#!/bin/sh

# Strict script
set -e
set -u

# Global variables
HOST=false
ROUTER=false

# Get options passed by user
getoption () {
    if [ $# -ge 1 ]; then
        FIRST_OPTION="$1"
    else
		usage
		exit 1
    fi
    case "$FIRST_OPTION" in
		host)
			HOST=true
			;;
		router)
            ROUTER=true
            ;;
		help)
            usage
			exit 0
            ;;
	esac
}

usage () {
	# Display usage
	echo "Usage: $0 option"
	echo " option can be:"
	echo " -host: tuning tips for end-host"
	echo " -router: tuning tips for router"
	echo " -help: display this help"
}

sysctl_atleast () {
    # Check the sysctl parameter is at least configured to the value
    # $1: sysctl parameter
    # $2: recommended value
    if [ `sysctl -n $1` -lt $2 ]; then
		echo "$1=\"$2\""
	fi
}

sysctl_equal () {
	# Check the sysctl parameter match (equal) the value
	# $1: sysctl parameter
	# $2: recommended value
	if [ `sysctl -n $1` -ne $2 ]; then
		echo "$1=\"$2\""
	fi
}

if_have_cap () {
	# Check if the interface have capabilities and if it's enabled return true (0)
	# $1: if
	# $2: capability
    ifconfig -m $1 | egrep -q "capabilities.*$2" && return 0 || return 1
}

if_is_cap_enabled () {
	# Check if the interface capability is enabled: return true
	# $1: if
	# $2: capability
	ifconfig -m $1 | egrep -q "options.*$2" && return 0 || return 1
}

tune () {
	# Major function that call the others
	# RO values need to be put in /boot/loader.conf.local
	# RW values need to be put in /etc/sysctl.conf
	echo "#==> Start of /boot/loader.conf.local"
	echo "#TO DO All these values doesn't works here and need to be put in /etc/sysctl.conf"
	net_nic
	($ROUTER) && net_forward
	($ROUTER) && net_inet
	($ROUTER) && cpu_ht
	net_netisr
	# Autombuf tunning avoid the need fo checking them
	# net_mbuf
	echo "#<== End of /boot/loader.conf.local"
	net_netstat
	echo "#==> To be added to /etc/rc.conf"
	#Need to replace this part by a list of modules (ichsmb coretemp aesni watchdog?)
	if grep -q 'Feature.*AESNI' /var/run/dmesg.boot ; then
		#CPU support, check if not allready loaded
		if ! kldstat -n aesni 2>&1 | grep -q aesni; then
			echo "#CPU has AESNI feature, you should load the modules"
			echo 'kld_list="cryptodev aesni"'
		fi
	fi
	for NIC in `ifconfig -l "ether"`; do
		#skip loopback interfaces
        echo $NIC | egrep -q "lo[[:digit:]]" && continue
		# Check only interface with vlan configured on them
		if egrep -q "ifconfig_vlan.*${NIC}" /etc/rc.conf; then
			if if_have_cap ${NIC} VLAN_HWFILTER; then
				if ! if_is_cap_enabled ${NIC} VLAN_HWFILTER; then
					echo "${NIC} configured with vlan and support VLAN_HWFILTER"
					echo " => add the vlanhwfilter option to ifconfig_${NIC} in /etc/rc.conf"
				fi
			fi # if_have_cap
		fi # rc.conf
		# Router should not have LRO/TSO4/TSO6 enabled
		if ($ROUTER); then
			for CAP in LRO TSO4 TSO6; do
				if_is_cap_enabled ${NIC} ${CAP} && \
					echo "A router MUST NOT have ${CAP} enabled on forwarding NIC"
			done
		fi
		# Check for virtualized
		if (${VM_DETECTED}); then
			if if_have_cap ${NIC} POLLING; then
				if ! if_is_cap_enabled ${NIC} POLLING; then
					# http://lists.freebsd.org/pipermail/freebsd-net/2013-May/035626.html
					echo "In VM, ${NIC} can have better behavior if POLLING was enabled"
					echo " => Enable polling in /etc/rc.conf.misc"
				fi
			fi
			# need disable hardware assisted features (need to found reference about this tip)
			for CAP in RXCSUM TXCSUM TSO4 TSO6; do
            	if_is_cap_enabled ${NIC} ${CAP} && \
					echo "${NIC} have ${CAP} enabled: Disable it in a VM"
        	done # end for CAP
		fi
	done

}

cpu_ht () {
	# Check if HyperThreading is enabled
	grep -q "^FreeBSD/SMP.*SMT" /var/run/dmesg.boot && echo "You should disable HyperThreading"
}

net_nic () {
	# NIC drivers tuning
	# em tuning
	if ifconfig -l "ether" | egrep -q 'em[[:digit:]]'; then
		# Need to check if there are not 82542 or 82543 (support 256 max)
		for VALUE in hw.em.txd hw.em.rxd; do
			sysctl_equal ${VALUE} 2048
		done
		# Disabling limit of rx frame
		#sysctl_atleast hw.em.rx_process_limit 500
		sysctl_equal hw.em.rx_process_limit -1
		# Increasing interrupt delay is bad for a router because it add latency
		# http://www.intel.com/design/network/applnots/ap450.htm
		if ! ($ROUTER); then
			for VALUE in hw.em.rx_abs_int_delay hw.em.tx_abs_int_delay; do
            	sysctl_equal ${VALUE} 1024
        	done
			for VALUE in hw.em.rx_int_delay hw.em.tx_int_delay; do
            	sysctl_equal ${VALUE} 512
        	done
		fi
	fi
	# igb tuning
	if ifconfig -l "ether" | egrep -q 'igb[[:digit:]]'; then
		for VALUE in hw.igb.txd hw.igb.rxd; do
            sysctl_equal ${VALUE} 2048
    	done
		# http://wiki.freebsd.org/NetworkPerformanceTuning
		sysctl_atleast hw.igb.max_interrupt_rate 32000
		# Need mbuf be increased if rx_process limit increased?	
		#sysctl_atleast hw.igb.rx_process_limit 500
		# Disabling limit of rx frame
		sysctl_equal hw.igb.rx_process_limit -1
		# Default kern.ipc.nmbclusters is too small for multiple igb card and multicore
		# Need to add a test [ `sysctl -n hw.ncpu` -gt 1 ] and arch (i386 don't like too much)
		[ `sysctl -n hw.ncpu` -ge 4 ] && sysctl_atleast kern.ipc.nmbclusters 262144
    fi
	# ixgbe tuning
	if ifconfig -l "ether" | egrep -q 'ix[[:digit:]]'; then
	# source: README in drivers source
	#	 When using the ixgbe driver with RSS autoconfigured based on the number of
  	# cores (the default setting) and that number is larger than 4, increase the
 	# memory resources allocated for the mbuf pool as follows:
		# Warning: Where to put this valuedepends if drivers is a module or in kernel
		if [ `sysctl -n hw.ncpu` -ge 4 ]; then
			#loader.conf ?
			sysctl_atleast kern.ipc.nmbclusters 262144
			#sysctl.conf ?
 			sysctl_atleast kern.ipc.nmbjumbop 262144
		fi
	fi
	# General
	# default value of send interface queue length
	# If em or igb were tunned, need to change this too
	# http://lists.freebsd.org/pipermail/freebsd-net/2012-July/032712.html	
	sysctl_atleast net.link.ifqmaxlen 10240	

	# cxgbe (Chelsio) tuning
	if ifconfig -l "ether" | egrep -q 'cxl[[:digit:]]'; then
		# source: http://bsdrp.net/documentation/examples/forwarding_performance_lab_of_a_hp_proliant_dl360p_gen8_with_10-gigabit_with_10-gigabit_chelsio_t540-cr
		# If ncpu=8 and drivers is using 8 queue (default), should reduce the NIC queue to 4
		if [  `sysctl -n hw.ncpu` -eq 8 ]; then
			[ `sysctl -n dev.cxl.0.nrxq` -eq 8 ] && echo 'echo hw.cxgbe.nrxq10g="4" >> /boot/loader.conf.local'
			[ `sysctl -n dev.cxl.0.ntxq` -eq 8 ] && echo 'echo hw.cxgbe.ntxq10g="4" >> /boot/loader.conf.local'
		fi
	fi
}

net_netstat () {
	# Analysing netstat -m output and provide advice regarding the result
	echo "TODO: netstat"
}

net_forward () {
	# http://wiki.freebsd.org/NetworkPerformanceTuning
	sysctl_equal net.inet.ip.forwarding 1
	if [ `sysrc -n ipsec_enable` == "YES" ]; then
		sysctl_equal net.inet.ip.fastforwarding 0
		echo "IPSec configuration detected"
		echo "Check if your hardware support one of these crypto accelerator modules:"
		echo " aesni(4), glxsb(4), hifn(4), ipsec(4), padlock(4), safe(4) or ubsec(4)"
	fi
}

net_inet () {
	# http://wiki.freebsd.org/NetworkPerformanceTuning
	for VALUE in net.inet.raw.maxdgram net.inet.raw.recvspace; do
        sysctl_atleast ${VALUE} 16384
    done
	for VALUE in net.inet.ip.redirect; do
        sysctl_equal ${VALUE} 0
    done
}

net_netisr () {
	# http://wiki.freebsd.org/NetworkPerformanceTuning
	sysctl_atleast net.route.netisr_maxqlen 2048
}

net_mbuf () {
	# kern.ipc.nmbclusters must be increased if multiple igb (man igb)
	# Need to calculate (regarding RAM size) this value
	# Does vm.kmem_size should be upgraded too ?
	# igb loads 8 rings, each of 1024 per NIC
	sysctl_atleast kern.ipc.nmbclusters 262144
	echo "On the following output, with all your NIC already configured"
	echo "  check the current and total"
	netstat -m | grep "mbuf clusters in use"
}



timecounter () {
	# Are the timecounter tuning usefull ?
	TIMECOUTER_CHOICE=`sysctl -n sysctl kern.timecounter.choice`
	if echo "${TIMECOUTER_CHOICE}" | grep -q "HPET"	; then
		sysctl -n kern.timecounter.hardware | grep -q HPET || \
			echo "sysctl kern.timecounter.hardware=HPET"
	fi
}
###################
## Main function ##
###################

getoption $*
#system_inventory
# Look for VM environnement
sysctl -n kern.vm_guest | grep -q -v "none" && VM_DETECTED=true || VM_DETECTED=false
tune
echo "Done!"