Apple Network Framework Socket Changes #662
Conversation
source/darwin/nw_socket.c
Outdated
| new_nw_socket->connection_setup = true; | ||
|
|
||
| // Setup socket state to start read/write operations. | ||
| s_set_socket_state(new_nw_socket, new_socket, CONNECTED_READ | CONNECTED_WRITE); |
There was a problem hiding this comment.
Unsure if it's necessary but we are not holding the synced data lock at the point this is called.
There was a problem hiding this comment.
Looking into it, We didn't lock here as we are in initializing process, no other process will touch the socket state. Updated the comment to explain it.
source/channel_bootstrap.c
Outdated
| return; | ||
| } | ||
|
|
||
| struct aws_socket *aws_server_bootstrap_new_socket_listener_async( |
There was a problem hiding this comment.
Trivial: Apart from an acquire on the server_connection_args, this is identical to the non async function. Is there any way to combine the two and perform the acquire conditionally? Or is it possible to have the acquire also happen on non-async and release? Seems like a lot of duplicated code.
There was a problem hiding this comment.
Eventually, I removed the aws_server_bootstrap_new_socket_listener_async. Now we have a setup_callback in aws_server_socket_channel_bootstrap_options. If setup_callback is set, it will asynchronously invoke the setup_callback when the listener is ready for use.
|
The main changes since last code review:
|
source/posix/socket.c
Outdated
| @@ -1135,21 +1135,56 @@ static void s_socket_accept_event( | |||
| socket->io_handle.data.fd); | |||
| } | |||
|
|
|||
| struct on_start_accept_result_args { | |||
There was a problem hiding this comment.
Unsure if it's worth de-duping this across posix/windows.
There was a problem hiding this comment.
Moved it to the private header, I'm not sure if that's better either though 🤔
| @@ -847,6 +847,8 @@ static void s_process_connection_result_task(struct aws_task *task, void *arg, e | |||
| struct nw_socket_scheduled_task_args *task_args = arg; | |||
| struct nw_socket *nw_socket = task_args->nw_socket; | |||
|
|
|||
| AWS_LOGF_TRACE(AWS_LS_IO_SOCKET, "id=%p: start to process connection result task.", (void *)nw_socket); | |||
There was a problem hiding this comment.
Connection/listener state changes would be better as debug I think
There was a problem hiding this comment.
Updated the state_change log to debug. For this specific line, since the function is invoking the connection result callback, I think trace would be a good to track for this callback.
| new_socket->remote_endpoint.port = port; | ||
| } | ||
| if (hostname != NULL) { | ||
| size_t address_strlen; |
There was a problem hiding this comment.
We almost never want to use libc memory/string functions directly. Let's just use aws_byte_cursor_from_c_str and make a local aws_byte_buf that refers to the memory and use an appropriate copy function.
source/darwin/nw_socket.c
Outdated
| if (nw_socket->on_accept_started_fn) { | ||
| nw_socket->on_accept_started_fn(aws_socket, crt_error_code, nw_socket->listen_accept_started_user_data); | ||
| } | ||
| s_unlock_base_socket(nw_socket); |
There was a problem hiding this comment.
Is there any chance that something in the accept started callback could invoke something that tries to lock the socket base?
There was a problem hiding this comment.
The base socket is only modified(write) at two places:
(1) socket creation/init : The lock is just inited, and would not hold at this time.
(2) aws_socket_cleanup(): This is where the base socket get nulled out. We only lock the base socket if the function is not on the event loop to avoid dead lock. Use this callback as an example, if we call aws_socket_cleanup() in on_accept_started_fn, as it is on the event loop, we will not lock the base socket here.
Apple Network Framework socket integration
By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.