On 3/27/19 7:30 PM, Alexei Starovoitov wrote:

quoted

quoted

imo it would be cleaner not to mess with cfg.
Ideally it should be marked 'const'.

Existing code sets those flags but on a fib6_info. This is not used for
nexthop objects and is kept here to not duplicate this if branch in the
create_info that uses it. This check affects both which device is used
as well as the flags.

What stopping you from doing fib6_nh->nh_flags |= RTF_REJECT | RTF_NONEXTHOP ?
cfg should really be const.

Because it is not a nexthop flag, it is a prefix flag.

quoted

quoted

lwtstate_put() is missing in the error path of existing code.
Is this a bug fix?
Why there is nothing about this in commit log?

Existing code has a different cleanup path.

This is done explicitly here per a request from Ido in v1 that the new
function be symmetric in its cleanup on an error.

I saw that comment.
What I don't see is where existing code doing that cleanup.
Could you please point it out?

fib6_info_release -> fib6_info_destroy_rcu