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 # Default fallback location for database directories.
  16 # These can be overwritten from CTDB configuration
  17 CTDB_DBDIR="${CTDB_VARDIR}"
  18 CTDB_DBDIR_PERSISTENT="${CTDB_VARDIR}/persistent"
  19
  20 loadconfig ctdb
  21
  22 (
  23     flock -n 9 || exit 1
  24
  25     echo "===== Start of debug locks PID=$$ ====="
  26
  27     # Create sed expression to convert inodes to names.
  28     # Filenames don't contain dashes and we want basenames
  29     # shellcheck disable=SC2035
  30     sed_cmd=$(cd "$CTDB_DBDIR" &&
  31                   stat -c "s#[0-9a-f]*:[0-9a-f]*:%i #%n #" *.tdb.* 2>/dev/null ;
  32               cd "$CTDB_DBDIR_PERSISTENT" &&
  33                   stat -c "s#[0-9a-f]*:[0-9a-f]*:%i #%n #" *.tdb.* 2>/dev/null)
  34
  35     # Parse /proc/locks and extract following information
  36     #    pid process_name tdb_name offsets [W]
  37     out=$( grep -F "POSIX  ADVISORY  WRITE" /proc/locks |
  38     awk '{ if($2 == "->") { print $6, $7, $8, $9, "W" } else { print $5, $6, $7, $8 } }' |
  39     while read pid rest ; do
  40         pname=$(readlink "/proc/${pid}/exe")
  41         echo "$pid $pname $rest"
  42     done | sed -e "$sed_cmd" | grep "\.tdb" )
  43
  44     if [ -n "$out" ]; then
  45         # Log information about locks
  46         echo "$out"
  47
  48         # Find processes that are waiting for locks
  49         dbs=$(echo "$out" | grep "W$" | awk '{print $3}')
  50         all_pids=""
  51         for db in $dbs ; do
  52             pids=$(echo "$out" | grep -v "W$" | grep "$db" | grep -v ctdbd | awk '{print $1}')
  53             all_pids="$all_pids $pids"
  54         done
  55         # Use word splitting to squash whitespace
  56         # shellcheck disable=SC2086
  57         pids=$(echo $all_pids | tr " " "\n" | sort -u)
  58
  59         # For each process waiting, log stack trace
  60         for pid in $pids ; do
  61             echo "----- Stack trace for PID=$pid -----"
  62             # x is intentionally ignored
  63             # shellcheck disable=SC2034
  64             read x x state x <"/proc/${pid}/stat"
  65             if [ "$state" = "D" ] ; then
  66                 # Don't run gstack on a process in D state since
  67                 # gstack will hang until the process exits D state.
  68                 # Although it is possible for a process to transition
  69                 # to D state after this check, it is unlikely because
  70                 # if a process is stuck in D state then it is probably
  71                 # the reason why this script was called.  Note that a
  72                 # kernel stack almost certainly won't help diagnose a
  73                 # deadlock... but it will probably give us someone to
  74                 # blame!
  75                 echo "----- Process in D state, printing kernel stack only"
  76                 cat "/proc/${pid}/stack"
  77             else
  78                 gstack "$pid"
  79                 # gcore -o /var/log/core-deadlock-ctdb $pid
  80             fi
  81         done
  82     fi
  83
  84     echo "===== End of debug locks PID=$$ ====="
  85 )9>"${CTDB_SCRIPT_VARDIR}/debug_locks.lock" | script_log "ctdbd-lock"
  86
  87 exit 0