Example for custom generic netlink family with custom attributes #103

phip1611 · 2020-12-05T19:28:03Z

I'd like to add an example on how to use neli with custom properties for custom generic netlink families. I use this at an uni project where I develop a kernel module (which registers a custom generic netlink family), a userland component in Rust (using neli), and a other userland component using libnl. Therefore I can ensure that my approach works with the kernel / other netlink implementations.

Let's assume we have a custom kernel driver that registers a "your_family_name_here" family. The subject of our problem knows the operations/commands echo, create device, and delete device. It also knows the attributes msg (string) and device name (string).

In this example (see code) an echo message is send to the kernel and the reply received:

PS: Example kernel module that fulfills this contract: https://github.com/phip1611/generic-netlink-user-kernel-rust

PPS: When I started doing this project (with neli 0.4.3) I spend so much time on this. Therefore I'd love to have an easy example in this great lib so that others can profit.

Let me know what you think :)

phip1611 · 2020-12-05T19:29:43Z

Of course this will not build/terminate in any CI system because the kernel module doesn't exist there. But it will fail, not stuck in a loop or so.

phip1611 · 2020-12-05T23:05:15Z

examples/custom_generic_nl_family_custom_types.rs

+        sock.recv().expect("Should receive a message").unwrap();
+
+    // According to Netlink Linux kernel code nl_type is either 0x2 (error) or our family id (ok)
+    if res.nl_type == 0x2 {


Ah I have to refactor this. Looks like this check is already covered by sock.recv(). It returns Err if nl_type is 0x2

No, neli tells The buffer was not large enough to complete the deserialize operation. Doesn't look what I would neli expect to do. Technically an "well formed error message" is okay but not error

phip1611 · 2020-12-05T23:15:33Z

Probably found a general neli problem that is not related to this specific PR. I just want to note it here:
If I reply from kernel with a netlink error message (nl_type in netlink-header is NLMSG_ERR (0x2)) then ´socket.recv()´ returns ERR because of The buffer was not large enough to complete the deserialize operation. My kernel module added no payload to the response, so no struct nlmsgerr. I don't know if this is right.. this is valid (generic) netlink behaviour. A well formed error message from the kernel should be okay, even if it doesn't contain any payload.. or am I wrong? Especially because struct nlmsgerr comes from "the old netlink times" without generic netlink. I think signaling an error with custom attributes is much better. The lib shoud support that.. but this is subject to another issue/PR.

Netlinks man page also encourages developers to use NLMSG_ERR to signalize error.
.recv() should only return Err if it couldn't parse the buffer. (which is in this case kinda true because neli is probably having a problem that the massage having no payload, but it should be parsed as Ok)

jbaublitz · 2020-12-06T17:07:18Z

@phip1611 Thanks for the report! So just to confirm, you returned an error packet with no nlmsgerr payload? According to the netlink manpage, this does not appear to be valid behavior at this point in time.

NLMSG_ERROR message signals an error and the payload contains an nlmsgerr structure

from here.

Unless you can give me a specific link to documentation that specifies that this payload is optional, I think this is a problem in your implementation, not in mine.

jbaublitz · 2020-12-06T18:24:13Z

I am happy to create an issue to improve error messaging if that would be helpful for you. I've also bumped into some situations where the error messaging is not adequate and I'd like to improve the error provided to the user in certain cases.

phip1611 · 2020-12-06T21:12:33Z

@jbaublitz You are right. I'm not a netlink expert and I just tried to use it in a way that it works for me. My Kernel module didn't include struct nlmsgerr in the payload. I was not following the standard.

But can you imagine a more convenient way to recognize the absence of this struct, when nl_type is NLMSG_ERR? Because the current message "The buffer was not large enough to complete the deserialize operation" is not effective.

PS: I update my kernel module to include the struct in the payload in error case. I edit this comment when I have an update on this.

jbaublitz · 2020-12-07T00:41:57Z

Perfect! I'll open an issue and tag you in it so we can discuss details.

jbaublitz · 2020-12-07T02:15:21Z

I have a few comments on the code so far. I'll review this soon and request some changes just so that people looking at the code are guided towards neli best practices when they look at the example.

jbaublitz

Here are some initial notes.

jbaublitz · 2021-01-04T00:31:58Z

examples/custom_generic_nl_family_custom_types.rs

+use neli::consts::nl::Nlmsg;
+use neli::consts::nl::{NlmF, NlmFFlags};
+use neli::consts::socket::NlFamily;
+use neli::genl::{Genlmsghdr, Nlattr};
+use neli::nl::{NlPayload, Nlmsghdr};
+use neli::socket::NlSocketHandle;
+use neli::types::{Buffer, GenlBuffer};
+use neli::utils::U32Bitmask;
+use neli::Nl;
+use std::process;


Could you switch this to nested imports? I've been meaning to put out some formal coding style guidelines but haven't gotten around to that quite yet.

jbaublitz · 2021-01-04T00:37:11Z

examples/custom_generic_nl_family_custom_types.rs

+        sock.recv().expect("Should receive a message").unwrap();
+
+    // According to Netlink Linux kernel code nl_type is either 0x2 (error) or our family id (ok)
+    if res.nl_type == u16::from(Nlmsg::Error)


neli has special handling for the Nlmsg::Error value as of 0.5.0. This value indicates either an actual error if the error code is non-zero or an ACK if the error code is 0. neli will return these as values of either NlPayload::Ack or NlPayload::Err in the resulting Nlmsghdr. In the error case, recv() will detect this as an error and return it as an Err(Nlmsgerr) value. Can you explain a little bit more why you chose to do the checking manually here? Does this use case not work for you?

Oh, apparently I was using the neli lib wrong in my uni project and therefore also in this demo.
I didn't had a proper example at the first place that fit my use case, therefore I just tried to figure everything out by myself. I'll improve the code and let you know if it works better. @jbaublitz

jbaublitz · 2021-01-04T00:38:31Z

examples/custom_generic_nl_family_custom_types.rs

+    let received = attr_handle.get_attribute(FoobarAttribute::Msg).unwrap();
+    let received = String::deserialize(received.nla_payload.as_ref()).unwrap();


This can be more concisely represented by attr_handle.get_attr_payload_as::<String>(FoobarAttribute::Msg).

phip1611 · 2021-01-29T14:23:30Z

@jbaublitz FYI I'll fix this in the next weeks, I didn't forget it. I'm having a deadline for an uni project this sunday (where I use neli :))

phip1611 · 2021-02-10T13:49:42Z

@jbaublitz Sorry it took a while, I updated the PR. Any more comments on the code? Of course I tested it and it's working.

PS: I don't get why this fails: https://github.com/jbaublitz/neli/pull/103/checks?check_run_id=1871771582

jbaublitz · 2021-02-11T23:53:17Z

I just reran the failed test and realized that this is a new clippy error in my code. I'll fix that.

jbaublitz · 2021-02-12T13:16:36Z

@phip1611 Clippy checks are now optional in CI. If you rebase onto master, you should be good to go.

phip1611 · 2021-02-12T16:14:21Z

@jbaublitz done; pushed the rebased branch

jbaublitz

Once this last change is made, I'll merge this. Thanks for putting so much effort into the documentation!

examples/custom_generic_nl_family_custom_types.rs

phip1611 · 2021-02-14T08:11:39Z

@jbaublitz I fixed it.

phip1611 mentioned this pull request Dec 5, 2020

"the trait neli::consts::genl::Cmd is not implemented" Problem with impl_var/impl_trait for custom types #99

Closed

phip1611 commented Dec 5, 2020

View reviewed changes

jbaublitz mentioned this pull request Dec 7, 2020

Improve error messaging #105

Closed

jbaublitz added the documentation label Dec 7, 2020

jbaublitz assigned jbaublitz and phip1611 and unassigned jbaublitz Dec 7, 2020

jbaublitz self-requested a review December 7, 2020 02:16

jbaublitz requested changes Jan 4, 2021

View reviewed changes

phip1611 mentioned this pull request Jan 4, 2021

Refactor code to fit latest neli best practises phip1611/generic-netlink-user-kernel-rust#3

Closed

jbaublitz mentioned this pull request Feb 12, 2021

Fix clippy #128

Merged

neli example for custom netlink family with custom types

4ba14d7

Typo

805e05b

jbaublitz requested changes Feb 13, 2021

View reviewed changes

examples/custom_generic_nl_family_custom_types.rs Outdated Show resolved Hide resolved

phip1611 added 2 commits February 14, 2021 09:08

Comments

ddc8e0f

Code improvements after review

35355fc

phip1611 mentioned this pull request Feb 22, 2021

Missing CAN bus constants here and libc #140

Closed

phip1611 requested a review from jbaublitz March 4, 2021 11:51

jbaublitz approved these changes Mar 13, 2021

View reviewed changes

jbaublitz merged commit ac03202 into jbaublitz:master Mar 13, 2021

phip1611 deleted the example-custom-types branch March 15, 2021 08:33

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Example for custom generic netlink family with custom attributes #103

Example for custom generic netlink family with custom attributes #103

phip1611 commented Dec 5, 2020 •

edited

Loading

phip1611 commented Dec 5, 2020 •

edited

Loading

phip1611 Dec 5, 2020

phip1611 Dec 5, 2020

phip1611 commented Dec 5, 2020 •

edited

Loading

jbaublitz commented Dec 6, 2020

jbaublitz commented Dec 6, 2020

phip1611 commented Dec 6, 2020 •

edited

Loading

jbaublitz commented Dec 7, 2020

jbaublitz commented Dec 7, 2020

jbaublitz left a comment

jbaublitz Jan 4, 2021

jbaublitz Jan 4, 2021

phip1611 Jan 4, 2021

jbaublitz Jan 4, 2021

phip1611 commented Jan 29, 2021

phip1611 commented Feb 10, 2021 •

edited

Loading

jbaublitz commented Feb 11, 2021

jbaublitz commented Feb 12, 2021

phip1611 commented Feb 12, 2021

jbaublitz left a comment

phip1611 commented Feb 14, 2021

		let received = attr_handle.get_attribute(FoobarAttribute::Msg).unwrap();
		let received = String::deserialize(received.nla_payload.as_ref()).unwrap();

Example for custom generic netlink family with custom attributes #103

Example for custom generic netlink family with custom attributes #103

Conversation

phip1611 commented Dec 5, 2020 • edited Loading

phip1611 commented Dec 5, 2020 • edited Loading

phip1611 Dec 5, 2020

Choose a reason for hiding this comment

phip1611 Dec 5, 2020

Choose a reason for hiding this comment

phip1611 commented Dec 5, 2020 • edited Loading

jbaublitz commented Dec 6, 2020

jbaublitz commented Dec 6, 2020

phip1611 commented Dec 6, 2020 • edited Loading

jbaublitz commented Dec 7, 2020

jbaublitz commented Dec 7, 2020

jbaublitz left a comment

Choose a reason for hiding this comment

jbaublitz Jan 4, 2021

Choose a reason for hiding this comment

jbaublitz Jan 4, 2021

Choose a reason for hiding this comment

phip1611 Jan 4, 2021

Choose a reason for hiding this comment

jbaublitz Jan 4, 2021

Choose a reason for hiding this comment

phip1611 commented Jan 29, 2021

phip1611 commented Feb 10, 2021 • edited Loading

jbaublitz commented Feb 11, 2021

jbaublitz commented Feb 12, 2021

phip1611 commented Feb 12, 2021

jbaublitz left a comment

Choose a reason for hiding this comment

phip1611 commented Feb 14, 2021

phip1611 commented Dec 5, 2020 •

edited

Loading

phip1611 commented Dec 5, 2020 •

edited

Loading

phip1611 commented Dec 5, 2020 •

edited

Loading

phip1611 commented Dec 6, 2020 •

edited

Loading

phip1611 commented Feb 10, 2021 •

edited

Loading