struct fib_alias *fa, *new_fa;
>          struct key_vector *l, *tp;
> +       unsigned int nlflags = 0;
>          struct fib_info *fi;
>          u8 plen = cfg->fc_dst_len;
>          u8 slen = KEYLENGTH - plen;
> @@ -1189,8 +1190,9 @@ int fib_table_insert(struct fib_table *tb, struct
> fib_config *cfg)
>                          fib_release_info(fi_drop);
>                          if (state & FA_S_ACCESSED)
> rt_cache_flush(cfg->fc_nlinfo.nl_net);
> +                       nlflags |= NLM_F_REPLACE;
>                          rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
> -                               tb->tb_id, &cfg->fc_nlinfo, NLM_F_REPLACE);
> +                               tb->tb_id, &cfg->fc_nlinfo, nlflags);
>
>                          goto succeeded;
>

Why even bother modifying this part?  Is this actually needed at all, 
are there some other flags you plan to drop into nlflags as well that 
would be passed as a part of this message?

> @@ -1201,7 +1203,9 @@ int fib_table_insert(struct fib_table *tb, struct
> fib_config *cfg)
>                  if (fa_match)
>                          goto out;
>
> -               if (!(cfg->fc_nlflags & NLM_F_APPEND))
> +               if (cfg->fc_nlflags & NLM_F_APPEND)
> +                       nlflags |= NLM_F_APPEND;
> +               else
>                          fa = fa_first;
>          }
>          err = -ENOENT;

I'm not sure I see the point of using the |=.   Why not just use a = and 
save yourself an instruction or two since you don't really need the OR 
operator in this case.

> @@ -1238,7 +1242,7 @@ int fib_table_insert(struct fib_table *tb, struct
> fib_config *cfg)
>
>          rt_cache_flush(cfg->fc_nlinfo.nl_net);
>          rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
> -                 &cfg->fc_nlinfo, 0);
> +                 &cfg->fc_nlinfo, nlflags);
>   succeeded:
>          return 0;
>
>