ctdb/config/debug_locks.sh

   1 #!/bin/sh
   2
   3 # This script parses /proc/locks and finds the processes that are holding
   4 # locks on CTDB databases.  For all those processes the script dumps a
   5 # stack trace.
   6 #
   7 # This script can be used only if Samba is configured to use fcntl locks
   8 # rather than mutex locks.
   9
  10 [ -n "$CTDB_BASE" ] || \
  11     CTDB_BASE=$(d=$(dirname "$0") ; cd -P "$d" ; echo "$PWD")
  12
  13 . "${CTDB_BASE}/functions"
  14
  15 # type is at least mentioned in POSIX and more is portable than which(1)
  16 # shellcheck disable=SC2039
  17 if ! type gstack >/dev/null 2>&1 ; then
  18         gstack ()
  19         {
  20                 _pid="$1"
  21
  22                 gdb -batch --quiet -nx "/proc/${_pid}/exe" "$_pid" \
  23                     -ex "thread apply all bt" 2>/dev/null |
  24                         grep '^\(#\|Thread \)'
  25         }
  26 fi
  27
  28 # Load/cache database options from configuration file
  29 ctdb_get_db_options
  30
  31 (
  32     flock -n 9 || exit 1
  33
  34     echo "===== Start of debug locks PID=$$ ====="
  35
  36     # Create sed expression to convert inodes to names.
  37     # Filenames don't contain dashes and we want basenames
  38     # shellcheck disable=SC2035
  39     sed_cmd=$(cd "$CTDB_DBDIR" &&
  40                   stat -c "s#[0-9a-f]*:[0-9a-f]*:%i #%n #" *.tdb.* 2>/dev/null ;
  41               cd "$CTDB_DBDIR_PERSISTENT" &&
  42                   stat -c "s#[0-9a-f]*:[0-9a-f]*:%i #%n #" *.tdb.* 2>/dev/null)
  43
  44     # Parse /proc/locks and extract following information
  45     #    pid process_name tdb_name offsets [W]
  46     out=$( grep -F "POSIX  ADVISORY  WRITE" /proc/locks |
  47     awk '{ if($2 == "->") { print $6, $7, $8, $9, "W" } else { print $5, $6, $7, $8 } }' |
  48     while read pid rest ; do
  49         pname=$(readlink "/proc/${pid}/exe")
  50         echo "$pid $pname $rest"
  51     done | sed -e "$sed_cmd" | grep '\.tdb' )
  52
  53     if [ -n "$out" ]; then
  54         # Log information about locks
  55         echo "$out"
  56
  57         # Find processes that are waiting for locks
  58         dbs=$(echo "$out" | grep "W$" | awk '{print $3}')
  59         all_pids=""
  60         for db in $dbs ; do
  61             pids=$(echo "$out" | grep -v "W$" | grep "$db" | grep -v ctdbd | awk '{print $1}')
  62             all_pids="$all_pids $pids"
  63         done
  64         # Use word splitting to squash whitespace
  65         # shellcheck disable=SC2086
  66         pids=$(echo $all_pids | tr ' ' '\n' | sort -u)
  67
  68         # For each process waiting, log stack trace
  69         for pid in $pids ; do
  70             echo "----- Stack trace for PID=$pid -----"
  71             # x is intentionally ignored
  72             # shellcheck disable=SC2034
  73             read x x state x <"/proc/${pid}/stat"
  74             if [ "$state" = "D" ] ; then
  75                 # Don't run gstack on a process in D state since
  76                 # gstack will hang until the process exits D state.
  77                 # Although it is possible for a process to transition
  78                 # to D state after this check, it is unlikely because
  79                 # if a process is stuck in D state then it is probably
  80                 # the reason why this script was called.  Note that a
  81                 # kernel stack almost certainly won't help diagnose a
  82                 # deadlock... but it will probably give us someone to
  83                 # blame!
  84                 echo "----- Process in D state, printing kernel stack only"
  85                 cat "/proc/${pid}/stack"
  86             else
  87                 gstack "$pid"
  88             fi
  89         done
  90     fi
  91
  92     echo "===== End of debug locks PID=$$ ====="
  93 )9>"${CTDB_SCRIPT_VARDIR}/debug_locks.lock" | script_log "ctdbd-lock"
  94
  95 exit 0