From 4a26059d00e6cac5d82286b4f268a39c4d68a7c7 Mon Sep 17 00:00:00 2001
From: Zach Brown <zab@versity.com>
Date: Thu, 31 Oct 2024 11:21:51 -0700
Subject: [PATCH] Add lock-shrink-read-race test

Add a quick test that races readers and shrinking to stress lock object
refcount racing between concurrent lock request handling threads in the
lock server.

Signed-off-by: Zach Brown <zab@versity.com>
---
 tests/golden/lock-shrink-read-race   |  2 ++
 tests/sequence                       |  1 +
 tests/tests/lock-shrink-read-race.sh | 40 ++++++++++++++++++++++++++++
 3 files changed, 43 insertions(+)
 create mode 100644 tests/golden/lock-shrink-read-race
 create mode 100644 tests/tests/lock-shrink-read-race.sh

diff --git a/tests/golden/lock-shrink-read-race b/tests/golden/lock-shrink-read-race
new file mode 100644
index 00000000..eb528076
--- /dev/null
+++ b/tests/golden/lock-shrink-read-race
@@ -0,0 +1,2 @@
+=== setup
+=== spin reading and shrinking
diff --git a/tests/sequence b/tests/sequence
index 16e0e465..78001e59 100644
--- a/tests/sequence
+++ b/tests/sequence
@@ -25,6 +25,7 @@ totl-xattr-tag.sh
 quota.sh
 lock-refleak.sh
 lock-shrink-consistency.sh
+lock-shrink-read-race.sh
 lock-pr-cw-conflict.sh
 lock-revoke-getcwd.sh
 lock-recover-invalidate.sh
diff --git a/tests/tests/lock-shrink-read-race.sh b/tests/tests/lock-shrink-read-race.sh
new file mode 100644
index 00000000..18693bf0
--- /dev/null
+++ b/tests/tests/lock-shrink-read-race.sh
@@ -0,0 +1,40 @@
+#
+# We had a lock server refcounting bug that could let one thread get a
+# reference on a lock struct that was being freed by another thread.  We
+# were able to reproduce this by having all clients try and produce a
+# lot of read and null requests.
+#
+# This will manfiest as a hung lock and timed out test runs, probably
+# with hung task messages on the console.  Depending on how the race
+# turns out, it can trigger KASAN warnings in
+# process_waiting_requests().
+#
+
+READERS_PER=3
+SECS=30
+
+echo "=== setup"
+touch "$T_D0/file"
+
+echo "=== spin reading and shrinking"
+END=$((SECONDS + SECS))
+for m in $(t_fs_nrs); do
+	eval file="\$T_D${m}/file"
+
+	# lots of tasks reading as fast as they can
+	for t in $(seq 1 $READERS_PER); do
+		(while [ $SECONDS -lt $END ]; do
+			stat $file > /dev/null
+		 done) &
+	done
+	# one task shrinking (triggering null requests) and reading
+	(while [ $SECONDS -lt $END ]; do
+		stat $file > /dev/null
+		t_trigger_arm_silent statfs_lock_purge $m
+		stat -f "$file" > /dev/null
+	 done) &
+done
+
+wait
+
+t_pass