What were you trying to do that didn't work?

When a ssh connection is performed using multiplexing (ControlPath, ControlMaster=auto, ...) and the master connection timed out a few milliseconds before mux_client_hello_exchange() is called, this leads to having ssh die in SIGPIPE, which is a bug.
Indeed, it's a bug because multiplexing is supposed to have a fallback as noted in the ssh_config(5) manpage:

     ControlMaster

             [...] These sessions will try to reuse the master instance's network connection rather than initiating new ones, but will fall back to connecting normally if the control socket does not exist, or is not listening.

The reason for ssh dying with SIGPIPE is at the time mux_client_hello_exchange() and underlying mux_client_write_packet() executes, the SIGPIPE signal is not yet ignored, causing line 1513 to raise the signal and kill ssh:

1488 static int
1489 mux_client_write_packet(int fd, struct sshbuf *m)
1490 {
 :
1513                 len = write(fd, ptr + have, need - have);
 :

A quick fix is to ignore the SIGPIPE while doing the hello:

Original code:

1575 static int
1576 mux_client_hello_exchange(int fd)
1577 {
 :
1589         if (mux_client_write_packet(fd, m) != 0) {
1590                 debug_f("write packet: %s", strerror(errno));
1591                 goto out;
1592         }
 :

Modified code:

1575 static int      
1576 mux_client_hello_exchange(int fd)
1577 {       
 :
1581         sshsig_t old_sigpipe;
 :
1590         old_sigpipe = ssh_signal(SIGPIPE, SIG_IGN);
1591         r = mux_client_write_packet(fd, m);
1592         ssh_signal(SIGPIPE, old_sigpipe);
1593         if (r != 0) {
1594                 debug_f("write packet: %s", strerror(errno));
1595                 goto out;
1596         }
 :

EDIT: Upstream fixed that through 96faa0de6c673a2ce84736eba37fc9fb723d9e5c.

Please provide the package NVR for which bug is seen:

openssh-clients-8.7p1-38.el9
openssh-clients-8.0p1-19.el8_9.2

How reproducible:

Often using a quickly closing connection

Steps to reproduce

1. Execute the following command in loop

   $ while :; do echo; date +%s.%N; /usr/bin/ssh -o ControlPath=/tmp/%r@%h:%p -o ControlPersist=2 -o ControlMaster=auto localhost hostname || { echo Exited with: $? ; break ; }; sleep 1.6s ; done

   The sleep delay may be adjusted, depending on the hardware, the idea being the new connection will happen just while master connection times out

Expected results

Connection to the system never failing in exit code 141.

Actual results

1716372217.124957684
vm-ssh9

1716372219.174926179
muxclient: master hello exchange failed
vm-ssh9

1716372221.187231796
muxclient: master hello exchange failed
vm-ssh9

1716372223.184413163
muxclient: master hello exchange failed
vm-ssh9

1716372225.178581753
Exited with: 141

Other reproducer using a systemtap script

1. In a terminal start the following script

   # stap -g -v -e 'probe process("/usr/bin/ssh").statement("*@mux.c:1513") { raise(%{SIGPIPE%}); exit() }'

2. In another terminal execute the following command in loop

   $ while :; do echo; date +%s.%N; /usr/bin/ssh -o ControlPath=/tmp/%r@%h:%p -o ControlPersist=20 -o ControlMaster=auto localhost hostname || { echo Exited with: $? ; break ; }; sleep 1.6s ; done

   Here above ControlPersist can be larger because we will inject a SIGPIPE directly into the write() location, which shows SIGPIPE is not ignored at the time of the call.