[PATCH v3 3/3] nvme: add test for controller rescan under I/O load

[Martin Wilck]

Add a test that repeatedly rescans nvme controllers while doing IO
on an nvme namespace connected to these controllers. The purpose
of the test is to make sure that no I/O errors or data corruption
occurs because of the rescan operations. The test uses sub-second
sleeps, which can't be easily accomplished in bash because of
missing floating-point arithmetic (and because usleep(1) isn't
portable). Therefore an awk program is used to trigger the
device rescans.

Link: https://lore.kernel.org/linux-nvme/20240822201413.112268-1-mwilck@suse.com/
Signed-off-by: Martin Wilck <mwilck at suse.com>
Reviewed-by: Nilay Shroff (nilay at linux.ibm.com)
---
v3: (all changes suggested by Shinichiro Kawasaki)
    - add "Link:" tag
    - add comment with patch description
    - declare variable "st" local
    - use "mapfile -t" for array assignment
v2: - don't use usleep (Nilay Shroff). Use an awk program to do floating
      point arithmetic and achieve more accurate sub-second sleep times.
    - add 053.out (Nilay Shroff).
---
 tests/nvme/053     | 74 ++++++++++++++++++++++++++++++++++++++++++++++
 tests/nvme/053.out |  2 ++
 tests/nvme/rc      | 18 +++++++++++
 3 files changed, 94 insertions(+)
 create mode 100755 tests/nvme/053
 create mode 100644 tests/nvme/053.out

diff --git a/tests/nvme/053 b/tests/nvme/053
new file mode 100755
index 0000000..df643ec
--- /dev/null
+++ b/tests/nvme/053
@@ -0,0 +1,74 @@
+#!/bin/bash
+# SPDX-License-Identifier: GPL-3.0+
+# Copyright (C) 2024 Martin Wilck, SUSE LLC
+#
+# Repeatedly rescans nvme controllers while doing IO on an nvme namespace
+# connected to these controllers, and make sure that no I/O errors or data
+# corruption occurs.
+
+. tests/nvme/rc
+
+DESCRIPTION="test controller rescan under I/O load"
+TIMED=1
+: "${TIMEOUT:=60}"
+
+rescan_controller() {
+	local path
+	path="$1/rescan_controller"
+
+	[[ -f "$path" ]] || {
+		echo "cannot rescan $1"
+		return 1
+	}
+
+	awk -f "$TMPDIR/rescan.awk" \
+	    -v path="$path" -v timeout="$TIMEOUT" -v seed="$2" &
+}
+
+create_rescan_script() {
+	cat >"$TMPDIR/rescan.awk" <<EOF
+ at load "time"
+
+BEGIN {
+    srand(seed);
+    finish = gettimeofday() + strtonum(timeout);
+    while (gettimeofday() < finish) {
+	sleep(0.1 + 5 * rand());
+	printf("1\n") > path;
+	close(path);
+    }
+}
+EOF
+}
+
+test_device() {
+	local -a ctrls
+	local i st
+
+	echo "Running ${TEST_NAME}"
+	create_rescan_script
+
+	mapfile -t ctrls < <(_nvme_get_ctrl_list)
+	_run_fio_verify_io --filename="$TEST_DEV" --time_based &> "$FULL" &
+
+	for i in "${!ctrls[@]}"; do
+		rescan_controller "${ctrls[$i]}" "$i"
+	done
+
+	while true; do
+		wait -n &>/dev/null
+		st=$?
+		case $st in
+			127)
+				break
+				;;
+			0)
+				;;
+			*)
+				echo "child process exited with $st!"
+				;;
+		esac
+	done
+
+	echo "Test complete"
+}
diff --git a/tests/nvme/053.out b/tests/nvme/053.out
new file mode 100644
index 0000000..e8086ce
--- /dev/null
+++ b/tests/nvme/053.out
@@ -0,0 +1,2 @@
+Running nvme/053
+Test complete
diff --git a/tests/nvme/rc b/tests/nvme/rc
index b702a57..a877de3 100644
--- a/tests/nvme/rc
+++ b/tests/nvme/rc
@@ -192,6 +192,24 @@ _test_dev_nvme_nsid() {
 	cat "${TEST_DEV_SYSFS}/nsid"
 }
 
+_nvme_get_ctrl_list() {
+	local subsys
+	local c
+
+	subsys=$(readlink  "${TEST_DEV_SYSFS}/device/subsystem")
+	case $subsys in
+		*/nvme)
+			readlink -f "${TEST_DEV_SYSFS}/device"
+			;;
+		*/nvme-subsystem)
+			for c in "${TEST_DEV_SYSFS}"/device/nvme*; do
+				[[ -L "$c" ]] || continue
+				[[ -f "$c/dev" ]] && readlink -f "$c"
+			done
+			;;
+	esac
+}
+
 _nvme_calc_rand_io_size() {
 	local img_size_mb
 	local io_size_kb
-- 
2.46.0