Support sdbus clients #54

mardy · 2024-03-25T09:39:38Z

This is built on top of #26. I've kept the original commit by @refi64 and added one to address the issue raised during the review.

Whether my commit actually succeeds in addressing that problem is something I'm not sure of, since I couldn't actually reproduce the original issue. But from my first tests it appears that sdbus clients can connect and send methods calls.

Possible fix for #21

bilelmoussaoui

various code style comments

flatpak-proxy.c

sophie-h · 2024-05-07T12:41:41Z

I cannot confirm that the patch fixes the issue for me.

mardy · 2024-05-08T06:47:42Z

I cannot confirm that the patch fixes the issue for me.

Hi Sophie, can you please tell me how to reproduce the issue?

bilelmoussaoui · 2024-05-08T11:07:39Z

I cannot confirm that the patch fixes the issue for me.

Hi Sophie, can you please tell me how to reproduce the issue?

You can use this flatpak https://github.com/zecakeh/zbus-flatpak-portal, you might have to run cargo update.

sophie-h · 2024-05-08T11:54:04Z

You can use this flatpak https://github.com/zecakeh/zbus-flatpak-portal

If you just use cargo run outside a Flatpak it should run properly. Afterwards, you can run it from a flatpak by

flatpak run --filesystem=host --command=./target/debug/zbus-flatpak-portal app.drey.Warp

Where you can use any id of an installed app instead of app.drey.Warp.

This then gives

Creating connection…
thread 'main' panicked at src/main.rs:7:60:
called `Result::unwrap()` on an `Err` value: InputOutput(Custom { kind: BrokenPipe, error: "failed to read from socket" })
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

(Sorry if this is all obvious for you. Just wanted to make sure you can reproduce it.)

bilelmoussaoui · 2024-05-09T15:27:23Z

You can use this flatpak https://github.com/zecakeh/zbus-flatpak-portal

If you just use cargo run outside a Flatpak it should run properly. Afterwards, you can run it from a flatpak by
flatpak run --filesystem=host --command=./target/debug/zbus-flatpak-portal app.drey.Warp
Where you can use any id of an installed app instead of app.drey.Warp.

This then gives
Creating connection…
thread 'main' panicked at src/main.rs:7:60:
called `Result::unwrap()` on an `Err` value: InputOutput(Custom { kind: BrokenPipe, error: "failed to read from socket" })
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
(Sorry if this is all obvious for you. Just wanted to make sure you can reproduce it.)

Note that zbus 4.2.1 includes a hack to workaround this issue, so you would have to make sure it is using zbus 4.2.0

mardy · 2024-05-16T12:56:44Z

Thanks for your help! I added one more commit (I guess it can be squashed with the previous ones, but for now I'll keep it separate as the commit message sheds some light on it). Now I can run the zbus-flatpak-portal program.

I added one more member (sent) to the Buffer structure. It may be possible to do without it, but the code was hard to understand because the pos and size members were being used with a different meaning depending on whether we were reading from or writing to the buffer; and the fact that we pass buffers from one socket to the other increases on this confusion.

I didn't want to make a too invasive change, but if you agree I'd rename the members like this (suggestions welcome!):

size -> capacity ("size" can also be OK, if you insist :-) )
pos -> "length", "used", "data_size"?
sent -> keep as is?

sophie-h

The latest version does seem to fix the problems for me!

@mardy could you maybe address the small review comments that are open such that we can hopefully move forward with this soon? :)

flatpak-proxy.c

sophie-h · 2024-06-22T12:58:42Z

I didn't want to make a too invasive change, but if you agree I'd rename the members like this (suggestions welcome!):

size -> capacity ("size" can also be OK, if you insist :-) )

pos -> "length", "used", "data_size"?

sent -> keep as is?

Maintainers have been very open to making the code clearer. However, maybe you can add these changes as an additional commit to simplify the review?

flatpak-proxy.c

swick · 2024-06-24T15:13:02Z

This contains commits which are broken by themselves. Can you please squash thing properly so that each commit is working?

swick · 2024-06-24T15:18:55Z

flatpak-proxy.c

@@ -298,6 +298,7 @@ struct FlatpakProxyClient
  AuthState     auth_state;


I don't understand what this commit changes. What is being fixed up exactly?

mardy · 2024-06-25T05:28:05Z

This contains commits which are broken by themselves. Can you please squash thing properly so that each commit is working?

By "broken" you mean "not building" or "not fixing the problem"? If it's the former, then I guess the problem is the warning/error mentioned by Sophie (which I will fix); if the latter, then I guess I should just squash all commits into one, because only the last commit fixes the issue for good.

flatpak-proxy.c

alexlarsson · 2024-08-16T13:26:46Z

I pushed some lightly tested fixes for the issues I found. Please have a look at these.
Otherwise this looks good to me.

I'd like to not have any individual commits in the tree where things are broken, so that bisect works, so most commits here should imho be squashed into one commit. However, I made them separate for easier review.

flatpak-proxy.c

swick · 2024-08-19T10:52:39Z

The changes look fine to me.

sd-bus sends the BEGIN command in at the same time as the AUTH command, assuming that the authentication will succeed. This ends up confusing xdg-dbus-proxy, as it assumes that BEGIN is not sent at the same time as any other messages. As a result, the server's authentication replies end up interpreted as standard D-Bus messages, failing parsing and resulting in the client being unable to connect. This changes the code to keep track of the number of authentication requests and the number ofresponses from the server to know when the authentication phase of the connection has actually completed. Until the authentication is completed, hold off all client messages received after BEGIN. We only send to the server the authentication commands up (and including) the BEGIN command, but everything past the line terminators is queued into the `extra_input_data` buffer and we stop reading from the client socket. Once the authentication is completed, the (partial) message we saved into `extra_input_data` is queued for the D-Bus server into the first outgoing buffer, and reading from the client socket resumes. Note that in order to cleanly process the partial sendind of data, another offset is introduced into the Buffer structure, which holds the bytes of buffers which have already been sent over the socket. Fixes flatpak#21 (finally!) This is based on initial work by: * Ryan Gonzalez <ryan.gonzalez@collabora.com> * Alberto Mardegan <a.mardegan@omp.ru>

This way we can avoid multiple reads in the pipelined case.

We indent one tab less by moving the check for buffer->pos == 0 to a separate check.

alexlarsson · 2024-08-20T10:21:22Z

I found one more issue in testing, but now it seems to work, so I squashed the functional changes into one commit (with props to ryan and alberto).

I think this is fine to go, but please take a second look and test it.

swick · 2024-08-20T13:02:43Z

Hard to see what changed because of the indentation changes and rebase.

alexlarsson · 2024-08-20T13:17:18Z

The only real change was to the handling of auth messages. We now delay the setting of AUTH_COMPLETE to after calling got_buffer_from_side() as that would otherwise try to parse the "BEGIN" auth line as a dbus message. See the new_auth_state variable.

But, it wouldn't be a bad idea to do a more complete review of the combined change, to ensure fresh eyes on this.

swick · 2024-08-20T14:50:46Z

I've taken another look and didn't find anything that stood out.

@mardy can you also please take another look?

mardy · 2024-08-20T16:17:42Z

@mardy can you also please take another look?

Will do tomorrow :-)

mardy · 2024-08-21T08:17:27Z

@mardy can you also please take another look?

Will do tomorrow :-)

I confirm that this continues to work for me :-)

alexlarsson · 2024-08-21T09:42:00Z

merging then

bilelmoussaoui reviewed May 7, 2024

View reviewed changes

flatpak-proxy.c Outdated Show resolved Hide resolved

flatpak-proxy.c Outdated Show resolved Hide resolved

flatpak-proxy.c Outdated Show resolved Hide resolved

mardy force-pushed the sdbus-auth branch from 5064e57 to 0e63504 Compare May 8, 2024 06:31

sophie-h suggested changes Jun 22, 2024

View reviewed changes

flatpak-proxy.c Outdated Show resolved Hide resolved

swick reviewed Jun 24, 2024

View reviewed changes

flatpak-proxy.c Outdated Show resolved Hide resolved

swick reviewed Jun 24, 2024

View reviewed changes

flatpak-proxy.c Outdated Show resolved Hide resolved

swick reviewed Jun 24, 2024

View reviewed changes

flatpak-proxy.c Outdated Show resolved Hide resolved

swick reviewed Jun 24, 2024

View reviewed changes

mardy force-pushed the sdbus-auth branch from 02db72e to 48855d1 Compare June 25, 2024 05:41

alexlarsson reviewed Aug 16, 2024

View reviewed changes

flatpak-proxy.c Outdated Show resolved Hide resolved

flatpak-proxy.c Outdated Show resolved Hide resolved

flatpak-proxy.c Outdated Show resolved Hide resolved

flatpak-proxy.c Outdated Show resolved Hide resolved

swick reviewed Aug 19, 2024

View reviewed changes

flatpak-proxy.c Show resolved Hide resolved

refi64 and others added 3 commits August 20, 2024 12:19

Now that we support pipelined auth, make default auth buffer larger

65cde98

This way we can avoid multiple reads in the pipelined case.

Minor code cleanup of side_in_cb()

5073e0c

We indent one tab less by moving the check for buffer->pos == 0 to a separate check.

alexlarsson force-pushed the sdbus-auth branch from ddb0ed8 to 5073e0c Compare August 20, 2024 10:19

alexlarsson merged commit 32982fc into flatpak:main Aug 21, 2024
5 checks passed

This was referenced Aug 21, 2024

Add support for pipelined SASL handshakes #26

Closed

auth: Support sd-bus authentication pipelining #48

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Support sdbus clients #54

Support sdbus clients #54

mardy commented Mar 25, 2024 •

edited

Loading

bilelmoussaoui left a comment

sophie-h commented May 7, 2024

mardy commented May 8, 2024 •

edited

Loading

bilelmoussaoui commented May 8, 2024

sophie-h commented May 8, 2024

bilelmoussaoui commented May 9, 2024

mardy commented May 16, 2024 •

edited

Loading

sophie-h left a comment

sophie-h commented Jun 22, 2024 •

edited

Loading

swick commented Jun 24, 2024

swick Jun 24, 2024

mardy commented Jun 25, 2024

alexlarsson commented Aug 16, 2024

swick commented Aug 19, 2024

alexlarsson commented Aug 20, 2024

swick commented Aug 20, 2024

alexlarsson commented Aug 20, 2024

swick commented Aug 20, 2024

mardy commented Aug 20, 2024

mardy commented Aug 21, 2024

alexlarsson commented Aug 21, 2024

		@@ -298,6 +298,7 @@ struct FlatpakProxyClient
		AuthState auth_state;

Support sdbus clients #54

Support sdbus clients #54

Conversation

mardy commented Mar 25, 2024 • edited Loading

bilelmoussaoui left a comment

Choose a reason for hiding this comment

sophie-h commented May 7, 2024

mardy commented May 8, 2024 • edited Loading

bilelmoussaoui commented May 8, 2024

sophie-h commented May 8, 2024

bilelmoussaoui commented May 9, 2024

mardy commented May 16, 2024 • edited Loading

sophie-h left a comment

Choose a reason for hiding this comment

sophie-h commented Jun 22, 2024 • edited Loading

swick commented Jun 24, 2024

swick Jun 24, 2024

Choose a reason for hiding this comment

mardy commented Jun 25, 2024

alexlarsson commented Aug 16, 2024

swick commented Aug 19, 2024

alexlarsson commented Aug 20, 2024

swick commented Aug 20, 2024

alexlarsson commented Aug 20, 2024

swick commented Aug 20, 2024

mardy commented Aug 20, 2024

mardy commented Aug 21, 2024

alexlarsson commented Aug 21, 2024

mardy commented Mar 25, 2024 •

edited

Loading

mardy commented May 8, 2024 •

edited

Loading

mardy commented May 16, 2024 •

edited

Loading

sophie-h commented Jun 22, 2024 •

edited

Loading