#!/bin/selfclean /bin/sh
main_pid=$$

log()
{
  echo "$@" >&2
}


fs_writable()
{
  local dir="$1"
  if ( : >"$dir/.check.tmp" ) 2>/dev/null; then
    rm -f "$dir/.check.tmp"
    return 0
  else
    return 1
  fi
}


label()
{
  "$@" 2>&1 | (
    shift
    while read line; do
      echo "$*: $line"
    done
  )
}


abort()
{
  if [ -z "$DONT_ABORT" ]; then
    log 'Error encountered and $DONT_ABORT is not set: aborting.'
    sleep 1
    kill -TERM $main_pid
    exit 98
  else
    log 'Not aborting - $DONT_ABORT is set.'
    sleep 10
  fi
}


retry()
{
  while :; do
    if "$@"; then
      sleep 5
      log "restarting '$*'"
    else
      log
      log "FATAL: process ($*) died unexpectedly with code $?"
      abort
    fi
  done
}


should_never_die()
{
  while :; do
    "$@"
    log
    log "FATAL: process ($*) died unexpectedly with code $?"
    abort
    log "restarting '$*'"
  done
}

main()
{
  # We originally tried to stress *both* ethernet and MoCA here, but that's
  # too hard if the bridge device is configured (which is the normal case).
  # So let's just send packets out the bridge interface, if it exists, or
  # eth0, otherwise.  The net effect is we'll end up sending packets over
  # MoCA if the thin bruno is connected over MoCA, or ethernet if it's
  # connected over ethernet, which is a lot like what should happen in real
  # life.
  if is-storage-box; then
    log "Fat bruno: operating in server mode."
    label should_never_die iperf -s &
    label should_never_die iperf -s -u &
  else
    if [ -z "$SERVER_IP" ]; then
      while :; do
        log "Thin bruno: broadcasting for servers..."
        SERVERS=$(find-servers)
        [ -n "$SERVERS" ] && break
      done
      log "Found servers: " $SERVERS
    else
      SERVERS=$SERVER_IP
      log "Hardcoded \$SERVER_IP: $SERVERS"
    fi

    for server in $SERVERS; do
      while ! nc $server 5001 </dev/null; do
        log "Waiting for iperf server to listen on $server:5001..."
        sleep 2
      done
    done

    for server in $SERVERS; do
      if [ -n "$MAX_BANDWIDTH" ]; then
        # iperf can't explicitly limit bandwidth in TCP mode.  Sigh.  So
        # skip TCP and use fast UDP instead.
        log "MAX_BANDWIDTH=$MAX_BANDWIDTH; testing UDP only."
        label retry iperf -t60 -i10 -u -c $server -b "${MAX_BANDWIDTH}M" -d &
      else
        log "MAX_BANDWIDTH not set; testing max rate TCP and slow UDP."
        label retry iperf -t60 -i10 -u -c $server -d &
        label retry iperf -t60 -i10    -c $server -d &
      fi
    done
  fi

  if fs_writable /var/media; then
    # fast random writes
    should_never_die stress-disk -p10 --random --write /var/media &
  else
    log "No /var/media; skipping disk-write test."
  fi

  # raw disk/usb reads
  for d in /dev/sd?; do
    if [ -e "$d" ]; then
      # fast sequential read (to force some seeks)
      should_never_die stress-disk -p10 "$d" &

      # slow small random read
      should_never_die stress-disk -p0 -c4096 -m10 --random "$d" &
    fi
  done

  # NAND reads
  #TODO(apenwarr): -c65536 causes memory allocation problems.
  # This seems to be because the MTD driver tries to allocate kernel memory
  # the same size as the read() parameter as a single contiguous block,
  # which, under heavy load, might not be available.  There's no good reason
  # for MTD to do this, afaik, so we could try to debug the kernel so we
  # can increase this parameter.  It could theoretically cause problems on
  # a normal system under high load, not just one being stresstested.
  if [ -e /dev/mtd/rootfs0 ]; then
    should_never_die stress-disk -p10 -c4096 /dev/mtd/rootfs0 &
  fi

  # NOR reads
  #TODO(apenwarr): see above about -c65536.
  if [ -e /dev/mtd/hnvram ]; then
    should_never_die stress-disk -p10 -c4096 /dev/mtd/hnvram &
  fi

  sleep 5
  #TODO(apenwarr): actually bind the following to cpu0 and cpu1
  should_never_die nice -n 19 spin cpu0-idle &
  should_never_die nice -n 19 spin cpu1-idle &

  wait
}


main 2>&1 | tee /proc/self/fd/2 | logos stresstest