Skip to contentSkip to conversation

pids.sh: Wait for killed tasks to leave the cgroup

PATCH In review Avinesh Kumar via ltp 3 participants 3 messages Nov 16

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 pids.sh: Wait for killed tasks to leave the cgroup

From: Avinesh Kumar <[email protected]>

pids_task1 leaves an orphaned child in the cgroup, so "wait" in stop_pids_tasks_path() can't wait for it, and kill -9 only queues the signal. The rmdir that follows can fail with EBUSY (unchecked), leaving the child cgroup behind and breaking the cgroup cleanup:

rmdir: failed to remove '.../test-3200/child1': Device or resource busy tst_cgroup.c:1052: TBROK: unlinkat(..., 'test-3200', AT_REMOVEDIR): EBUSY

Wait for the cgroup to become empty at the end of stop_pids_tasks_path() and in cleanup() before the test cgroup is removed.

Fixes dd2b901a9 ("controllers: Update pids.sh to use newer cgroup lib")
---
 testcases/kernel/controllers/pids/pids.sh | 21 +++++++++++++++++++++
 1 file changed, 21 insertions(+)
testcases/kernel/controllers/pids/pids.sh +21 -0
index dd27b5616f0c..32434ab0dba0 100755
@@ -32,11 +32,28 @@ OPTIONS
 EOF
 }
 
+wait_cgroup_empty()
+{
+	local delay=1000
+
+	while [ -n "$(cat "$1/$task_list" 2>/dev/null)" ]; do
+		if [ $delay -gt 4000000 ]; then
+			# tst_brk() only warns in cleanup(), so return after it
+			tst_brk TBROK "cgroup $1 is not empty"
+			return
+		fi
+		tst_sleep ${delay}us
+		delay=$((delay * 2))
+	done
+}
+
 cleanup()
 {
 	killall -9 pids_task2 >/dev/null 2>&1
 	killall -9 pids_task1 >/dev/null 2>&1
 
+	[ -n "$testpath" ] && wait_cgroup_empty "$testpath"
+
 	cgroup_cleanup
 }
 
@@ -83,6 +100,10 @@ stop_pids_tasks_path()
 		kill -9 $i 2>/dev/null
 		wait $i 2>/dev/null
 	done
+
+	# wait can't reap the orphaned pids_task1 children, so wait for them
+	# to leave the cgroup instead
+	wait_cgroup_empty "$path"
 }
 
 case1()
Signature · 5 lines
2.56.0


-- 
Mailing list info: https://lists.linux.it/listinfo/ltp
#1 lore raw

Hi Avinesh,

Quoted 1 line · Avinesh Kumar

pids.sh: Wait for killed tasks to leave the cgroup

Verdict - Reviewed

---
Note:

The agent can sometimes produce false positives although often its findings are genuine. If you find issues with the review, please comment this email or ignore the suggestions.

Regards,
LTP AI Reviewer

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
#2 lore raw
In reply to Avinesh Kumar via ltp

Hi!

Quoted 51 lines

pids_task1 leaves an orphaned child in the cgroup, so "wait" in stop_pids_tasks_path() can't wait for it, and kill -9 only queues the signal. The rmdir that follows can fail with EBUSY (unchecked), leaving the child cgroup behind and breaking the cgroup cleanup:

rmdir: failed to remove '.../test-3200/child1': Device or resource busy tst_cgroup.c:1052: TBROK: unlinkat(..., 'test-3200', AT_REMOVEDIR): EBUSY

Wait for the cgroup to become empty at the end of stop_pids_tasks_path() and in cleanup() before the test cgroup is removed.

Fixes dd2b901a9 ("controllers: Update pids.sh to use newer cgroup lib")
---
 testcases/kernel/controllers/pids/pids.sh | 21 +++++++++++++++++++++
 1 file changed, 21 insertions(+)
testcases/kernel/controllers/pids/pids.sh +17 -0
index dd27b5616f0c..32434ab0dba0 100755
@@ -32,11 +32,28 @@ OPTIONS
 EOF
 }
 
+wait_cgroup_empty()
+{
+	local delay=1000
+
+	while [ -n "$(cat "$1/$task_list" 2>/dev/null)" ]; do
+		if [ $delay -gt 4000000 ]; then
+			# tst_brk() only warns in cleanup(), so return after it
+			tst_brk TBROK "cgroup $1 is not empty"
+			return
+		fi
+		tst_sleep ${delay}us
+		delay=$((delay * 2))
+	done
+}
+
 cleanup()
 {
 	killall -9 pids_task2 >/dev/null 2>&1
 	killall -9 pids_task1 >/dev/null 2>&1
 
+	[ -n "$testpath" ] && wait_cgroup_empty "$testpath"
+
 	cgroup_cleanup
 }
 

testcases/kernel/controllers/pids/pids.sh stop_pids_tasks_path() · line 107

Quoted 8 lines
testcases/kernel/controllers/pids/pids.sh +4 -0
@@ -83,6 +100,10 @@ stop_pids_tasks_path()
 		kill -9 $i 2>/dev/null
 		wait $i 2>/dev/null
 	done
+
+	# wait can't reap the orphaned pids_task1 children, so wait for them
+	# to leave the cgroup instead
+	wait_cgroup_empty "$path"
 }

This is working around the real problem. The tst_cg_cleanup() function in lib/tst_cgroup.c that is eventually called from cgroup_cleanup (testcases/lib/tst_cgctl.c) makes sure there are no processes left before unlinking the cgroups.

The ugly part of the shell cgroup library is that it dumps the cgroup library state into a file and then re-loads it in each command. So my guess is that there is a bug somewhere in that code.

And I also suppose that it would be easier to rewrite the test into C so that we can use the library natively than trying to find the problem.

Signature · 5 lines
Cyril Hrubis
[email protected]

-- 
Mailing list info: https://lists.linux.it/listinfo/ltp