| Message ID | 20240605161815.34923-1-gulam.mohamed@oracle.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [V4,blktests] loop: Detect a race condition between loop detach and open | expand |
On 6/5/2024 9:18 AM, Gulam Mohamed wrote: > When one process opens a loop device partition and another process detaches > it, there will be a race condition due to which stale loop partitions are > created causing IO errors. This test will detect the race. > > Signed-off-by: Gulam Mohamed<gulam.mohamed@oracle.com> Looks good. Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com> -ck
On Jun 05, 2024 / 16:18, Gulam Mohamed wrote: > When one process opens a loop device partition and another process detaches > it, there will be a race condition due to which stale loop partitions are > created causing IO errors. This test will detect the race. > > Signed-off-by: Gulam Mohamed <gulam.mohamed@oracle.com> > --- > v4<-v3: > 1. Resolved formatting issues > 2. Using long options for commands instead of short options Thank you for the v4 patch. Looks good to me. Before applying it, I will wait for some more days in case anyone has other comments.
Hi Shinichiro, > -----Original Message----- > From: Shinichiro Kawasaki <shinichiro.kawasaki@wdc.com> > Sent: Thursday, June 6, 2024 11:33 AM > To: Gulam Mohamed <gulam.mohamed@oracle.com> > Cc: linux-block@vger.kernel.org; chaitanyak@nvidia.com > Subject: Re: [PATCH V4 blktests] loop: Detect a race condition between loop > detach and open > > On Jun 05, 2024 / 16:18, Gulam Mohamed wrote: > > When one process opens a loop device partition and another process > > detaches it, there will be a race condition due to which stale loop > > partitions are created causing IO errors. This test will detect the race. > > > > Signed-off-by: Gulam Mohamed <gulam.mohamed@oracle.com> > > --- > > v4<-v3: > > 1. Resolved formatting issues > > 2. Using long options for commands instead of short options > > Thank you for the v4 patch. Looks good to me. Before applying it, I will wait > for some more days in case anyone has other comments. Thank you very much for reviewing all these versions.
On Jun 06, 2024 / 09:11, Gulam Mohamed wrote: > Hi Shinichiro, > > > -----Original Message----- > > From: Shinichiro Kawasaki <shinichiro.kawasaki@wdc.com> > > Sent: Thursday, June 6, 2024 11:33 AM > > To: Gulam Mohamed <gulam.mohamed@oracle.com> > > Cc: linux-block@vger.kernel.org; chaitanyak@nvidia.com > > Subject: Re: [PATCH V4 blktests] loop: Detect a race condition between loop > > detach and open > > > > On Jun 05, 2024 / 16:18, Gulam Mohamed wrote: > > > When one process opens a loop device partition and another process > > > detaches it, there will be a race condition due to which stale loop > > > partitions are created causing IO errors. This test will detect the race. > > > > > > Signed-off-by: Gulam Mohamed <gulam.mohamed@oracle.com> > > > --- > > > v4<-v3: > > > 1. Resolved formatting issues > > > 2. Using long options for commands instead of short options > > > > Thank you for the v4 patch. Looks good to me. Before applying it, I will wait > > for some more days in case anyone has other comments. > > Thank you very much for reviewing all these versions. FYI, I've applied this patch. Please note that I folded in a minor change. When I took another look in the patch, I noticed that $TMPDIR was referred out of the test() function. Actually, $TMPDIR works only within test(), so I moved it into test(). Anyway, thank you for this work.
diff --git a/tests/loop/010 b/tests/loop/010 new file mode 100755 index 000000000000..f5d1bf1f9c18 --- /dev/null +++ b/tests/loop/010 @@ -0,0 +1,75 @@ +#!/bin/bash +# SPDX-License-Identifier: GPL-3.0+ +# Copyright (C) 2024, Oracle and/or its affiliates. +# +# Test to detect a race between loop detach and loop open which creates +# stale loop partitions when one process opens the loop partition and +# another process detaches the loop device. +# +. tests/loop/rc +DESCRIPTION="check stale loop partition" +TIMED=1 + +requires() { + _have_program parted + _have_program mkfs.xfs +} + +image_file="$TMPDIR/loopImg" + +create_loop() { + while true + do + loop_device="$(losetup --partscan --find --show "${image_file}")" + blkid /dev/loop0p1 >& /dev/null + done +} + +detach_loop() { + while true + do + if [ -e /dev/loop0 ]; then + losetup --detach /dev/loop0 >& /dev/null + fi + done +} + +test() { + echo "Running ${TEST_NAME}" + local loop_device + local create_pid + local detach_pid + + truncate --size 1G "${image_file}" + parted --align none --script "${image_file}" mklabel gpt + loop_device="$(losetup --partscan --find --show "${image_file}")" + parted --align none --script "${loop_device}" mkpart primary 64s 109051s + + udevadm settle + + if [ ! -e "${loop_device}" ]; then + return 1 + fi + + mkfs.xfs --force "${loop_device}p1" >& /dev/null + losetup --detach "${loop_device}" >& /dev/null + + create_loop & + create_pid=$! + detach_loop & + detach_pid=$! + + sleep "${TIMEOUT:-90}" + { + kill -9 $create_pid + kill -9 $detach_pid + wait + sleep 1 + } 2>/dev/null + + losetup --detach-all >& /dev/null + if _dmesg_since_test_start | grep --quiet "partition scan of loop0 failed (rc=-16)"; then + echo "Fail" + fi + echo "Test complete" +} diff --git a/tests/loop/010.out b/tests/loop/010.out new file mode 100644 index 000000000000..64a6aee00b8a --- /dev/null +++ b/tests/loop/010.out @@ -0,0 +1,2 @@ +Running loop/010 +Test complete
When one process opens a loop device partition and another process detaches it, there will be a race condition due to which stale loop partitions are created causing IO errors. This test will detect the race. Signed-off-by: Gulam Mohamed <gulam.mohamed@oracle.com> --- v4<-v3: 1. Resolved formatting issues 2. Using long options for commands instead of short options tests/loop/010 | 75 ++++++++++++++++++++++++++++++++++++++++++++++ tests/loop/010.out | 2 ++ 2 files changed, 77 insertions(+) create mode 100755 tests/loop/010 create mode 100644 tests/loop/010.out