RE: [net-next 0/4] devlink: add dry run support for flash update
From: "Keller, Jacob E" <jacob.e.keller@intel.com>
Date: 2021-10-09 00:32:59
-----Original Message----- From: Jakub Kicinski <kuba@kernel.org> Sent: Friday, October 08, 2021 5:18 PM To: Keller, Jacob E <jacob.e.keller@intel.com> Cc: Jiri Pirko <jiri@resnulli.us>; netdev@vger.kernel.org Subject: Re: [net-next 0/4] devlink: add dry run support for flash update On Fri, 8 Oct 2021 23:58:45 +0000 Keller, Jacob E wrote:quoted
quoted
quoted
Doesn't the policy checks prevent any unknown attributes? Or are unknown attributes silently ignored?Did you test it? DEVLINK_CMD_FLASH_UPDATE has GENL_DONT_VALIDATE_STRICT set.Hmm. I did run into an issue while initially testing where DEVLINK_ATTR_DRY_RUN wasn't in the devlink_nla_policy table and it rejected the command with an unknown attribute. I had to add the attribute to the policy table to fix this. I'm double checking on a different kernel now with the new userspace to see if I get the same behavior.Weird.quoted
I'm not super familiar with the validation code or what GENL_DONT_VALIDATE_STRICT means... Indeed.. I just did a search for GENL_DONT_VALIDATE_STRICT and the only uses I can find are ones which *set* the flag. Nothing ever checks it!! I suspect it got removed at some point.. still digging into exact history though...It's passed by genl_family_rcv_msg_doit() to genl_family_rcv_msg_attrs_parse() where it chooses the netlink policy.
Ah.. I see how its done. It's passed as the argument so you don't see a direct comparison which makes it look like there isn't one... Feels like there could probably be a better abstraction that was more readable here... Anyways. I'll confirm what happens on the kernel that doesn't have the attribute defined at all. I wonder if the thing I saw differently was because the attribute *was* known but wasn't in policy. I.e. because it was defined it was validated.... Yep, I confirm that on a kernel without the DRY_RUN flag that it would allow the run because we aren't being strict. I am guessing that we can't convert devlink over to strict validation? Thanks, Jake