#!/usr/bin/env bash
#
# JbTecWiz Support Centre -- generated fix script
#
# Fault : NFS: "Stale file handle" and hung mounts
# Fix   : Recover from a hung hard mount, and stop it recurring
# Source: https://jbtecwiz.com/support/lnx-sto-nfs-stale
#
# Run as     : Root shell on the client
# Expect     : 30 minutes
# Risk       : medium
# Reversible : yes
#
# WHEN THIS IS THE RIGHT FIX
#   Everything touching the path hangs.
#
# HOW TO UNDO IT
#   Restore the original fstab line.
#
# Walks the fix one step at a time and asks before each. Steps with no
# command are yours to do -- it prints those and waits. DRYRUN=1 prints
# without executing; UNATTENDED=1 does not ask.
#
# --------------------------------------------------------------------
# NO WARRANTY - USE AT YOUR OWN RISK
#
# This script is provided by JbTecWiz as-is and with no warranty of any
# kind, express or implied. You run it entirely at your own risk.
#
# JbTecWiz accepts no liability for any loss or damage arising from its
# use, including but not limited to data loss, downtime, or configuration
# changes that turn out to be wrong for your system.
#
# You are responsible for reading this script before running it, for
# satisfying yourself that it suits the machine in front of you, and for
# having a working backup first. Some steps cannot be undone.
# --------------------------------------------------------------------

set -uo pipefail
DRYRUN="${DRYRUN:-0}"
UNATTENDED="${UNATTENDED:-0}"
failed=0

if [ "$(id -u)" -ne 0 ]; then
  echo "  This fix is documented as needing root. Re-run with sudo." >&2
  exit 3
fi

rule() { printf "\n%s\n" "$(printf '-%.0s' $(seq 1 70))"; if [ $# -gt 0 ]; then echo "$1"; fi; }
prose() { echo "$1" | fold -s -w 74 | sed "s/^/  /"; }

# Returns 0 when the caller should run the command, 1 when it should not.
# A manual step always returns 1 -- there is nothing for the caller to run.
step() {  # step <n> <do> <why> <manual|cmd> [command lines...]
  local n="$1" dotext="$2" why="$3" mode="$4"; shift 4
  rule "  Step $n of 6"
  prose "$dotext"
  if [ -n "$why" ]; then echo; prose "$why"; fi
  if [ "$mode" = "manual" ]; then
    echo; echo "  -> Do this yourself, then press Enter to carry on."
    if [ "$UNATTENDED" = "0" ] && [ "$DRYRUN" = "0" ]; then read -r _; fi
    return 1
  fi
  echo; printf "    %s\n" "$@"; echo
  if [ "$DRYRUN" = "1" ]; then echo "  (dry run -- not executed)"; return 1; fi
  if [ "$UNATTENDED" = "0" ]; then
    read -r -p "  Run this step? [Y]es / [S]kip / [Q]uit " a
    case "$a" in
      [Qq]*) echo "  Stopped at your request."; exit 0 ;;
      [Ss]*) echo "  Skipped."; return 1 ;;
    esac
  fi
  return 0
}

rule
echo "  NFS: "Stale file handle" and hung mounts"
echo "  Recover from a hung hard mount, and stop it recurring"
echo
echo "  Risk: medium   Reversible   30 minutes"
echo
prose 'No warranty. Use at your own risk - JbTecWiz accepts no liability. Read it before you run it, and have a backup.'
rule
echo
if [ "$UNATTENDED" = "0" ] && [ "$DRYRUN" = "0" ]; then
  read -r -p "  Ready? [y/N] " go
  case "$go" in [Yy]*) ;; *) echo "  Nothing was changed."; exit 0;; esac
fi

if step 1 'Confirm the diagnosis -- processes in D state on that mount.' '' cmd 'ps -eo pid,stat,wchan:24,comm | awk '\''$2 ~ /D/'\''' 'dmesg -T | grep -i '\''nfs.*not responding'\'' | tail'; then
  ps -eo pid,stat,wchan:24,comm | awk '$2 ~ /D/'
  dmesg -T | grep -i 'nfs.*not responding' | tail
  if [ $? -ne 0 ]; then
    failed=$((failed+1))
    echo "  Step 1 failed. The rest of the fix may depend on it." >&2
  fi
fi

if step 2 'Detach it lazily so new commands stop hanging.' '' cmd 'sudo umount -l /mnt/nfs'; then
  sudo umount -l /mnt/nfs
  if [ $? -ne 0 ]; then
    failed=$((failed+1))
    echo "  Step 2 failed. The rest of the fix may depend on it." >&2
  fi
fi

step 3 'Understand the trade-off before changing the mount options. A hard mount blocks forever, which protects data -- writes are retried until the server returns. A soft mount gives up and returns an error, which can silently corrupt a file being written.' 'Soft mounts are the usual advice for this fault and they trade a hang for possible data loss. For read-only data that is a fine trade; for anything being written it is not.' manual || true

if step 4 'Add intr and a timeout where a hard mount is right but responsiveness matters.' '' cmd 'sudo sed -i '\''s|\(nfs.*defaults\)|\1,_netdev,hard,timeo=600,retrans=2|'\'' /etc/fstab' 'sudo mount -a'; then
  sudo sed -i 's|\(nfs.*defaults\)|\1,_netdev,hard,timeo=600,retrans=2|' /etc/fstab
  sudo mount -a
  if [ $? -ne 0 ]; then
    failed=$((failed+1))
    echo "  Step 4 failed. The rest of the fix may depend on it." >&2
  fi
fi

if step 5 'Use autofs for mounts that are not always needed -- it mounts on access and unmounts when idle, which removes the whole class of boot-time hangs.' '' cmd 'sudo apt install autofs'; then
  sudo apt install autofs
  if [ $? -ne 0 ]; then
    failed=$((failed+1))
    echo "  Step 5 failed. The rest of the fix may depend on it." >&2
  fi
fi

step 6 'Always add _netdev to an NFS entry in fstab, or the machine will hang at boot waiting for a mount before the network exists.' '' manual || true

rule "  Confirm it worked"
prose 'The mount recovers when the server returns, and a boot with the server unavailable still completes.'
if [ "$DRYRUN" = "0" ]; then
  mount | grep nfs
  systemd-analyze blame | head
fi

rule
if [ "$failed" -gt 0 ]; then
  echo "  Finished with $failed failed step(s)."
  echo "  Read the full write-up at https://jbtecwiz.com/support/lnx-sto-nfs-stale"
else
  echo "  Finished."
fi
echo
prose 'To undo: Restore the original fstab line.'
rule