Skip to contentSkip to conversation

Four blktests patches

PATCH In review Bart Van Assche 3 participants 11 messages Aug 8, 2019 – Sep 4, 2019 Block layer

lore
Share & tools
lore
#0 lore raw

cover Four blktests patches

Hi Omar,

This series includes one improvement for the NVMe tests, two improvements for the NVMeOF-multipath tests and version two of the SRP test that triggers a SCSI reset while I/O is ongoing. Please consider these for the official blktests git repository.

Thanks,

Bart.

Bart Van Assche (4):
tests/nvme/rc: Modify the approach for disabling and re-enabling
Ctrl-C
tests/nvmeof-mp/rc: Make simulate_network_failure_loop() more robust tests/nvmeof-mp/rc: Make the NVMeOF multipath tests more reliable tests/srp/014: Add a test that triggers a SCSI reset while I/O is
ongoing

 tests/nvme/rc      |   4 +-
 tests/nvmeof-mp/rc |   7 +--
 tests/srp/014      | 107 +++++++++++++++++++++++++++++++++++++++++++++
 tests/srp/014.out  |   2 +
 4 files changed, 115 insertions(+), 5 deletions(-)
 create mode 100755 tests/srp/014
 create mode 100644 tests/srp/014.out
Signature · 1 line
2.22.0.770.g0f2c4a37fd-goog
#1 lore raw

PATCH 1/4 tests/nvme/rc: Modify the approach for disabling and re-enabling Ctrl-C

Avoid that the following error messages are reported when redirecting stdin:

stty: 'standard input': Inappropriate ioctl for device
stty: 'standard input': Inappropriate ioctl for device

Cc Logan Gunthorpe <[email protected]>
Cc Johannes Thumshirn <[email protected]>
Fixes a987b10bc179 ("nvme: Ensure all ports and subsystems are removed on cleanup")
---
 tests/nvme/rc | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
tests/nvme/rc +2 -2
index d4e18e635dea..40f0413d32d2 100644
@@ -36,7 +36,7 @@ _cleanup_nvmet() {
 	fi
 
 	# Don't let successive Ctrl-Cs interrupt the cleanup processes
-	stty -isig
+	trap '' SIGINT
 
 	shopt -s nullglob
 
@@ -66,7 +66,7 @@ _cleanup_nvmet() {
 	done
 
 	shopt -u nullglob
-	stty isig
+	trap SIGINT
 
 	modprobe -r nvme-loop 2>/dev/null
 	modprobe -r nvmet 2>/dev/null
Signature · 1 line
2.22.0.770.g0f2c4a37fd-goog
#2 lore raw

Re: [PATCH blktests 1/4] tests/nvme/rc: Modify the approach for disabling and re-enabling Ctrl-C

Quoted 18 lines · Bart Van Assche

Avoid that the following error messages are reported when redirecting stdin:

stty: 'standard input': Inappropriate ioctl for device
stty: 'standard input': Inappropriate ioctl for device

Cc Logan Gunthorpe <[email protected]>
Cc Johannes Thumshirn <[email protected]>
Fixes a987b10bc179 ("nvme: Ensure all ports and subsystems are removed on cleanup")
---
 tests/nvme/rc | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
tests/nvme/rc +0 -0
index d4e18e635dea..40f0413d32d2 100644

tests/nvme/rc _cleanup_nvmet() { · line 39

Quoted 5 lines
tests/nvme/rc +1 -1
@@ -36,7 +36,7 @@ _cleanup_nvmet() {
 	fi
 
 	# Don't let successive Ctrl-Cs interrupt the cleanup processes
-	stty -isig
+	trap '' SIGINT

Did you test this? Pretty sure I tried using trap and it didn't work, probably because it's already running inside a trapped SIGINT.

Maybe it'd be better to just ignore any errors stty produces and pipe to /dev/null?

Logan

#3 lore raw

Re: [PATCH blktests 1/4] tests/nvme/rc: Modify the approach for disabling and re-enabling Ctrl-C

Quoted 28 lines · Logan Gunthorpe
Quoted 18 lines · Bart Van Assche

Avoid that the following error messages are reported when redirecting stdin:

stty: 'standard input': Inappropriate ioctl for device
stty: 'standard input': Inappropriate ioctl for device

Cc Logan Gunthorpe <[email protected]>
Cc Johannes Thumshirn <[email protected]>
Fixes a987b10bc179 ("nvme: Ensure all ports and subsystems are removed on cleanup")
---
  tests/nvme/rc | 4 ++--
  1 file changed, 2 insertions(+), 2 deletions(-)
tests/nvme/rc +0 -0
index d4e18e635dea..40f0413d32d2 100644

tests/nvme/rc _cleanup_nvmet() { · line 39

Quoted 5 lines
tests/nvme/rc +1 -1
@@ -36,7 +36,7 @@ _cleanup_nvmet() {
  	fi
 
  	# Don't let successive Ctrl-Cs interrupt the cleanup processes
-	stty -isig
+	trap '' SIGINT

Did you test this? Pretty sure I tried using trap and it didn't work, probably because it's already running inside a trapped SIGINT.

Quoted 2 lines

Maybe it'd be better to just ignore any errors stty produces and pipe to /dev/null?

Hi Logan,

I don't think that redirecting the stty errors would be sufficient because Ctrl-C still works even if stdin, stdout and stderr are redirected. A command like sleep 60 </dev/null >&/dev/null can be interrupted with Ctrl-C but stty -isig >&/dev/null does not suppress Ctrl-C if stdin is redirected.

Are there other trap SIGINT statements in the blktests code? Does that mean that I overlooked something?

Thanks,

Bart.

Reviewed-by
#4 lore raw

Re: [PATCH blktests 1/4] tests/nvme/rc: Modify the approach for disabling and re-enabling Ctrl-C

Quoted 33 lines · Bart Van Assche
Quoted 29 lines · Logan Gunthorpe
Quoted 20 lines · Bart Van Assche

Avoid that the following error messages are reported when redirecting stdin:

stty: 'standard input': Inappropriate ioctl for device
stty: 'standard input': Inappropriate ioctl for device

Cc Logan Gunthorpe <[email protected]>
Cc Johannes Thumshirn <[email protected]>
Fixes a987b10bc179 ("nvme: Ensure all ports and subsystems are

removed on cleanup")

---
  tests/nvme/rc | 4 ++--

  1 file changed, 2 insertions(+), 2 deletions(-)
tests/nvme/rc +0 -0
index d4e18e635dea..40f0413d32d2 100644
@@ -36,7 +36,7 @@ _cleanup_nvmet() {
Quoted 4 lines

      fi
        # Don't let successive Ctrl-Cs interrupt the cleanup processes -    stty -isig
+    trap '' SIGINT

Did you test this? Pretty sure I tried using trap and it didn't work, probably because it's already running inside a trapped SIGINT.

Quoted 2 lines

Maybe it'd be better to just ignore any errors stty produces and pipe to /dev/null?

Quoted 7 lines

Hi Logan,

I don't think that redirecting the stty errors would be sufficient because Ctrl-C still works even if stdin, stdout and stderr are redirected. A command like sleep 60 </dev/null >&/dev/null can be interrupted with Ctrl-C but stty -isig >&/dev/null does not suppress Ctrl-C if stdin is redirected.

Ok, actually I just tested your change and it does work. I must have done something slightly differently when I first tried it (I think I added a handler which echoed messages which still let the SIGINTs through to child processes but it works with the null string).

So:

Quoted 2 lines

Are there other trap SIGINT statements in the blktests code? Does that mean that I overlooked something?

The main code traps EXIT which calls the cleanup handler that this trap then overrides SIGINT with, so I'm not really sure how they interact.

Thanks,

Logan

#5 lore raw

Re: [PATCH blktests 1/4] tests/nvme/rc: Modify the approach for disabling and re-enabling Ctrl-C

Quoted 2 lines · Logan Gunthorpe

The main code traps EXIT which calls the cleanup handler that this trap then overrides SIGINT with, so I'm not really sure how they interact.

Hi Logan,

You may want to know that there is already code in the blktests project that modifies a trap handler from inside a trap handler:

trap 'trap "" EXIT; teardown' EXIT

Bart.

#6 lore raw

PATCH 2/4 tests/nvmeof-mp/rc: Make simulate_network_failure_loop() more robust

In reply to Bart Van Assche

Avoid that the following is logged in the nvmeof-mp .full log files:

ls: cannot access '/sys/class/nvme/*/device/*/nvme0n1/reset_controller': No such file or directory
tests/nvmeof-mp/rc: line 124: : No such file or directory

---
 tests/nvmeof-mp/rc | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
tests/nvmeof-mp/rc +3 -2
index 9324dd1e8e4f..7352b1628cd3 100755
@@ -117,8 +117,9 @@ simulate_network_failure_loop() {
 	while [ $rc = 0 ]; do
 		sleep_until 5 ${deadline} || break
 		for d in $(held_by "$dev"); do
-			sf=$(ls -d /sys/class/nvme/*/device/*/"${d#/dev/}/reset_controller")
-			echo 1 > "$sf"
+			for sf in /sys/class/nvme/*/device/*/"${d#/dev/}/reset_controller"; do
+				[ -e "$sf" ] && echo 1 > "$sf"
+			done
 		done
 	done 2>>"$FULL"
 
Signature · 1 line
2.22.0.770.g0f2c4a37fd-goog
#7 lore raw

PATCH 3/4 tests/nvmeof-mp/rc: Make the NVMeOF multipath tests more reliable

In reply to Bart Van Assche

Fix the following failure for tests 002, 003, 004 and 011:

+tests/nvmeof-mp/rc: line 99: echo: write error: Operation already in progress

---
 tests/nvmeof-mp/rc | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tests/nvmeof-mp/rc +1 -1
index 7352b1628cd3..2493fcee12de 100755
@@ -124,7 +124,7 @@ simulate_network_failure_loop() {
 	done 2>>"$FULL"
 
 	for ((i=0;i<5;i++)); do
-		log_in && break
+		log_in 2>/dev/null && break
 		sleep 1
 	done
 }
Signature · 1 line
2.22.0.770.g0f2c4a37fd-goog
#8 lore raw

PATCH 4/4 tests/srp/014: Add a test that triggers a SCSI reset while I/O is ongoing

In reply to Bart Van Assche

This test triggers the task management function handling code in the SRP initiator and target drivers.

---
 tests/srp/014     | 107 ++++++++++++++++++++++++++++++++++++++++++++++
 tests/srp/014.out |   2 +
 2 files changed, 109 insertions(+)
 create mode 100755 tests/srp/014
 create mode 100644 tests/srp/014.out
tests/srp/014 new file +107 -0
new file mode 100755
index 000000000000..8ecd8a439a82
@@ -0,0 +1,107 @@
+#!/bin/bash
+# SPDX-License-Identifier: GPL-2.0+
+# Copyright (c) 2016-2018 Western Digital Corporation or its affiliates
+#
+# Submit SCSI resets while a fio --verify job is running. This is a regression
+# test for Linux kernel commit fd5614124406 ("scsi: RDMA/srp: Fix a
+# sleep-in-invalid-context bug") # v5.3.
+
+. tests/srp/rc
+
+DESCRIPTION="Run sg_reset while I/O is ongoing"
+TIMED=1
+
+# $1: SCSI device of which to change the state to running, e.g. sdc.
+make_running() {
+	local dev=$1 sp state
+
+	for sp in /sys/class/scsi_device/*/device/block/"$dev"; do
+		if [ -e "$sp" ]; then
+			break
+		else
+			return 1
+		fi
+	done
+	sp=$(dirname "$(dirname "$sp")")/state
+	# If the SCSI error handler changed the device state to offline,
+	# change the state back to running.
+	state=$(<"$sp")
+	if [ "$state" = offline ]; then
+		echo running > "$sp"
+		echo "$dev: state $state -> running" >>"$FULL"
+	else
+		echo "$dev: state $state" >>"$FULL"
+	fi
+}
+
+# $1: dm device to examine, e.g.
+# /dev/disk/by-id/dm-uuid-mpath-360014056e756c6c62300000000000000
+make_all_running() {
+	local d h dev=$1
+
+	while [ -L "$dev" ]; do
+		dev=$(realpath "$dev")
+	done
+	dev=${dev#/dev/}
+	for h in /sys/class/block/*/holders/"$dev"; do
+		[ -e "$h" ] || continue
+		d=$(basename "$(dirname "$(dirname "$h")")")
+		make_running "$d"
+	done
+}
+
+# $1: dm device to act on.
+set_running_loop() {
+	local dev="$1"
+
+	[ -e "$dev" ] || return $?
+	while true; do
+		sleep 1
+		make_all_running "$dev"
+	done
+	echo "set_running_loop $dev finished" >>"$FULL"
+}
+
+# $1: dm device to reset periodically; $2: how long the reset loop should run.
+sg_reset_loop() {
+	local cmd dev="$1" duration="$2" deadline i=0 reset_type
+
+	[ -e "$dev" ] || return $?
+	[ -n "$duration" ] || return $?
+	reset_type=(-d -b)
+	deadline=$(($(uptime_s) + duration))
+	while true; do
+		sleep_until 1 ${deadline} || break
+		cmd="sg_reset --no-esc ${reset_type[i++ % 2]} $dev"
+		{ echo "+ $cmd"; eval "$cmd"; } >>"$FULL" 2>&1
+	done
+	echo "sg_reset_loop $dev finished" >>"$FULL"
+}
+
+test_sg_reset() {
+	local dev fio_status m job
+
+	use_blk_mq y y || return $?
+	dev=$(get_bdev 0) || return $?
+	sg_reset_loop "$dev" "$TIMEOUT" &
+	# Redirect stderr to suppress the bash "Terminated" message.
+	set_running_loop "$dev" 2>/dev/null &
+	job=$!
+	run_fio --verify=md5 --rw=randwrite --bs=64K --loops=$((10**6)) \
+		--iodepth=16 --group_reporting --sync=1 --direct=1 \
+		--ioengine=libaio --runtime="${TIMEOUT}" \
+		--filename="$dev" --name=sg-reset-test --thread --numjobs=1 \
+		--output="${RESULTS_DIR}/srp/fio-output-014.txt" \
+		>>"$FULL"
+	fio_status=$?
+	kill "$job"
+	make_all_running "$dev"
+	wait
+	return $fio_status
+}
+
+test() {
+	: "${TIMEOUT:=30}"
+	trap 'trap "" EXIT; teardown' EXIT
+	setup && test_sg_reset && echo Passed
+}
tests/srp/014.out new file +2 -0
new file mode 100644
index 000000000000..5e25d8e8672d
@@ -0,0 +1,2 @@
+Configured SRP target driver
+Passed
Signature · 1 line
2.22.0.770.g0f2c4a37fd-goog
#9 lore raw
In reply to Bart Van Assche
Quoted 6 lines · Bart Van Assche

Hi Omar,

This series includes one improvement for the NVMe tests, two improvements for the NVMeOF-multipath tests and version two of the SRP test that triggers a SCSI reset while I/O is ongoing. Please consider these for the official blktests git repository.

(replying to my own e-mail)

Hi Omar,

This patch series was posted about one month ago. Feedback about how to proceed with this patch series is still welcome.

Thanks,

Bart.

#10 lore raw
Quoted 8 lines · Bart Van Assche
Quoted 6 lines · Bart Van Assche

Hi Omar,

This series includes one improvement for the NVMe tests, two improvements for the NVMeOF-multipath tests and version two of the SRP test that triggers a SCSI reset while I/O is ongoing. Please consider these for the official blktests git repository.

Quoted 6 lines

(replying to my own e-mail)

Hi Omar,

This patch series was posted about one month ago. Feedback about how to proceed with this patch series is still welcome.

I was actually just in the middle of merging it. Pushed now, thanks and sorry for the delay.