[PATCH] generic: add test for XFS forkoff miscalcution on 32-bit platform

[PATCH] generic: add test for XFS forkoff miscalcution on 32-bit platform

[Gao Xiang]

There is a regression that recent XFS_LITINO(mp) update causes
xfs_attr_shortform_bytesfit() returns maxforkoff rather than 0.

Therefore, one result is
  "ASSERT(new_size <= XFS_IFORK_SIZE(ip, whichfork));"

Add a regression test in fstests generic to look after that since
the testcase itself isn't xfs-specific.

Signed-off-by: Gao Xiang <hsiangkao@redhat.com>
---
I have no usable 32-bit test environment to run xfstests, that is
what I have checked:
 - checked this new script can pass on x86_64;
 - manually ran script commands on i386 buildroot with problematic
   kernel and the filesystem got stuck on getfattr command.

 tests/generic/618     | 56 +++++++++++++++++++++++++++++++++++++++++++
 tests/generic/618.out |  4 ++++
 tests/generic/group   |  1 +
 3 files changed, 61 insertions(+)
 create mode 100755 tests/generic/618
 create mode 100644 tests/generic/618.out

diff --git a/tests/generic/618 b/tests/generic/618
new file mode 100755
index 00000000..997c6f75
--- /dev/null
+++ b/tests/generic/618
@@ -0,0 +1,56 @@
+#! /bin/bash
+# SPDX-License-Identifier: GPL-2.0
+# Copyright (c) 2020 Red Hat, Inc. All Rights Reserved.
+#
+# FS QA Test 618
+#
+# Verify that forkoff can be returned as 0 properly if it isn't
+# able to fit inline for XFS.
+# However, this test is fs-neutral and can be done quickly so
+# leave it in generic
+# This test verifies the problem fixed in kernel with commit
+# xxxxxxxxxxxx ("xfs: fix forkoff miscalculation related to XFS_LITINO(mp)")
+
+seq=`basename $0`
+seqres=$RESULT_DIR/$seq
+echo "QA output created by $seq"
+
+here=`pwd`
+tmp=/tmp/$$
+status=1	# failure is the default!
+trap "_cleanup; exit \$status" 0 1 2 3 15
+
+_cleanup()
+{
+	cd /
+	rm -f $tmp.*
+}
+
+# get standard environment, filters and checks
+. ./common/rc
+. ./common/filter
+. ./common/attr
+
+# remove previous $seqres.full before test
+rm -f $seqres.full
+
+# real QA test starts here
+
+_supported_fs generic
+_require_test
+_require_attrs user
+
+localfile="${TEST_DIR}/testfile"
+
+touch "${localfile}"
+"${SETFATTR_PROG}" -n user.0 -v "`seq 0 80`" "${localfile}"
+"${SETFATTR_PROG}" -n user.1 -v "`seq 0 80`" "${localfile}"
+
+# Make sure that changes are written to disk
+_test_cycle_mount
+
+# check getfattr result as well
+_getfattr --absolute-names -ebase64 -d $localfile | tail -n +2 | sort
+
+status=0
+exit
diff --git a/tests/generic/618.out b/tests/generic/618.out
new file mode 100644
index 00000000..848fdc58
--- /dev/null
+++ b/tests/generic/618.out
@@ -0,0 +1,4 @@
+QA output created by 618
+
+user.0=0sMAoxCjIKMwo0CjUKNgo3CjgKOQoxMAoxMQoxMgoxMwoxNAoxNQoxNgoxNwoxOAoxOQoyMAoyMQoyMgoyMwoyNAoyNQoyNgoyNwoyOAoyOQozMAozMQozMgozMwozNAozNQozNgozNwozOAozOQo0MAo0MQo0Mgo0Mwo0NAo0NQo0Ngo0Nwo0OAo0OQo1MAo1MQo1Mgo1Mwo1NAo1NQo1Ngo1Nwo1OAo1OQo2MAo2MQo2Mgo2Mwo2NAo2NQo2Ngo2Nwo2OAo2OQo3MAo3MQo3Mgo3Mwo3NAo3NQo3Ngo3Nwo3OAo3OQo4MA==
+user.1=0sMAoxCjIKMwo0CjUKNgo3CjgKOQoxMAoxMQoxMgoxMwoxNAoxNQoxNgoxNwoxOAoxOQoyMAoyMQoyMgoyMwoyNAoyNQoyNgoyNwoyOAoyOQozMAozMQozMgozMwozNAozNQozNgozNwozOAozOQo0MAo0MQo0Mgo0Mwo0NAo0NQo0Ngo0Nwo0OAo0OQo1MAo1MQo1Mgo1Mwo1NAo1NQo1Ngo1Nwo1OAo1OQo2MAo2MQo2Mgo2Mwo2NAo2NQo2Ngo2Nwo2OAo2OQo3MAo3MQo3Mgo3Mwo3NAo3NQo3Ngo3Nwo3OAo3OQo4MA==
diff --git a/tests/generic/group b/tests/generic/group
index 94e860b8..eca9d619 100644
--- a/tests/generic/group
+++ b/tests/generic/group
@@ -620,3 +620,4 @@
 615 auto rw
 616 auto rw io_uring stress
 617 auto rw io_uring stress
+618 auto quick attr
-- 
2.18.4

Re: [PATCH] generic: add test for XFS forkoff miscalcution on 32-bit platform

[Eryu Guan]

On Wed, Nov 18, 2020 at 02:02:58PM +0800, Gao Xiang wrote:
> There is a regression that recent XFS_LITINO(mp) update causes
> xfs_attr_shortform_bytesfit() returns maxforkoff rather than 0.
> 
> Therefore, one result is
>   "ASSERT(new_size <= XFS_IFORK_SIZE(ip, whichfork));"
> 
> Add a regression test in fstests generic to look after that since
> the testcase itself isn't xfs-specific.
> 
> Signed-off-by: Gao Xiang <hsiangkao@redhat.com>
> ---
> I have no usable 32-bit test environment to run xfstests, that is
> what I have checked:
>  - checked this new script can pass on x86_64;
>  - manually ran script commands on i386 buildroot with problematic
>    kernel and the filesystem got stuck on getfattr command.
> 
>  tests/generic/618     | 56 +++++++++++++++++++++++++++++++++++++++++++
>  tests/generic/618.out |  4 ++++
>  tests/generic/group   |  1 +
>  3 files changed, 61 insertions(+)
>  create mode 100755 tests/generic/618
>  create mode 100644 tests/generic/618.out
> 
> diff --git a/tests/generic/618 b/tests/generic/618
> new file mode 100755
> index 00000000..997c6f75
> --- /dev/null
> +++ b/tests/generic/618
> @@ -0,0 +1,56 @@
> +#! /bin/bash
> +# SPDX-License-Identifier: GPL-2.0
> +# Copyright (c) 2020 Red Hat, Inc. All Rights Reserved.
> +#
> +# FS QA Test 618
> +#
> +# Verify that forkoff can be returned as 0 properly if it isn't
> +# able to fit inline for XFS.
> +# However, this test is fs-neutral and can be done quickly so
> +# leave it in generic
> +# This test verifies the problem fixed in kernel with commit
> +# xxxxxxxxxxxx ("xfs: fix forkoff miscalculation related to XFS_LITINO(mp)")

Would you please re-post when the commit is upstream? With the commit ID
updated.

> +
> +seq=`basename $0`
> +seqres=$RESULT_DIR/$seq
> +echo "QA output created by $seq"
> +
> +here=`pwd`
> +tmp=/tmp/$$
> +status=1	# failure is the default!
> +trap "_cleanup; exit \$status" 0 1 2 3 15
> +
> +_cleanup()
> +{
> +	cd /
> +	rm -f $tmp.*
> +}
> +
> +# get standard environment, filters and checks
> +. ./common/rc
> +. ./common/filter
> +. ./common/attr
> +
> +# remove previous $seqres.full before test
> +rm -f $seqres.full
> +
> +# real QA test starts here
> +
> +_supported_fs generic
> +_require_test
> +_require_attrs user
> +
> +localfile="${TEST_DIR}/testfile"

Usually we use a testfile prefixed with $seq, e.g.

localfile=${TEST_DIR}/$seq.testfile

And remove it before test to avoid side effects from previous runs.

rm -f $localfile
touch $localfile

> +
> +touch "${localfile}"
> +"${SETFATTR_PROG}" -n user.0 -v "`seq 0 80`" "${localfile}"
> +"${SETFATTR_PROG}" -n user.1 -v "`seq 0 80`" "${localfile}"

I'd be better to add comments on why we need two user attrs and why we
need such long attr value.

> +
> +# Make sure that changes are written to disk
> +_test_cycle_mount
> +
> +# check getfattr result as well

Also, better to document the test failure behavior, e.g. kernel crash or
hung or just a getfattr failure.

Thanks,
Eryu

> +_getfattr --absolute-names -ebase64 -d $localfile | tail -n +2 | sort
> +
> +status=0
> +exit
> diff --git a/tests/generic/618.out b/tests/generic/618.out
> new file mode 100644
> index 00000000..848fdc58
> --- /dev/null
> +++ b/tests/generic/618.out
> @@ -0,0 +1,4 @@
> +QA output created by 618
> +
> +user.0=0sMAoxCjIKMwo0CjUKNgo3CjgKOQoxMAoxMQoxMgoxMwoxNAoxNQoxNgoxNwoxOAoxOQoyMAoyMQoyMgoyMwoyNAoyNQoyNgoyNwoyOAoyOQozMAozMQozMgozMwozNAozNQozNgozNwozOAozOQo0MAo0MQo0Mgo0Mwo0NAo0NQo0Ngo0Nwo0OAo0OQo1MAo1MQo1Mgo1Mwo1NAo1NQo1Ngo1Nwo1OAo1OQo2MAo2MQo2Mgo2Mwo2NAo2NQo2Ngo2Nwo2OAo2OQo3MAo3MQo3Mgo3Mwo3NAo3NQo3Ngo3Nwo3OAo3OQo4MA==
> +user.1=0sMAoxCjIKMwo0CjUKNgo3CjgKOQoxMAoxMQoxMgoxMwoxNAoxNQoxNgoxNwoxOAoxOQoyMAoyMQoyMgoyMwoyNAoyNQoyNgoyNwoyOAoyOQozMAozMQozMgozMwozNAozNQozNgozNwozOAozOQo0MAo0MQo0Mgo0Mwo0NAo0NQo0Ngo0Nwo0OAo0OQo1MAo1MQo1Mgo1Mwo1NAo1NQo1Ngo1Nwo1OAo1OQo2MAo2MQo2Mgo2Mwo2NAo2NQo2Ngo2Nwo2OAo2OQo3MAo3MQo3Mgo3Mwo3NAo3NQo3Ngo3Nwo3OAo3OQo4MA==
> diff --git a/tests/generic/group b/tests/generic/group
> index 94e860b8..eca9d619 100644
> --- a/tests/generic/group
> +++ b/tests/generic/group
> @@ -620,3 +620,4 @@
>  615 auto rw
>  616 auto rw io_uring stress
>  617 auto rw io_uring stress
> +618 auto quick attr
> -- 
> 2.18.4

Re: [PATCH] generic: add test for XFS forkoff miscalcution on 32-bit platform

[Gao Xiang]

Hi Eryu,

On Sun, Nov 22, 2020 at 10:46:33PM +0800, Eryu Guan wrote:
> On Wed, Nov 18, 2020 at 02:02:58PM +0800, Gao Xiang wrote:

...

> > +# xxxxxxxxxxxx ("xfs: fix forkoff miscalculation related to XFS_LITINO(mp)")
> 
> Would you please re-post when the commit is upstream? With the commit ID
> updated.

Sure will definitely do tomorrow.

> 
> > +
> > +seq=`basename $0`
> > +seqres=$RESULT_DIR/$seq
> > +echo "QA output created by $seq"
> > +
> > +here=`pwd`
> > +tmp=/tmp/$$
> > +status=1	# failure is the default!
> > +trap "_cleanup; exit \$status" 0 1 2 3 15
> > +
> > +_cleanup()
> > +{
> > +	cd /
> > +	rm -f $tmp.*
> > +}
> > +
> > +# get standard environment, filters and checks
> > +. ./common/rc
> > +. ./common/filter
> > +. ./common/attr
> > +
> > +# remove previous $seqres.full before test
> > +rm -f $seqres.full
> > +
> > +# real QA test starts here
> > +
> > +_supported_fs generic
> > +_require_test
> > +_require_attrs user
> > +
> > +localfile="${TEST_DIR}/testfile"
> 
> Usually we use a testfile prefixed with $seq, e.g.
> 
> localfile=${TEST_DIR}/$seq.testfile
> 
> And remove it before test to avoid side effects from previous runs.
> 
> rm -f $localfile
> touch $localfile

okay, will fix.

Actually I didn't take much time on this (just copy from
generic/611 and use $TEST_DIR instead.) 

> 
> > +
> > +touch "${localfile}"
> > +"${SETFATTR_PROG}" -n user.0 -v "`seq 0 80`" "${localfile}"
> > +"${SETFATTR_PROG}" -n user.1 -v "`seq 0 80`" "${localfile}"
> 
> I'd be better to add comments on why we need two user attrs and why we
> need such long attr value.

There is no specific reason of using "user attrs" and this
combination, just the example in
commit ada49d64fb35 ("xfs: fix forkoff miscalculation related to XFS_LITINO(mp)")

long xattrs which acrosses inline xattr shortform fork can
trigger this issue. As I said in the patch, the bug itself
is xfs-specific, but the testcase is generic.

> 
> > +
> > +# Make sure that changes are written to disk
> > +_test_cycle_mount
> > +
> > +# check getfattr result as well
> 
> Also, better to document the test failure behavior, e.g. kernel crash or
> hung or just a getfattr failure.

It depends, since ASSERT already fails in the second setfattr
and similar as
https://bugzilla.redhat.com/show_bug.cgi?id=1894177

if CONFIG_XFS_ASSERT_FATAL=n, what I saw was that hung on
the _getfattr line. But in any case, _getfattr won't success
with the correct result. So I don't think it needs to document
such unstable status.

Thanks,
Gao Xiang

> 
> Thanks,
> Eryu
> 
> > +_getfattr --absolute-names -ebase64 -d $localfile | tail -n +2 | sort
> > +
> > +status=0
> > +exit
> > diff --git a/tests/generic/618.out b/tests/generic/618.out
> > new file mode 100644
> > index 00000000..848fdc58
> > --- /dev/null
> > +++ b/tests/generic/618.out
> > @@ -0,0 +1,4 @@
> > +QA output created by 618
> > +
> > +user.0=0sMAoxCjIKMwo0CjUKNgo3CjgKOQoxMAoxMQoxMgoxMwoxNAoxNQoxNgoxNwoxOAoxOQoyMAoyMQoyMgoyMwoyNAoyNQoyNgoyNwoyOAoyOQozMAozMQozMgozMwozNAozNQozNgozNwozOAozOQo0MAo0MQo0Mgo0Mwo0NAo0NQo0Ngo0Nwo0OAo0OQo1MAo1MQo1Mgo1Mwo1NAo1NQo1Ngo1Nwo1OAo1OQo2MAo2MQo2Mgo2Mwo2NAo2NQo2Ngo2Nwo2OAo2OQo3MAo3MQo3Mgo3Mwo3NAo3NQo3Ngo3Nwo3OAo3OQo4MA==
> > +user.1=0sMAoxCjIKMwo0CjUKNgo3CjgKOQoxMAoxMQoxMgoxMwoxNAoxNQoxNgoxNwoxOAoxOQoyMAoyMQoyMgoyMwoyNAoyNQoyNgoyNwoyOAoyOQozMAozMQozMgozMwozNAozNQozNgozNwozOAozOQo0MAo0MQo0Mgo0Mwo0NAo0NQo0Ngo0Nwo0OAo0OQo1MAo1MQo1Mgo1Mwo1NAo1NQo1Ngo1Nwo1OAo1OQo2MAo2MQo2Mgo2Mwo2NAo2NQo2Ngo2Nwo2OAo2OQo3MAo3MQo3Mgo3Mwo3NAo3NQo3Ngo3Nwo3OAo3OQo4MA==
> > diff --git a/tests/generic/group b/tests/generic/group
> > index 94e860b8..eca9d619 100644
> > --- a/tests/generic/group
> > +++ b/tests/generic/group
> > @@ -620,3 +620,4 @@
> >  615 auto rw
> >  616 auto rw io_uring stress
> >  617 auto rw io_uring stress
> > +618 auto quick attr
> > -- 
> > 2.18.4
> 

[PATCH v2] generic: add test for XFS forkoff miscalcution on 32-bit platform

[Gao Xiang]

There is a regression that recent XFS_LITINO(mp) update causes
xfs_attr_shortform_bytesfit() returns maxforkoff rather than 0.

Therefore, one result is
  "ASSERT(new_size <= XFS_IFORK_SIZE(ip, whichfork));"

Add a regression test in fstests generic to look after that since
the testcase itself isn't xfs-specific.

Signed-off-by: Gao Xiang <hsiangkao@redhat.com>
---
changes since v1:
 - update commit id since the fix has been upstreamed (Eryu);
 - switch to use scratch_mkfs instead since fixed inode size is
   prefered to keep XFS_LITINO value can trigger the issue for
   v4 (412) and v5 (336); and no need to use TEST_DIR
   "rm localfile" way
 - refine some inlined comments.

 tests/generic/618     | 67 +++++++++++++++++++++++++++++++++++++++++++
 tests/generic/618.out |  4 +++
 tests/generic/group   |  1 +
 3 files changed, 72 insertions(+)
 create mode 100755 tests/generic/618
 create mode 100644 tests/generic/618.out

diff --git a/tests/generic/618 b/tests/generic/618
new file mode 100755
index 00000000..bdaa3874
--- /dev/null
+++ b/tests/generic/618
@@ -0,0 +1,67 @@
+#! /bin/bash
+# SPDX-License-Identifier: GPL-2.0
+# Copyright (c) 2020 Red Hat, Inc. All Rights Reserved.
+#
+# FS QA Test 618
+#
+# Verify that forkoff can be returned as 0 properly if it isn't
+# able to fit inline for XFS.
+# However, this test is fs-neutral and can be done quickly so
+# leave it in generic
+# This test verifies the problem fixed in kernel with commit
+# ada49d64fb35 ("xfs: fix forkoff miscalculation related to XFS_LITINO(mp)")
+
+seq=`basename $0`
+seqres=$RESULT_DIR/$seq
+echo "QA output created by $seq"
+
+here=`pwd`
+tmp=/tmp/$$
+status=1	# failure is the default!
+trap "_cleanup; exit \$status" 0 1 2 3 15
+
+_cleanup()
+{
+	cd /
+	rm -f $tmp.*
+}
+
+# get standard environment, filters and checks
+. ./common/rc
+. ./common/filter
+. ./common/attr
+
+# remove previous $seqres.full before test
+rm -f $seqres.full
+
+# real QA test starts here
+
+_supported_fs generic
+_require_scratch
+_require_attrs user
+
+# Use fixed inode size 512, so both v4 and v5 can be tested,
+# and also make sure the issue can be triggered if the default
+# inode size is changed later.
+[ $FSTYP = "xfs" ] && MKFS_OPTIONS="$MKFS_OPTIONS -i size=512"
+_scratch_mkfs > $seqres.full 2>&1
+_scratch_mount
+
+localfile="${SCRATCH_MNT}/testfile"
+touch $localfile
+
+# value cannot exceed XFS_ATTR_SF_ENTSIZE_MAX (256) or it will turn into
+# leaf form directly; the following combination can trigger the issue for
+# both v4 (XFS_LITINO = 412) & v5 (XFS_LITINO = 336) fses.
+"${SETFATTR_PROG}" -n user.0 -v "`seq 0 80`" "${localfile}"
+"${SETFATTR_PROG}" -n user.1 -v "`seq 0 80`" "${localfile}"
+
+# Make sure that changes are written to disk
+_scratch_cycle_mount
+
+# getfattr won't succeed with the expected result if fails
+_getfattr --absolute-names -ebase64 -d $localfile | tail -n +2 | sort
+
+_scratch_unmount
+status=0
+exit
diff --git a/tests/generic/618.out b/tests/generic/618.out
new file mode 100644
index 00000000..848fdc58
--- /dev/null
+++ b/tests/generic/618.out
@@ -0,0 +1,4 @@
+QA output created by 618
+
+user.0=0sMAoxCjIKMwo0CjUKNgo3CjgKOQoxMAoxMQoxMgoxMwoxNAoxNQoxNgoxNwoxOAoxOQoyMAoyMQoyMgoyMwoyNAoyNQoyNgoyNwoyOAoyOQozMAozMQozMgozMwozNAozNQozNgozNwozOAozOQo0MAo0MQo0Mgo0Mwo0NAo0NQo0Ngo0Nwo0OAo0OQo1MAo1MQo1Mgo1Mwo1NAo1NQo1Ngo1Nwo1OAo1OQo2MAo2MQo2Mgo2Mwo2NAo2NQo2Ngo2Nwo2OAo2OQo3MAo3MQo3Mgo3Mwo3NAo3NQo3Ngo3Nwo3OAo3OQo4MA==
+user.1=0sMAoxCjIKMwo0CjUKNgo3CjgKOQoxMAoxMQoxMgoxMwoxNAoxNQoxNgoxNwoxOAoxOQoyMAoyMQoyMgoyMwoyNAoyNQoyNgoyNwoyOAoyOQozMAozMQozMgozMwozNAozNQozNgozNwozOAozOQo0MAo0MQo0Mgo0Mwo0NAo0NQo0Ngo0Nwo0OAo0OQo1MAo1MQo1Mgo1Mwo1NAo1NQo1Ngo1Nwo1OAo1OQo2MAo2MQo2Mgo2Mwo2NAo2NQo2Ngo2Nwo2OAo2OQo3MAo3MQo3Mgo3Mwo3NAo3NQo3Ngo3Nwo3OAo3OQo4MA==
diff --git a/tests/generic/group b/tests/generic/group
index 94e860b8..eca9d619 100644
--- a/tests/generic/group
+++ b/tests/generic/group
@@ -620,3 +620,4 @@
 615 auto rw
 616 auto rw io_uring stress
 617 auto rw io_uring stress
+618 auto quick attr
-- 
2.18.4

Re: [PATCH v2] generic: add test for XFS forkoff miscalcution on 32-bit platform

[Darrick J. Wong]

On Mon, Nov 23, 2020 at 04:20:47PM +0800, Gao Xiang wrote:
> There is a regression that recent XFS_LITINO(mp) update causes
> xfs_attr_shortform_bytesfit() returns maxforkoff rather than 0.
> 
> Therefore, one result is
>   "ASSERT(new_size <= XFS_IFORK_SIZE(ip, whichfork));"
> 
> Add a regression test in fstests generic to look after that since
> the testcase itself isn't xfs-specific.
> 
> Signed-off-by: Gao Xiang <hsiangkao@redhat.com>
> ---
> changes since v1:
>  - update commit id since the fix has been upstreamed (Eryu);
>  - switch to use scratch_mkfs instead since fixed inode size is
>    prefered to keep XFS_LITINO value can trigger the issue for
>    v4 (412) and v5 (336); and no need to use TEST_DIR
>    "rm localfile" way
>  - refine some inlined comments.
> 
>  tests/generic/618     | 67 +++++++++++++++++++++++++++++++++++++++++++
>  tests/generic/618.out |  4 +++
>  tests/generic/group   |  1 +
>  3 files changed, 72 insertions(+)
>  create mode 100755 tests/generic/618
>  create mode 100644 tests/generic/618.out
> 
> diff --git a/tests/generic/618 b/tests/generic/618
> new file mode 100755
> index 00000000..bdaa3874
> --- /dev/null
> +++ b/tests/generic/618
> @@ -0,0 +1,67 @@
> +#! /bin/bash
> +# SPDX-License-Identifier: GPL-2.0
> +# Copyright (c) 2020 Red Hat, Inc. All Rights Reserved.
> +#
> +# FS QA Test 618
> +#
> +# Verify that forkoff can be returned as 0 properly if it isn't
> +# able to fit inline for XFS.
> +# However, this test is fs-neutral and can be done quickly so
> +# leave it in generic
> +# This test verifies the problem fixed in kernel with commit
> +# ada49d64fb35 ("xfs: fix forkoff miscalculation related to XFS_LITINO(mp)")
> +
> +seq=`basename $0`
> +seqres=$RESULT_DIR/$seq
> +echo "QA output created by $seq"
> +
> +here=`pwd`
> +tmp=/tmp/$$
> +status=1	# failure is the default!
> +trap "_cleanup; exit \$status" 0 1 2 3 15
> +
> +_cleanup()
> +{
> +	cd /
> +	rm -f $tmp.*
> +}
> +
> +# get standard environment, filters and checks
> +. ./common/rc
> +. ./common/filter
> +. ./common/attr
> +
> +# remove previous $seqres.full before test
> +rm -f $seqres.full
> +
> +# real QA test starts here
> +
> +_supported_fs generic
> +_require_scratch
> +_require_attrs user

Does this require

_require_no_xfs_bug_on_assert ?

> +
> +# Use fixed inode size 512, so both v4 and v5 can be tested,
> +# and also make sure the issue can be triggered if the default
> +# inode size is changed later.
> +[ $FSTYP = "xfs" ] && MKFS_OPTIONS="$MKFS_OPTIONS -i size=512"
> +_scratch_mkfs > $seqres.full 2>&1
> +_scratch_mount
> +
> +localfile="${SCRATCH_MNT}/testfile"
> +touch $localfile
> +
> +# value cannot exceed XFS_ATTR_SF_ENTSIZE_MAX (256) or it will turn into
> +# leaf form directly; the following combination can trigger the issue for
> +# both v4 (XFS_LITINO = 412) & v5 (XFS_LITINO = 336) fses.
> +"${SETFATTR_PROG}" -n user.0 -v "`seq 0 80`" "${localfile}"
> +"${SETFATTR_PROG}" -n user.1 -v "`seq 0 80`" "${localfile}"

It's probably worth mentioning that the second setattr causes an integer
underflow that is incorrectly typecast, leading to the assert
triggering.  Otherwise this seems reasonable to me.

--D

> +
> +# Make sure that changes are written to disk
> +_scratch_cycle_mount
> +
> +# getfattr won't succeed with the expected result if fails
> +_getfattr --absolute-names -ebase64 -d $localfile | tail -n +2 | sort
> +
> +_scratch_unmount
> +status=0
> +exit
> diff --git a/tests/generic/618.out b/tests/generic/618.out
> new file mode 100644
> index 00000000..848fdc58
> --- /dev/null
> +++ b/tests/generic/618.out
> @@ -0,0 +1,4 @@
> +QA output created by 618
> +
> +user.0=0sMAoxCjIKMwo0CjUKNgo3CjgKOQoxMAoxMQoxMgoxMwoxNAoxNQoxNgoxNwoxOAoxOQoyMAoyMQoyMgoyMwoyNAoyNQoyNgoyNwoyOAoyOQozMAozMQozMgozMwozNAozNQozNgozNwozOAozOQo0MAo0MQo0Mgo0Mwo0NAo0NQo0Ngo0Nwo0OAo0OQo1MAo1MQo1Mgo1Mwo1NAo1NQo1Ngo1Nwo1OAo1OQo2MAo2MQo2Mgo2Mwo2NAo2NQo2Ngo2Nwo2OAo2OQo3MAo3MQo3Mgo3Mwo3NAo3NQo3Ngo3Nwo3OAo3OQo4MA==
> +user.1=0sMAoxCjIKMwo0CjUKNgo3CjgKOQoxMAoxMQoxMgoxMwoxNAoxNQoxNgoxNwoxOAoxOQoyMAoyMQoyMgoyMwoyNAoyNQoyNgoyNwoyOAoyOQozMAozMQozMgozMwozNAozNQozNgozNwozOAozOQo0MAo0MQo0Mgo0Mwo0NAo0NQo0Ngo0Nwo0OAo0OQo1MAo1MQo1Mgo1Mwo1NAo1NQo1Ngo1Nwo1OAo1OQo2MAo2MQo2Mgo2Mwo2NAo2NQo2Ngo2Nwo2OAo2OQo3MAo3MQo3Mgo3Mwo3NAo3NQo3Ngo3Nwo3OAo3OQo4MA==
> diff --git a/tests/generic/group b/tests/generic/group
> index 94e860b8..eca9d619 100644
> --- a/tests/generic/group
> +++ b/tests/generic/group
> @@ -620,3 +620,4 @@
>  615 auto rw
>  616 auto rw io_uring stress
>  617 auto rw io_uring stress
> +618 auto quick attr
> -- 
> 2.18.4
> 

Re: [PATCH v2] generic: add test for XFS forkoff miscalcution on 32-bit platform

[Gao Xiang]

Hi Darrick,

On Mon, Nov 23, 2020 at 10:24:00AM -0800, Darrick J. Wong wrote:
> On Mon, Nov 23, 2020 at 04:20:47PM +0800, Gao Xiang wrote:

...

> > +_supported_fs generic
> > +_require_scratch
> > +_require_attrs user
> 
> Does this require
> 
> _require_no_xfs_bug_on_assert ?
> 

For the sake of harmless testcase, I think that is fine and I will
add it in the next version as
[ $FSTYP = "xfs" ] && _require_no_xfs_bug_on_assert


To clarify the synotom:
if bug_on_assert is on, it will bug_on on the second setfattr,
_scratch_cycle_mount will fail due to
umount: can't unmount /tmp/mnt: Device or resource busy

if _scratch_cycle_mount is removed, _getfattr will hung due to
(I think) unbalanced xfs_ilock(dp, XFS_ILOCK_EXCL) in xfs_attr_set()
and xfs_attr_get() will take the lock again.


if bug_on_assert is off, it will fail on _scratch_cycle_mount due to
log recovery replay fail when mounting as below

[  856.786226] XFS (sda): Mounting V5 Filesystem                       
[  856.802704] XFS (sda): Starting recovery (logdev: internal)                                                                                                                                                    
[  856.806362] XFS: Assertion failed: len <= XFS_DFORK_ASIZE(dip, mp), file: fs/xfs/xfs_inode_item_recover.c, line: 426

but if the line of _scratch_cycle_mount is removed, _getfattr will
success due to
	ifp->if_u1.if_data = krealloc(ifp->if_u1.if_data, roundup(new_size, 4),
				      GFP_NOFS | __GFP_NOFAIL);
	ifp->if_bytes = new_size;
in xfs_idata_realloc(), and xfs_attr_shortform_getvalue() will
temporary ok. yet this testcase have _scratch_cycle_mount anyway.

In any case, it's fine with _getfattr after _scratch_cycle_mount
for this testcase.

And "bug_on_assert is off" does no harm to the system stablity itself
so I think that's better.

> > +
> > +# Use fixed inode size 512, so both v4 and v5 can be tested,
> > +# and also make sure the issue can be triggered if the default
> > +# inode size is changed later.
> > +[ $FSTYP = "xfs" ] && MKFS_OPTIONS="$MKFS_OPTIONS -i size=512"
> > +_scratch_mkfs > $seqres.full 2>&1
> > +_scratch_mount
> > +
> > +localfile="${SCRATCH_MNT}/testfile"
> > +touch $localfile
> > +
> > +# value cannot exceed XFS_ATTR_SF_ENTSIZE_MAX (256) or it will turn into
> > +# leaf form directly; the following combination can trigger the issue for
> > +# both v4 (XFS_LITINO = 412) & v5 (XFS_LITINO = 336) fses.
> > +"${SETFATTR_PROG}" -n user.0 -v "`seq 0 80`" "${localfile}"
> > +"${SETFATTR_PROG}" -n user.1 -v "`seq 0 80`" "${localfile}"
> 
> It's probably worth mentioning that the second setattr causes an integer
> underflow that is incorrectly typecast, leading to the assert
> triggering.  Otherwise this seems reasonable to me.

Ok, will try to add in the next version as well.

Thanks,
Gao Xiang

> 
> --D
> 
> > +
> > +# Make sure that changes are written to disk
> > +_scratch_cycle_mount
> > +
> > +# getfattr won't succeed with the expected result if fails
> > +_getfattr --absolute-names -ebase64 -d $localfile | tail -n +2 | sort
> > +
> > +_scratch_unmount
> > +status=0
> > +exit
> > diff --git a/tests/generic/618.out b/tests/generic/618.out
> > new file mode 100644
> > index 00000000..848fdc58
> > --- /dev/null
> > +++ b/tests/generic/618.out
> > @@ -0,0 +1,4 @@
> > +QA output created by 618
> > +
> > +user.0=0sMAoxCjIKMwo0CjUKNgo3CjgKOQoxMAoxMQoxMgoxMwoxNAoxNQoxNgoxNwoxOAoxOQoyMAoyMQoyMgoyMwoyNAoyNQoyNgoyNwoyOAoyOQozMAozMQozMgozMwozNAozNQozNgozNwozOAozOQo0MAo0MQo0Mgo0Mwo0NAo0NQo0Ngo0Nwo0OAo0OQo1MAo1MQo1Mgo1Mwo1NAo1NQo1Ngo1Nwo1OAo1OQo2MAo2MQo2Mgo2Mwo2NAo2NQo2Ngo2Nwo2OAo2OQo3MAo3MQo3Mgo3Mwo3NAo3NQo3Ngo3Nwo3OAo3OQo4MA==
> > +user.1=0sMAoxCjIKMwo0CjUKNgo3CjgKOQoxMAoxMQoxMgoxMwoxNAoxNQoxNgoxNwoxOAoxOQoyMAoyMQoyMgoyMwoyNAoyNQoyNgoyNwoyOAoyOQozMAozMQozMgozMwozNAozNQozNgozNwozOAozOQo0MAo0MQo0Mgo0Mwo0NAo0NQo0Ngo0Nwo0OAo0OQo1MAo1MQo1Mgo1Mwo1NAo1NQo1Ngo1Nwo1OAo1OQo2MAo2MQo2Mgo2Mwo2NAo2NQo2Ngo2Nwo2OAo2OQo3MAo3MQo3Mgo3Mwo3NAo3NQo3Ngo3Nwo3OAo3OQo4MA==
> > diff --git a/tests/generic/group b/tests/generic/group
> > index 94e860b8..eca9d619 100644
> > --- a/tests/generic/group
> > +++ b/tests/generic/group
> > @@ -620,3 +620,4 @@
> >  615 auto rw
> >  616 auto rw io_uring stress
> >  617 auto rw io_uring stress
> > +618 auto quick attr
> > -- 
> > 2.18.4
> > 
> 

[PATCH v3] generic: add test for XFS forkoff miscalcution on 32-bit platform

[Gao Xiang]

There is a regression that recent XFS_LITINO(mp) update causes
xfs_attr_shortform_bytesfit() returns maxforkoff rather than 0.

Therefore, one result is
  "ASSERT(new_size <= XFS_IFORK_SIZE(ip, whichfork));"

Add a regression test in fstests generic to look after that since
the testcase itself isn't xfs-specific.

Signed-off-by: Gao Xiang <hsiangkao@redhat.com>
---
changes since v2:
 - "_require_no_xfs_bug_on_assert" to avoid crashing the system (Darrick);
 - refine a commit for more details (Darrick)

 tests/generic/618     | 75 +++++++++++++++++++++++++++++++++++++++++++
 tests/generic/618.out |  4 +++
 tests/generic/group   |  1 +
 3 files changed, 80 insertions(+)
 create mode 100755 tests/generic/618
 create mode 100644 tests/generic/618.out

diff --git a/tests/generic/618 b/tests/generic/618
new file mode 100755
index 00000000..f1c1605e
--- /dev/null
+++ b/tests/generic/618
@@ -0,0 +1,75 @@
+#! /bin/bash
+# SPDX-License-Identifier: GPL-2.0
+# Copyright (c) 2020 Red Hat, Inc. All Rights Reserved.
+#
+# FS QA Test 618
+#
+# Verify that forkoff can be returned as 0 properly if it isn't
+# able to fit inline for XFS.
+# However, this test is fs-neutral and can be done quickly so
+# leave it in generic
+# This test verifies the problem fixed in kernel with commit
+# ada49d64fb35 ("xfs: fix forkoff miscalculation related to XFS_LITINO(mp)")
+
+seq=`basename $0`
+seqres=$RESULT_DIR/$seq
+echo "QA output created by $seq"
+
+here=`pwd`
+tmp=/tmp/$$
+status=1	# failure is the default!
+trap "_cleanup; exit \$status" 0 1 2 3 15
+
+_cleanup()
+{
+	cd /
+	rm -f $tmp.*
+}
+
+# get standard environment, filters and checks
+. ./common/rc
+. ./common/filter
+. ./common/attr
+
+# remove previous $seqres.full before test
+rm -f $seqres.full
+
+# real QA test starts here
+
+_supported_fs generic
+_require_scratch
+_require_attrs user
+
+if [ $FSTYP = "xfs" ]; then
+	# avoid crashing the system if possible
+	_require_no_xfs_bug_on_assert
+
+	# Use fixed inode size 512, so both v4 and v5 can be tested,
+	# and also make sure the issue can be triggered if the default
+	# inode size is changed later.
+	MKFS_OPTIONS="$MKFS_OPTIONS -i size=512"
+fi
+
+_scratch_mkfs > $seqres.full 2>&1
+_scratch_mount
+
+localfile="${SCRATCH_MNT}/testfile"
+touch $localfile
+
+# value cannot exceed XFS_ATTR_SF_ENTSIZE_MAX (256) or it will turn into leaf
+# form directly; the following combination can trigger the issue for both v4
+# (XFS_LITINO = 412) & v5 (XFS_LITINO = 336) fses, in details the 2nd setattr
+# causes an integer underflow that is incorrectly typecast, leading to the
+# assert triggering.
+"${SETFATTR_PROG}" -n user.0 -v "`seq 0 80`" "${localfile}"
+"${SETFATTR_PROG}" -n user.1 -v "`seq 0 80`" "${localfile}"
+
+# Make sure that changes are written to disk
+_scratch_cycle_mount
+
+# getfattr won't succeed with the expected result if fails
+_getfattr --absolute-names -ebase64 -d $localfile | tail -n +2 | sort
+
+_scratch_unmount
+status=0
+exit
diff --git a/tests/generic/618.out b/tests/generic/618.out
new file mode 100644
index 00000000..848fdc58
--- /dev/null
+++ b/tests/generic/618.out
@@ -0,0 +1,4 @@
+QA output created by 618
+
+user.0=0sMAoxCjIKMwo0CjUKNgo3CjgKOQoxMAoxMQoxMgoxMwoxNAoxNQoxNgoxNwoxOAoxOQoyMAoyMQoyMgoyMwoyNAoyNQoyNgoyNwoyOAoyOQozMAozMQozMgozMwozNAozNQozNgozNwozOAozOQo0MAo0MQo0Mgo0Mwo0NAo0NQo0Ngo0Nwo0OAo0OQo1MAo1MQo1Mgo1Mwo1NAo1NQo1Ngo1Nwo1OAo1OQo2MAo2MQo2Mgo2Mwo2NAo2NQo2Ngo2Nwo2OAo2OQo3MAo3MQo3Mgo3Mwo3NAo3NQo3Ngo3Nwo3OAo3OQo4MA==
+user.1=0sMAoxCjIKMwo0CjUKNgo3CjgKOQoxMAoxMQoxMgoxMwoxNAoxNQoxNgoxNwoxOAoxOQoyMAoyMQoyMgoyMwoyNAoyNQoyNgoyNwoyOAoyOQozMAozMQozMgozMwozNAozNQozNgozNwozOAozOQo0MAo0MQo0Mgo0Mwo0NAo0NQo0Ngo0Nwo0OAo0OQo1MAo1MQo1Mgo1Mwo1NAo1NQo1Ngo1Nwo1OAo1OQo2MAo2MQo2Mgo2Mwo2NAo2NQo2Ngo2Nwo2OAo2OQo3MAo3MQo3Mgo3Mwo3NAo3NQo3Ngo3Nwo3OAo3OQo4MA==
diff --git a/tests/generic/group b/tests/generic/group
index 94e860b8..eca9d619 100644
--- a/tests/generic/group
+++ b/tests/generic/group
@@ -620,3 +620,4 @@
 615 auto rw
 616 auto rw io_uring stress
 617 auto rw io_uring stress
+618 auto quick attr
-- 
2.18.4

Re: [PATCH v3] generic: add test for XFS forkoff miscalcution on 32-bit platform

[Darrick J. Wong]

On Tue, Nov 24, 2020 at 06:11:45PM +0800, Gao Xiang wrote:
> There is a regression that recent XFS_LITINO(mp) update causes
> xfs_attr_shortform_bytesfit() returns maxforkoff rather than 0.
> 
> Therefore, one result is
>   "ASSERT(new_size <= XFS_IFORK_SIZE(ip, whichfork));"
> 
> Add a regression test in fstests generic to look after that since
> the testcase itself isn't xfs-specific.
> 
> Signed-off-by: Gao Xiang <hsiangkao@redhat.com>

Looks good to me,
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>

--D

> ---
> changes since v2:
>  - "_require_no_xfs_bug_on_assert" to avoid crashing the system (Darrick);
>  - refine a commit for more details (Darrick)
> 
>  tests/generic/618     | 75 +++++++++++++++++++++++++++++++++++++++++++
>  tests/generic/618.out |  4 +++
>  tests/generic/group   |  1 +
>  3 files changed, 80 insertions(+)
>  create mode 100755 tests/generic/618
>  create mode 100644 tests/generic/618.out
> 
> diff --git a/tests/generic/618 b/tests/generic/618
> new file mode 100755
> index 00000000..f1c1605e
> --- /dev/null
> +++ b/tests/generic/618
> @@ -0,0 +1,75 @@
> +#! /bin/bash
> +# SPDX-License-Identifier: GPL-2.0
> +# Copyright (c) 2020 Red Hat, Inc. All Rights Reserved.
> +#
> +# FS QA Test 618
> +#
> +# Verify that forkoff can be returned as 0 properly if it isn't
> +# able to fit inline for XFS.
> +# However, this test is fs-neutral and can be done quickly so
> +# leave it in generic
> +# This test verifies the problem fixed in kernel with commit
> +# ada49d64fb35 ("xfs: fix forkoff miscalculation related to XFS_LITINO(mp)")
> +
> +seq=`basename $0`
> +seqres=$RESULT_DIR/$seq
> +echo "QA output created by $seq"
> +
> +here=`pwd`
> +tmp=/tmp/$$
> +status=1	# failure is the default!
> +trap "_cleanup; exit \$status" 0 1 2 3 15
> +
> +_cleanup()
> +{
> +	cd /
> +	rm -f $tmp.*
> +}
> +
> +# get standard environment, filters and checks
> +. ./common/rc
> +. ./common/filter
> +. ./common/attr
> +
> +# remove previous $seqres.full before test
> +rm -f $seqres.full
> +
> +# real QA test starts here
> +
> +_supported_fs generic
> +_require_scratch
> +_require_attrs user
> +
> +if [ $FSTYP = "xfs" ]; then
> +	# avoid crashing the system if possible
> +	_require_no_xfs_bug_on_assert
> +
> +	# Use fixed inode size 512, so both v4 and v5 can be tested,
> +	# and also make sure the issue can be triggered if the default
> +	# inode size is changed later.
> +	MKFS_OPTIONS="$MKFS_OPTIONS -i size=512"
> +fi
> +
> +_scratch_mkfs > $seqres.full 2>&1
> +_scratch_mount
> +
> +localfile="${SCRATCH_MNT}/testfile"
> +touch $localfile
> +
> +# value cannot exceed XFS_ATTR_SF_ENTSIZE_MAX (256) or it will turn into leaf
> +# form directly; the following combination can trigger the issue for both v4
> +# (XFS_LITINO = 412) & v5 (XFS_LITINO = 336) fses, in details the 2nd setattr
> +# causes an integer underflow that is incorrectly typecast, leading to the
> +# assert triggering.
> +"${SETFATTR_PROG}" -n user.0 -v "`seq 0 80`" "${localfile}"
> +"${SETFATTR_PROG}" -n user.1 -v "`seq 0 80`" "${localfile}"
> +
> +# Make sure that changes are written to disk
> +_scratch_cycle_mount
> +
> +# getfattr won't succeed with the expected result if fails
> +_getfattr --absolute-names -ebase64 -d $localfile | tail -n +2 | sort
> +
> +_scratch_unmount
> +status=0
> +exit
> diff --git a/tests/generic/618.out b/tests/generic/618.out
> new file mode 100644
> index 00000000..848fdc58
> --- /dev/null
> +++ b/tests/generic/618.out
> @@ -0,0 +1,4 @@
> +QA output created by 618
> +
> +user.0=0sMAoxCjIKMwo0CjUKNgo3CjgKOQoxMAoxMQoxMgoxMwoxNAoxNQoxNgoxNwoxOAoxOQoyMAoyMQoyMgoyMwoyNAoyNQoyNgoyNwoyOAoyOQozMAozMQozMgozMwozNAozNQozNgozNwozOAozOQo0MAo0MQo0Mgo0Mwo0NAo0NQo0Ngo0Nwo0OAo0OQo1MAo1MQo1Mgo1Mwo1NAo1NQo1Ngo1Nwo1OAo1OQo2MAo2MQo2Mgo2Mwo2NAo2NQo2Ngo2Nwo2OAo2OQo3MAo3MQo3Mgo3Mwo3NAo3NQo3Ngo3Nwo3OAo3OQo4MA==
> +user.1=0sMAoxCjIKMwo0CjUKNgo3CjgKOQoxMAoxMQoxMgoxMwoxNAoxNQoxNgoxNwoxOAoxOQoyMAoyMQoyMgoyMwoyNAoyNQoyNgoyNwoyOAoyOQozMAozMQozMgozMwozNAozNQozNgozNwozOAozOQo0MAo0MQo0Mgo0Mwo0NAo0NQo0Ngo0Nwo0OAo0OQo1MAo1MQo1Mgo1Mwo1NAo1NQo1Ngo1Nwo1OAo1OQo2MAo2MQo2Mgo2Mwo2NAo2NQo2Ngo2Nwo2OAo2OQo3MAo3MQo3Mgo3Mwo3NAo3NQo3Ngo3Nwo3OAo3OQo4MA==
> diff --git a/tests/generic/group b/tests/generic/group
> index 94e860b8..eca9d619 100644
> --- a/tests/generic/group
> +++ b/tests/generic/group
> @@ -620,3 +620,4 @@
>  615 auto rw
>  616 auto rw io_uring stress
>  617 auto rw io_uring stress
> +618 auto quick attr
> -- 
> 2.18.4
>