#!/bin/sh
-# This script parses /proc/locks and finds the processes that are holding
-# locks on CTDB databases. For all those processes the script dumps a
-# stack trace.
+# This script attempts to find processes holding locks on a particular
+# CTDB database and dumps a stack trace for each such processe.
#
-# This script can be used only if Samba is configured to use fcntl locks
-# rather than mutex locks.
+# There are 2 cases:
+#
+# * Samba is configured to use fcntl locks
+#
+# In this case /proc/locks is parsed to find potential lock holders
+#
+# * Samba is configured to use POSIX robust mutexes
+#
+# In this case the helper program tdb_mutex_check is used to find
+# potential lock holders.
+#
+# This helper program uses a private glibc struct field, so is
+# neither portable nor supported. If this field is not available
+# then the helper is not built. Unexpected changes in internal
+# glibc structures may cause unexpected results, including crashes.
+# Bug reports for this helper program are not accepted without an
+# accompanying patch.
[ -n "$CTDB_BASE" ] || \
- CTDB_BASE=$(d=$(dirname "$0") ; cd -P "$d" ; dirname "$PWD")
+ CTDB_BASE=$(d=$(dirname "$0") && cd -P "$d" && echo "$PWD")
. "${CTDB_BASE}/functions"
-# Default fallback location for database directories.
-# These can be overwritten from CTDB configuration
-CTDB_DBDIR="${CTDB_VARDIR}"
-CTDB_DBDIR_PERSISTENT="${CTDB_VARDIR}/persistent"
+if [ $# -ne 4 ] ; then
+ die "usage: $0 <pid> { DB | RECORD } <tdb_path> { FCNTL | MUTEX }"
+fi
-loadconfig ctdb
+lock_helper_pid="$1"
+# lock_scope is unused for now
+# shellcheck disable=SC2034
+lock_scope="$2"
+tdb_path="$3"
+lock_type="$4"
-(
- flock -n 9 || exit 1
-
- echo "===== Start of debug locks PID=$$ ====="
-
- # Create sed expression to convert inodes to names.
- # Filenames don't contain dashes and we want basenames
- # shellcheck disable=SC2035
- sed_cmd=$(cd "$CTDB_DBDIR" &&
- stat -c "s#[0-9a-f]*:[0-9a-f]*:%i #%n #" *.tdb.* 2>/dev/null ;
- cd "$CTDB_DBDIR_PERSISTENT" &&
- stat -c "s#[0-9a-f]*:[0-9a-f]*:%i #%n #" *.tdb.* 2>/dev/null)
-
- # Parse /proc/locks and extract following information
- # pid process_name tdb_name offsets [W]
- out=$( grep -F "POSIX ADVISORY WRITE" /proc/locks |
- awk '{ if($2 == "->") { print $6, $7, $8, $9, "W" } else { print $5, $6, $7, $8 } }' |
- while read pid rest ; do
- pname=$(readlink "/proc/${pid}/exe")
- echo "$pid $pname $rest"
- done | sed -e "$sed_cmd" | grep "\.tdb" )
-
- if [ -n "$out" ]; then
- # Log information about locks
- echo "$out"
-
- # Find processes that are waiting for locks
- dbs=$(echo "$out" | grep "W$" | awk '{print $3}')
- all_pids=""
- for db in $dbs ; do
- pids=$(echo "$out" | grep -v "W$" | grep "$db" | grep -v ctdbd | awk '{print $1}')
- all_pids="$all_pids $pids"
- done
- # Use word splitting to squash whitespace
- # shellcheck disable=SC2086
- pids=$(echo $all_pids | tr " " "\n" | sort -u)
-
- # For each process waiting, log stack trace
- for pid in $pids ; do
- echo "----- Stack trace for PID=$pid -----"
- # x is intentionally ignored
- # shellcheck disable=SC2034
- read x x state x <"/proc/${pid}/stat"
- if [ "$state" = "D" ] ; then
+# type is at least mentioned in POSIX and more is portable than which(1)
+# shellcheck disable=SC2039
+if ! type gstack >/dev/null 2>&1 ; then
+ gstack ()
+ {
+ _pid="$1"
+
+ gdb -batch --quiet -nx "/proc/${_pid}/exe" "$_pid" \
+ -ex "thread apply all bt" 2>/dev/null |
+ grep '^\(#\|Thread \)'
+ }
+fi
+
+# Load/cache database options from configuration file
+ctdb_get_db_options
+
+dump_stack ()
+{
+ _pid="$1"
+
+ echo "----- Stack trace for PID=${_pid} -----"
+ _state=$(ps -p "$_pid" -o state= | cut -c 1)
+ if [ "$_state" = "D" ] ; then
# Don't run gstack on a process in D state since
# gstack will hang until the process exits D state.
# Although it is possible for a process to transition
# deadlock... but it will probably give us someone to
# blame!
echo "----- Process in D state, printing kernel stack only"
- cat "/proc/${pid}/stack"
- else
- gstack "$pid"
- # gcore -o /var/log/core-deadlock-ctdb $pid
- fi
+ get_proc "${_pid}/stack"
+ else
+ gstack "$_pid"
+ fi
+}
+
+dump_stacks ()
+{
+ _pids="$1"
+
+ # Use word splitting to squash whitespace
+ # shellcheck disable=SC2086
+ _pids=$(echo $_pids | tr ' ' '\n' | sort -u)
+
+ for _pid in $_pids; do
+ dump_stack "$_pid"
done
- fi
+}
+
+get_tdb_file_id ()
+{
+ if ! _device_inode=$(stat -c "%d:%i" "$tdb_path" 2>/dev/null) ; then
+ die "Unable to stat \"${tdb_path}\""
+ fi
+ _device="${_device_inode%%:*}"
+ _device_major=$((_device >> 8))
+ _device_minor=$((_device & 0xff))
+ _inode="${_device_inode#*:}"
+ printf '%02x:%02x:%u\n' "$_device_major" "$_device_minor" "$_inode"
+}
+
+debug_via_proc_locks ()
+{
+ # Get file ID to match relevant column in /proc/locks
+ _file_id=$(get_tdb_file_id)
+
+ # Log information from /proc/locks about the waiting process
+ _tdb=$(basename "$tdb_path")
+ _comm=$(ps -p "$lock_helper_pid" -o comm=)
+ _out=$(get_proc "locks" |
+ awk -v pid="$lock_helper_pid" \
+ -v file_id="$_file_id" \
+ -v file="$_tdb" \
+ -v comm="$_comm" \
+ '$2 == "->" &&
+ $3 == "POSIX" &&
+ $4 == "ADVISORY" &&
+ $5 == "WRITE" &&
+ $6 == pid &&
+ $7 == file_id { print $6, comm, file, $8, $9 }')
+ if [ -n "$_out" ] ; then
+ echo "Waiter:"
+ echo "$_out"
+ fi
+
+ # Parse /proc/locks and find process holding locks on $tdb_path
+ # extract following information
+ # pid process_name tdb_name offsets
+ _out=$(get_proc "locks" |
+ awk -v pid="$lock_helper_pid" \
+ -v file_id="$_file_id" \
+ -v file="$_tdb" \
+ '$2 == "POSIX" &&
+ $3 == "ADVISORY" &&
+ $4 == "WRITE" &&
+ $5 != pid &&
+ $6 == file_id { print $5, file, $7, $8 }' |
+ while read -r _pid _rest ; do
+ _pname=$(ps -p "$_pid" -o comm=)
+ echo "$_pid $_pname $_rest"
+ done)
+
+ if [ -z "$_out" ]; then
+ return
+ fi
+
+ # Log information about locks
+ echo "Lock holders:"
+ echo "$_out"
+
+ _pids=$(echo "$_out" | awk '{ print $1 }')
+
+ lock_holder_pids="${lock_holder_pids:+${lock_holder_pids} }${_pids}"
+}
+
+debug_via_tdb_mutex ()
+{
+ _helper="${CTDB_HELPER_BINDIR}/tdb_mutex_check"
+ if [ ! -x "$_helper" ] ; then
+ # Mutex helper not available - not supported?
+ # Avoid not found error...
+ return
+ fi
+
+ # Helper should always succeed
+ if ! _t=$("$_helper" "$tdb_path") ; then
+ return
+ fi
+
+ _out=$(echo "$_t" | sed -n -e 's#^\[\(.*\)\] pid=\(.*\)#\2 \1#p')
+
+ if [ -z "$_out" ]; then
+ if [ -n "$_t" ] ; then
+ echo "$_t" | grep -F 'trylock failed'
+ fi
+ return
+ fi
+
+ # Get process names, append $tdb_path
+ _out=$(echo "$_out" |
+ while read -r _pid _rest ; do
+ _pname=$(ps -p "$_pid" -o comm=)
+ _tdb=$(basename "$tdb_path")
+ echo "${_pid} ${_pname} ${_tdb} ${_rest}"
+ done)
+
+ # Log information about locks
+ echo "Lock holders:"
+ echo "$_out"
+
+ # Get PIDs of processes that are holding locks
+ _pids=$(echo "$_out" |
+ awk -v pid="$lock_helper_pid" '$1 != pid {print $1}')
+
+ lock_holder_pids="${lock_holder_pids:+${lock_holder_pids} }${_pids}"
+}
+
+(
+ flock -n 9 || exit 1
+
+ echo "===== Start of debug locks PID=$$ ====="
+
+ lock_holder_pids=""
+
+ debug_via_proc_locks
+
+ if [ "$lock_type" = "MUTEX" ] ; then
+ debug_via_tdb_mutex
+ fi
+
+ dump_stacks "$lock_holder_pids"
- echo "===== End of debug locks PID=$$ ====="
+ echo "===== End of debug locks PID=$$ ====="
)9>"${CTDB_SCRIPT_VARDIR}/debug_locks.lock" | script_log "ctdbd-lock"
exit 0
git.samba.org / vlendec / samba-autobuild / .git / blobdiff