Ubiquiti NanoBeam/XW compliance test

#!/bin/sh
# vim:sw=8:ts=8:ai:foldmethod=indent
# Compliance Test scanner tabajara.
# You should be using your country's channels only, but who is me to judge...
VERSION="2014.12.26.01"
# Tested and confirmed working on:
# NANOBEAM M5 19db with FW version(s):
# XW.v5.5.9
# XW.v5.5.10

# Ubiquiti AirOS NanoBeam/NanoStation/Rocket XW compliance test kludge script
# Please report other successful deployments in forums and share your improvements!
# If I can merge it back, please say so in your posts. Thanks ':)

cat <<EOF >/dev/null
******************************** READ ME FIRST ********************************
MODE OF OPERATION:
- Can be run manually for one scan, with no bandwidth change.
- Designed to be run from rc.poststart, with "auto" argument. Will background
  itself.
- Refuses to run two or more instances (checks for a lock file)
- Refuses to run in AP mode (only managed/"station")
- Refuses to run if channel list is enabled (if you have a list, you already
  did the hard work yourself!)
- Writes to the system logger and /tmp/scriptname.log (only to console when in
  manual mode) with all it is doing. Keep the system log enabled!
- Bandwitdh hopping: Read below!

THEORY OF OPERATION:
- Because of the quantity of available channels on compliance test mode, the
  station will take an eternity plus a day to link to some channels, and
  sometimes will not detect the AP on the site survey.
- If the radio is associated while the script is running, it just sleeps.
- If not associated, the script will limit the quantity of channels to be
  scanned, and submit a survey, pausing a little, then checking if it is
  associated. Rinse and repeat until no more channels are to be tested.
- If bandwidth hopping is disabled, it will just start from scratch trying all
  channels again. To enable hopping, see below.

BANDWIDTH HOPPING BEHAVIOR:
- After trying all channels on current bandwidth mode, if it STILL is not
  associated, it will:
  * Note original bandwidth and count of retries it can make
  * Change to another bandwidth, reboot, scan
  * When every bandwidth is tested, it adds to the counter of retries
  * When retries count is reached, reverts to original bandwidth and wait.
  * If the association is successful, reset counter and wait.
- This is managed by the "startup date" option. If it is UNCHECKED, the script
  will NEVER try to change BW mode. If CHECKED, the numeric DAY of the startup
  date is the count of tries the script will use to try scanning all the
  bandwidths. If it is 1, the script will stop trying new bandwidths after one
  try on all the ones. I would recommend setting it to 2 or 3 at most -- if it
  cannot do it in three times, the AP is dead.
- Yeah, you only have 31 tries at most, but it is better than rebooting ad
  infinitum if the AP is down. You can always ask someone to unplug/plug the
  power and it will start over again from where it stopped.
- We save every change. If you reboot the radio while it is not associated yet,
  it will continue scanning where it left off. Mind this!

PARAMETERS:
Set "bw_all" to the ordered list of your most used bandwidths.
Experiment with "try_count" and "try_sleep" variables, on a radio you can
locally SSH into and kill the script. In my experience too much "try_count" and
too little "try_sleep" IS BAD.

EOF

# CHANGELOG:
# - 2014.12.26.01
# + Fixes:
#   Script name was hardcoded on install() - using basename $0 now.
# - 2014.11.15.03
# + Fixes:
#   Randomization of undetected channels being re-sorted before use
#   Wait in auto mode until system is up, so can write logs
# + Cosmetic:
#   Split logged/printed channel list to at most 15 channels/line
#   Print board name and firmware version at startup
#   Print running parameters when unassociated
# + Improvements:
#   Parameters f_log, f_config, max_log_lines added
#   Truncate f_log to $max_log_lines to save memory
#   Will read file $f_config for parameters on each loop - this means that
#     you can now modify the parameters at runtime!
#   Removing $f_lock file (rm /var/lock/CTscan.lock) makes the script exit.
#
# - 2014.11.03.32
# + First internal release

### PARAMETERS ###
# List of bandwidth ranges - change ranges in this order after trying the
# current bandwidth and not associating.
# Notice the trailing space for consistency. Keep it.
# You could add 20MHz too but it is redundant as 20/40 does it.
bw_all="20/40 30 10 5 25 "
# Channel count to be locked on each try - will build a random list of N channels
try_count=50
# Wait time (seconds) for each channel list try.
try_sleep=5
# Wait time (seconds) on the main loop - the "are we associated yet?" scan
loop_sleep=5
# Max log lines to keep on $f_logfile
max_log_lines=100
### END OF PARAMETERS ###

# Files used (other than tmp ones)
f_lock=/var/lock/$(basename $0).lock # refuse to run while this file is in place.
f_persist=$0.persist # here we save our progress between reboots
f_syscfg=/tmp/system.cfg # when changing bandwidth, will write this
f_poststart=/etc/persistent/rc.poststart
f_log=/tmp/$(basename $0).log
f_config=$0.config

# Bandwidth variables.
bw_20_40="0 1 1 11naht40";
bw_30="30 1 0 11naht20";
bw_25="25 1 0 11naht20"
bw_20="0 1 0 11naht20";
bw_10="0 2 0 11naht20";
bw_5="0 4 0 11naht20"
getbw() { # get configured bandwidth in MHz, from the config file
    chanbw=$(sed -n 's/^radio\.1\.chanbw=//p' $f_syscfg)
    clksel=$(sed -n 's/^radio\.1\.clksel=//p' $f_syscfg)
    cwmmode=$(sed -n 's/^radio\.1\.cwm\.mode=//p' $f_syscfg)
    ieee_mode=$(sed -n 's/^radio\.1\.ieee_mode=//p' $f_syscfg)
    bw="$chanbw $clksel $cwmmode $ieee_mode"
    [ "$bw" = "$bw_20_40" ] && echo 20/40 && return 0
    [ "$bw" = "$bw_30" ] && echo 30 && return 0
    [ "$bw" = "$bw_25" ] && echo 25 && return 0
    [ "$bw" = "$bw_20" ] && echo 20 && return 0
    [ "$bw" = "$bw_10" ] && echo 10 && return 0
    [ "$bw" = "$bw_5"  ] && echo 5  && return 0
}
setbw() { # set bandwith on config file
    if   [ "$1" = "20/40" ]; then bw=$bw_20_40
    elif [ "$1" = "30" ]; then bw=$bw_30
    elif [ "$1" = "25" ]; then bw=$bw_25
    elif [ "$1" = "20" ]; then bw=$bw_20
    elif [ "$1" = "10" ]; then bw=$bw_10
    elif [ "$1" = "5" ];  then bw=$bw_5
    else return 1
    fi
    set - $bw
    sed -i "
    s/^\(radio\.1\.chanbw\)=.*$/\1=$1/
    s/^\(radio\.1\.clksel\)=.*$/\1=$2/
    s/^\(radio\.1\.cwm.mode\)=.*$/\1=$3/
    s/^\(radio\.1\.ieee_mode\)=.*$/\1=$4/
    " $f_syscfg
}

# Lines which the script will add/remove from poststart
enableCTkey="### CTSCAN-enable"
enableCTstr="echo '<option value=\"511\">Compliance Test</option>' >/etc/ccodes.inc $enableCTkey"
enableSCkey="### CTSCAN-start"
enableSCstr="/bin/sh /etc/persistent/$(basename $0) auto $enableSCkey"
poststartCreate() { # creates poststart if not found
    if ! [ -f "$f_poststart" ]; then
        echo '#!/bin/sh' > $f_poststart
        chmod +x $f_poststart
    fi
}
install() { # add autostart
    poststartCreate
    sed -i "/$enableSCkey/d; $ a\
$enableSCstr" $f_poststart
    echo "CTscan added to autostart, save and reboot to apply."
}
uninstall() { # remove autostart
    sed -i "/$enableSCkey/d" $f_poststart
    echo "CTscan removed from autostart, save and reboot to apply."
}
enableCT() { # enable compliance test.
    poststartCreate
    sed -i '/^radio\.\(1\.\)\?countrycode/s/=.*/=511/' $f_syscfg
    sed -i "/$enableCTkey/d; $ a\
$enableCTstr" $f_poststart
    echo "Compliance test enabled, save and reboot to apply."
}
disableCT() { # disable compliance test.
    sed -i '/^radio\.\(1\.\)\?countrycode/s/=.*/=840/' $f_syscfg
    sed -i "/$enableCTkey/d" $f_poststart
    echo "Compliance test disabled, save and reboot to apply."
    echo "NOTE: You will then be required to set your country on the web interface!"
}

getstartupday() { # get date enabled, and if so, try count
    if fgrep -q 'system.date.status=enabled' $f_syscfg; then
        day=$(sed -n 's/^system\.date\.timestamp=..\([0-9]\{2\}\).*$/\1/p' $f_syscfg)
        echo $((day)) # remove leading zero if exists
        return 0
    else
        echo 0
        return 1
    fi
}
checkassoc() {
    # returns true if associated (mind the starting !)
    ! iwconfig ath0 | grep -q 'Access Point: Not-Associated'
    return $?
}
grepchan() {
    # Parses "iwlist scanning" command and returns only the channels found,
    # sorted, unique; removes channels already tried (kept in $f_fail temp file)
    sed -n "
    # Look for ESSID
    # XXX FIXME I hate putting variables inside a script... This is a point of
    # failure, if YOUR OWN SSID have some special character.
    /ESSID:$1/{
        # Join two lines down that
        N; N;
        # Catch only the frequency
        s/.*Frequency:\([0-9.]\+\) GHz.*/\1/;
        # Format it in MHz
        s/\.//; s/$/000/; s/^\([0-9]\{4\}\).*/\1/;
        # Print
        p;
    }" $f_scan | sort -n | uniq | grep -v -f $f_fail
}
scan() { # Main script loop. Will check, find, change BW
    if [ "$p" = "echolog" ]; then # Wait the system do fs check, mount
        while true; do
            sleep 2
            pgrep syscheck >&- || break
        done
    fi
    $p "CTscan version $VERSION: starting up."
    $p "Firmware version $(cat /etc/version) on a $(sed -n '/^board.name=/s/.*=//p' /etc/board.info)"
    bw_changes_stop=0 # when we already tried it all and gave up change bw, this will be 1
    bw_curr=$(getbw)
    while true; do
        # Reload config
        [ -f "$f_config" ] && source $f_config
        checkdisabled
        [ "$p" = "echolog" ] && sleep $loop_sleep # Early sleep is by design
        if ! checkassoc; then
            # Keeps the log file small
            cur_log_lines=$(wc -l <$f_log)
            [ "$cur_log_lines" -gt "$max_log_lines" ] && sed -i 1,$((cur_log_lines-max_log_lines))d $f_log
            # Find!
            $p 'Trying to find AP by locking onto some channels. Current parameters:'
            $p "bw_curr=$bw_curr bw_all='$bw_all' try_count=$try_count try_sleep=$try_sleep loop_sleep=$loop_sleep"
            if findchan; then
                # Success, reset/remove session and quiet down.
                bw_changes_stop=0
                [ -f "$f_persist" ] && rm $f_persist && savecfg
                [ "$p" = "echo" ] && return 0
                continue
            fi
            [ "$p" = "echo" ] && return 1 # Manual mode ends here
            if [ "$bw_changes_stop" = "1" ]; then
                # We have given up, keep in this bw until someone reboots
                continue
            fi
            bw_changes_allowed=$(getstartupday)
            if [ "$bw_changes_allowed" = "0" ]; then
                # bw change is disabled (startup date not enabled!), respect it and keep bw
                bw_changes_stop=1
                [ -f "$f_persist" ] && rm $f_persist && savecfg
                continue
            fi
            # bw change enabled, look for active session (or create one)
            getsession
            if [ "$bw_changes_stop" = "1" ]; then
                # We already tried looping ($bw_changes_allowed times) and failed in all.
                # The session file is here just to hint me to give up. So I will.
                # We should be on the last bandwidth set that worked, keep on it until reboot
                [ -f "$f_persist" ] && rm $f_persist && savecfg
                continue
            fi
            checkdisabled
            # DOWN HERE IS THE DESPAIR... REBOOTS
            if [ -n "$bw_new" ]; then
                # New bw waiting to try, do it
                sed -i "\,^bw_new=$bw_new$,d" $f_persist
                $p 'Rebooting on new bandwidth:' $bw_new 'MHz' # Shouldn't have time to read this, but...
                setbw $bw_new; savecfg; restart
            else
                # THE END OF THE INTERNET
                $p 'Raising retry counter'
                bw_try_count=$((bw_try_count+1))
                if [ "$bw_try_count" -lt "$bw_changes_allowed" ]; then
                    # We can retry! Add bws again
                    $p 'Resetting bandwidths to try again (try '$bw_try_count')'
                    echobandwidths >> $f_persist
                    sed -i "/^bw_try_count=/s/=.*/=$bw_try_count/" $f_persist
                else
                    $p 'Retry count exceeded, will stay after reboot'
                    echo "bw_changes_stop=1" > $f_persist
                fi
            fi
            # Revert to first bw and reboot
            $p 'Reverting to first bandwidth ('$bw_orig'MHz)'
            setbw $bw_orig; savecfg; restart
            # Now it is the end. Really. Remember me. Goodbye. Sniff.
        fi
    done
}
splitN() {
    qt_split=$1
    awk '
    {
        if( (NR%'$qt_split')==0 ) {
            printf("%s\n",$0)
        } else {
            printf("%s ",$0)
        }
    }
    END {
        if( (NR%'$qt_split')!=0 )print ""
    }'
}
findchan() {
    f_scan=$(mktemp -t) # temp file for iwconfig scanned list output
    f_fail=$(mktemp -t) # temp file for failed channels tried
    # grep -f pattern_file will barf if this file is empty
    echo ABCDEFGHIJKLMNOPQRSTUVWXYZ > $f_fail

    $p 'Enabling all channels'
    athchans -i ath0 all

    $p 'Available channels:'
    full_range=$(athchans -i ath0 -g | sed 's/ /\n/g')
    echo "$full_range" | splitN 15 | while read line; do $p $line; done

    $p 'Searching all channels range'
    iwlist ath0 scanning > $f_scan
    $p 'Active channels search complete'

    ssid=$(sed -n '/wireless\.[0-9]\.ssid=/{s/.*=//;p;q;}' $f_syscfg)
    $p 'SSID is set to: "'$ssid'"'

    checkdisabled
    # First test - try the channels that match the SSID, if any
    test_channels=$(grepchan '"'$ssid'"')
    failcode=1
    if [ -n "$test_channels" ]; then
        $p 'SSID found on full scan, channel(s): '$test_channels
        trychans && failcode=0
        echo "$test_channels" > $f_fail
    fi
    checkdisabled
    # Second test - try every other channel detected
    # Maybe hidden SSID and stuff.
    if [ "$failcode" = "1" ]; then
        test_channels=$(grepchan '')
        if [ -n "$test_channels" ]; then
            $p 'Trying all detected channels'
            trychans && failcode=0
            echo "$test_channels" >> $f_fail
        fi
    fi
    checkdisabled
    # Third -test - try everything else.
    if [ "$failcode" = "1" ]; then
        # Randomize full range list so each channel has a chance to be first
        full_range=$(echo "$full_range" | awk 'BEGIN{srand();}{print rand()"\t"$0}'|sort -n|cut -f2-)
        test_channels=$(echo "$full_range" | grep -v -f $f_fail)
        if [ -n "$test_channels" ]; then
            $p 'Trying the other undetected channels'
            trychans && failcode=0
        fi
    fi
    if [ "$failcode" = "1" ]; then
        $p 'Still unassociated!'
    else
        $p 'We are now officially associated! :)'
        $p $(iwlist ath0 channel|grep Current)
    fi
    rm $f_scan $f_fail
    return $failcode
}
trychans() {
    echo "$test_channels" | splitN $try_count | while read chan_list; do
        checkdisabled
        $p 'Trying at most '$try_count' channels in '$bw_curr'MHz bandwidth:'
        # Sort the channels printed on the screen, easy to read ;)
        echo "$chan_list" | sed 's/ /\n/g' | sort -n | splitN 15 |
            while read line; do $p $line; done
        # Now set the channels and
        athchans -i ath0 $chan_list
        #iwconfig ath0 commit # beams do not need this, but potato
        $p 'Surveying and waiting for '$try_sleep' seconds'
        # This somehow works faster than just waiting for a fix
        iwlist ath0 scanning >/dev/null
        # Wait...
        waitfor=$try_sleep
        while true; do
            checkassoc && break
            waitfor=$((waitfor-1))
            [ "$waitfor" = "0" ] && break
            sleep 1
        done
        # Check and get outta here
        checkassoc && break
    done
    # Return the results of the last check
    checkassoc
    return $?
}
getsession() {
    bw_new=""
    if [ ! -f "$f_persist" ]; then # no session file, create it
        $p 'Creating session file list of pending bandwidths to try'
        echo bw_orig=$bw_curr > $f_persist
        echo bw_changes_stop=0 >> $f_persist
        echo bw_try_count=0 >> $f_persist
        echobandwidths >> $f_persist
    fi
    source $f_persist
}
echobandwidths() {
    # reverse bandwidths to try, so the more preferred gets set after all others
    echo "$bw_all" | sed 's/ /\n/g' | sed '1!G;h;$!d' | while read bw; do
        [ "$bw" = "$bw_curr" ] && continue
        [ -z "$bw" ] && continue
        echo bw_new=$bw
    done
}
savecfg() {
    $p 'Saving'
    cfgmtd -w -p /etc/
    return 0
}
restart() {
    # I won't take chances with softrestart :)
    #/usr/etc/rc.d/rc.softrestart save
    rm $f_lock # just in case
    reboot
    exit 1 # just in case too
}
echolog() {
    logger $@
    echo $(date) $@ >> $f_log
    return 0
}
checkdisabled() {
    if [ ! -f "$f_lock" ]; then
        $p 'lock file removed, exiting at user request'
        exit 99
    fi
}

if [ "$1" = "enableCT" -o "$1" = "disableCT" -o "$1" = "install" -o "$1" = "uninstall" ]; then
    $1 # XXX this can be easily be overlooked - it calls the functions of same name ok?
elif [ "$1" = "auto" -o "$1" = "manual" ]; then
    if ! grep -q 'countrycode=511' $f_syscfg; then
        echo 'Error: Compliance test mode is not enabled in config. Enable, save, reboot.'
    elif ! iwconfig ath0 | grep -q Mode:Managed; then
        echo 'Error: Only managed mode (station/client) supported!'
    elif ! grep -q 'wireless\.1\.scan_list\.status=disabled' $f_syscfg; then
        echo 'Error: Channel list has to be DISABLED!'
    elif [ -f $f_lock ]; then
        echo -e "Error: Refusing to run - file $f_lock already exists.\n\tKill existing script if any, remove it and try again."
    elif [ "$1" = "auto" ]; then
        p=echolog
        echo 'Backgrounding auto scan'
        scan &
        echo "$!" > $f_lock
    else
        p=echo
        scan
    fi
    return $?
elif [ "$1" = "getbw" ]; then
    echo $(getbw)MHz
elif [ "$1" = "setbw" -a -n "$2" ]; then
    setbw $2
    echo $(getbw)MHz
else
    echo 'Invalid usage. Arguments available:'
    echo $0 '[ install | uninstall | enableCT | disableCT ]'
    echo $0 '[ auto | manual | getbw | setbw ]'
    echo
    echo 'install | uninstall'
    echo '  Add/remove call to this script to rc.poststart.'
    echo 'enableCT | disableCT'
    echo '  Adds/removes compliance test (only) to web interface and enables it in the'
    echo '  config file. Reboot to apply.'
    echo 'auto'
    echo '  Runs continually in the background trying to associate on all available channels'
    echo 'manual'
    echo '  One-time scan/association try, only on current bandwidth.'
    echo 'getbw/setbw'
    echo '  Shortcut to get/set current bandwidth. Valid values for setting: 20/40 30 25 20 10 5'
    echo '  Do not forget to save/reboot after setting bandwidth.'
    exit 1
fi
exit 0