Fix race between network.service and NIC driver probe (issue #556)

etc/systemd/system/network.service

@@ -8,6 +8,10 @@ Wants=network.target
 EnvironmentFile=/etc/local/mxhost.conf
 Type=oneshot
 RemainAfterExit=yes
+# The NIC driver may still be probing when we get here (short userspace
+# startup after kexec outruns e.g. the ~1 s ixgbe probe, issue #556), so
+# wait for the device before renaming and configuring it.
+ExecStartPre=/usr/sbin/mxnetctl wait ${MX_NETDEV}
 ExecStart=/usr/sbin/mxnetctl start
 ExecStart=/sbin/ip addr add ${MX_IPADDR}/20 broadcast 141.14.31.255 dev ${MX_NETDEV}
 ExecStart=/sbin/ip link set up dev ${MX_NETDEV}

mxnetctl/mxnetctl

@@ -4,21 +4,24 @@ use warnings;
 
 use Getopt::Long;
 
-our ($opt_quiet,$opt_noop,$opt_ignore_hw);
+our ($opt_quiet,$opt_noop,$opt_ignore_hw,$opt_timeout);
 use constant OPTIONS => (
 	'quiet'     => \$opt_quiet,
 	'noop'      => \$opt_noop,
+	'timeout=i' => \$opt_timeout,
 );
 
 sub USAGE {
 	return <<"__EOF__";
 usage: $0
   start   [options]
   stop    [options] # (ignored)
+  wait    [options] DEVICE
 
 options:
   --quiet      : do not log actions
   --noop       : dry run
+  --timeout=S  : seconds to wait for DEVICE (wait only, default 30)
 
 __EOF__
 }
@@ -175,7 +178,26 @@ sub start {
 	}
 }
 
+sub wait_for_device {	# 'net04' -> 1 if device (or its configured hardware) is present, 0 on timeout
+	my ($dev)=@_;
+	read_mxnet();
+	my $hw=$DEV_TO_HW{$dev};
+	my $deadline=time+$opt_timeout;
+	while (1) {
+		-d "/sys/class/net/$dev" and return 1;
+		if (defined $hw) {
+			for my $d (network_hardware_devices()) {
+				my $addr=eval { get_hw_address($d) } // next;
+				lc($addr) eq lc($hw) and return 1;
+			}
+		}
+		time>=$deadline and return 0;
+		select undef,undef,undef,0.1;
+	}
+}
+
 GetOptions(OPTIONS) or die USAGE;
+$opt_timeout //= 30;	# default documented in USAGE
 
 @ARGV>=1 or die USAGE;
 
@@ -185,6 +207,13 @@ if ($cmd eq 'start') {
 	start();
 } elsif ($cmd eq 'stop') {
 	;
+} elsif ($cmd eq 'wait') {
+	@ARGV>=2 or die USAGE;
+	my $dev=$ARGV[1];
+	unless (wait_for_device($dev)) {
+		warn "timeout waiting for network device $dev after ${opt_timeout}s\n";
+		exit 1;
+	}
 } else {
 	die USAGE;
 }