Add `shutdown_socket()` function for follow up of #505 #506

Jakio815 · 2024-12-20T22:54:05Z

This PR follow ups #505

New Integrated Function: `shutdown_socket()`

int shutdown_socket(int* socket, bool read_before_closing);

Arguments

socket: A pointer to the socket descriptor that needs to be shut down and closed.
read_before_closing: A boolean indicating whether the socket should read any remaining incoming data before closing:
- true: Initiates a graceful shutdown by sending a FIN packet (SHUT_WR) and waits for EOF (0-length message) from the peer.
- false: Immediately shuts down both reading and writing directions (SHUT_RDWR).

Return Values

0: Indicates successful shutdown and closure of the socket.
-1: Indicates a failure during shutdown or closure, with errno set to describe the specific error.

Function Description
This function gracefully shuts down and closes a socket.

When read_before_closing is false:
- The function calls shutdown(SHUT_RDWR) to immediately stop both sending and receiving data. Then calls close().
When read_before_closing is true:
- The function calls shutdown(SHUT_WR) to signal the end of writing but allows reading to continue.
- It waits for the peer to send an EOF (indicated by read() returning 0), and discards all received bytes.

Refactoring on `close_inbound_socket()` and `close_outbound_socket()` from `federate.c`

`close_inbound_socket()`

The original code looked not updated. There were no use case when the argument flag was not 1. I removed the flag argument, and all unused code.

`close_outbound_socket()`

The original flags were only set by this one line code.
int flag = _lf_normal_termination ? 1 : -1;
So it depends on _lf_normal_termination. However, this function can directly use _lf_normal_termination, so I removed the flag, and directly used _lf_normal_termination.

… close_inbound_socket()

Jakio815 · 2024-12-21T01:17:31Z

core/federated/network/socket_common.c

@@ -155,7 +155,7 @@ static int create_server(uint16_t port, int* final_socket, uint16_t* final_port,
    return -1;
  }
  set_socket_timeout_option(socket_descriptor, &timeout_time);
-  int used_port = set_socket_bind_option(socket_descriptor, port, increment_port_on_retry);
+  uint16_t used_port = set_socket_bind_option(socket_descriptor, port, increment_port_on_retry);


Minor fix on port's data type.

Jakio815 · 2024-12-21T01:18:13Z

core/federated/federate.c

@@ -1943,9 +1918,11 @@ void lf_connect_to_rti(const char* hostname, int port) {

 void lf_create_server(int specified_port) {
  assert(specified_port <= UINT16_MAX && specified_port >= 0);
-  if (create_TCP_server(specified_port, &_fed.server_socket, (uint16_t*)&_fed.server_port, false)) {
+  uint16_t port;
+  if (create_TCP_server(specified_port, &_fed.server_socket, &port, false)) {


Minor fix on port's data type.

hokeun

Very nice refactoring. Just left minor suggestions.

hokeun · 2024-12-21T03:14:03Z

core/federated/RTI/rti_remote.c

@@ -1030,9 +1009,7 @@ void send_reject(int* socket_id, unsigned char error_code) {
    lf_print_warning("RTI failed to write MSG_TYPE_REJECT message on the socket.");
  }
  // Close the socket.


Minor: how about adding more information for this call? For example, "Close the socket without reading until EOF."

Thank you, I fixed it as suggested.

hokeun · 2024-12-21T03:14:15Z

core/federated/RTI/rti_remote.c

@@ -1418,9 +1395,7 @@ void lf_connect_to_federates(int socket_descriptor) {
      if (!authenticate_federate(&socket_id)) {
        lf_print_warning("RTI failed to authenticate the incoming federate.");
        // Close the socket.


Thank you, I fixed it as suggested.

hokeun · 2024-12-21T03:14:34Z

core/federated/RTI/rti_remote.c

@@ -1488,8 +1463,7 @@ void* respond_to_erroneous_connections(void* nothing) {
      lf_print_warning("RTI failed to write FEDERATION_ID_DOES_NOT_MATCH to erroneous incoming connection.");
    }
    // Close the socket.


Here as well.

Thank you, I fixed it as suggested.

Jakio815 · 2024-12-22T03:00:28Z

The macOS test is not passing for some unknown reason.

The CI test log seems to not work starting from Running: src/federated/DistributedCountPhysical.lf (28%), however if you run the federated tests on the local machine, it actually doesn't work starting from Dataflow.lf.

The reason for not passing is that the program blocks on the final shutdown phase.

First, this is how the LF code is intended to work.

The federate sends a RESIGN signal.
The RTI calls shutdown(socket, SHUT_WR), and then immediately calls read()
- The shutdown() sends a FIN packet to the federate.
The federate was blocked on read() to listen messages from the RTI.
- The federate receives the FIN packet, and finishes reading the data in the socket buffer, and then returns EOF.
The federate will also send a FIN packet and the RTI's read() will return 0.

This normally works on Linux, thus passing all tests. However, this fails on Mac.
Where it fails is exactly in step 3. The federate received a FIN packet, and the read() should return EOF, however it does not, and gets blocked forever.

I also checked that the FIN packet was properly sent, and the ACK from the federate to the RTI was also sent.

// netstat -ant | grep 15045
tcp4       0      0  127.0.0.1.15045        127.0.0.1.64842        FIN_WAIT_2
tcp4       0      0  127.0.0.1.64842        127.0.0.1.15045        CLOSE_WAIT

The RTI(15045) is in FIN_WAIT_2 state, which means that it sent the FIN, and received the ACK.
The federate(64842) is in CLOSE_WAIT which means that the remote side's(RTI) connection is closed, and waiting to connect its own connection.

This error can be reproduced by the lingua-franca's repo on branch refactor-only-comm-type and reactor-c branch on shutdown, and it should be done on a MacOS machine.

@hokeun @edwardalee Could anyone help investigating this problem?

Jakio815 added 5 commits December 20, 2024 15:52

Add shutdown_socket()

0af50a5

Use shutdown_socket() in rti_remote.c and federate.c

e5129c5

Fix comments.

1b00ed3

Fix port type && change to print_log on shutdown failures && Refactor…

de2fbde

… close_inbound_socket()

Refactor close_outbound_socket().

9ea9f1d

Jakio815 added the refactoring label Dec 21, 2024

Jakio815 mentioned this pull request Dec 21, 2024

Refactoring of socket related functions to a separate socket_common.c #505

Merged

Jakio815 commented Dec 21, 2024

View reviewed changes

Merge branch 'refactor-only-comm-type' into shutdown

d9a67a2

Jakio815 marked this pull request as ready for review December 21, 2024 01:18

Jakio815 changed the title ~~Draft: Add shutdown_socket() function.~~ Add shutdown_socket() function. Dec 21, 2024

Jakio815 changed the title ~~Add shutdown_socket() function.~~ Add shutdown_socket() function for follow up of #505 Dec 21, 2024

Add comment to shutdown_socket function.

996896c

Jakio815 requested review from hokeun and edwardalee December 21, 2024 01:24

hokeun approved these changes Dec 21, 2024

View reviewed changes

Jakio815 added 6 commits December 20, 2024 20:56

Add commnets.

e3b0ea1

Properly replace close() to shutdown_socket()

30a9446

Minor fix.

276d7b4

Minor fix.

93adf2e

Properly close federate's socket connected to the RTI.

d27c743

Minor fix.

ef3345b

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add `shutdown_socket()` function for follow up of #505 #506

Add `shutdown_socket()` function for follow up of #505 #506

Jakio815 commented Dec 20, 2024 •

edited

Loading

Jakio815 Dec 21, 2024

Jakio815 Dec 21, 2024

hokeun left a comment

hokeun Dec 21, 2024

Jakio815 Dec 21, 2024

hokeun Dec 21, 2024

Jakio815 Dec 21, 2024

hokeun Dec 21, 2024

Jakio815 Dec 21, 2024

Jakio815 commented Dec 22, 2024 •

edited

Loading

Add shutdown_socket() function for follow up of #505 #506

Are you sure you want to change the base?

Add shutdown_socket() function for follow up of #505 #506

Conversation

Jakio815 commented Dec 20, 2024 • edited Loading

New Integrated Function: shutdown_socket()

Refactoring on close_inbound_socket() and close_outbound_socket() from federate.c

close_inbound_socket()

close_outbound_socket()

Choose a reason for hiding this comment

Choose a reason for hiding this comment

hokeun left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Jakio815 commented Dec 22, 2024 • edited Loading

Add `shutdown_socket()` function for follow up of #505 #506

Add `shutdown_socket()` function for follow up of #505 #506

Jakio815 commented Dec 20, 2024 •

edited

Loading

New Integrated Function: `shutdown_socket()`

Refactoring on `close_inbound_socket()` and `close_outbound_socket()` from `federate.c`

`close_inbound_socket()`

`close_outbound_socket()`

Jakio815 commented Dec 22, 2024 •

edited

Loading