Re: [PATCH 3/4] generic: test that linking into a directory fails with EDQUOT

["Darrick J. Wong" <djwong@kernel.org>]

On Wed, Apr 13, 2022 at 01:17:23AM +0800, Zorro Lang wrote:
> On Mon, Apr 11, 2022 at 03:54:48PM -0700, Darrick J. Wong wrote:
> > From: Darrick J. Wong <djwong@kernel.org>
> > 
> > Add a regression test to make sure that unprivileged userspace linking
> > into a directory fails with EDQUOT when the directory quota limits have
> > been exceeded.
> > 
> > Signed-off-by: Darrick J. Wong <djwong@kernel.org>
> > ---
> >  tests/generic/832     |   67 +++++++++++++++++++++++++++++++++++++++++++++++++
> >  tests/generic/832.out |    3 ++
> >  2 files changed, 70 insertions(+)
> >  create mode 100755 tests/generic/832
> >  create mode 100644 tests/generic/832.out
> > 
> > 
> > diff --git a/tests/generic/832 b/tests/generic/832
> > new file mode 100755
> > index 00000000..1190b795
> > --- /dev/null
> > +++ b/tests/generic/832
> > @@ -0,0 +1,67 @@
> > +#! /bin/bash
> > +# SPDX-License-Identifier: GPL-2.0
> > +# Copyright (c) 2022 Oracle.  All Rights Reserved.
> > +#
> > +# FS QA Test No. 832
> > +#
> > +# Ensure that unprivileged userspace hits EDQUOT while linking files into a
> > +# directory when the directory's quota limits have been exceeded.
> > +#
> > +# Regression test for commit:
> > +#
> > +# 871b9316e7a7 ("xfs: reserve quota for dir expansion when linking/unlinking files")
> > +#
> > +. ./common/preamble
> > +_begin_fstest auto quick quota
> > +
> > +# Import common functions.
> > +. ./common/filter
> > +. ./common/quota
> > +
> > +# real QA test starts here
> > +
> > +# Modify as appropriate.
> > +_supported_fs generic
> > +_require_quota
> > +_require_user
> > +_require_scratch
> > +
> > +_scratch_mkfs > "$seqres.full" 2>&1
> > +_qmount_option usrquota
> > +_qmount
> > +
> > +blocksize=$(_get_block_size $SCRATCH_MNT)
> > +scratchdir=$SCRATCH_MNT/dir
> > +scratchfile=$SCRATCH_MNT/file
> > +mkdir $scratchdir
> > +touch $scratchfile
> > +
> > +# Create a 2-block directory for our 1-block quota limit
> > +total_size=$((blocksize * 2))
> > +dirents=$((total_size / 255))
> > +
> > +for ((i = 0; i < dirents; i++)); do
> > +	name=$(printf "x%0254d" $i)
> > +	ln $scratchfile $scratchdir/$name
> > +done
> > +
> > +# Set a low quota hardlimit for an unprivileged uid and chown the files to it
> > +echo "set up quota" >> $seqres.full
> > +setquota -u $qa_user 0 "$((blocksize / 1024))" 0 0 $SCRATCH_MNT
> > +chown $qa_user $scratchdir $scratchfile
> > +repquota -upn $SCRATCH_MNT >> $seqres.full
> > +
> > +# Fail at appending the directory as qa_user to ensure quota enforcement works
> > +echo "fail quota" >> $seqres.full
> > +for ((i = 0; i < dirents; i++)); do
> > +	name=$(printf "y%0254d" $i)
> > +	su - "$qa_user" -c "ln $scratchfile $scratchdir/$name" 2>&1 | \
> 
> All looks good to me. Only one question about this "su -". Is the "-" necessary?
> I checked all cases in fstests, no one use "--login" when try to su to $qa_user.
> I'm not sure if "login $qa_user" will affect the testing, I just know it affect
> environment variables.

It's not strictly necessary since it's unlikely that qa_user="-luser",
but it seems like a Good Idea to prevent su cli option injection
attacks.

--D

> Thanks,
> Zorro
> 
> > +		_filter_scratch | sed -e 's/y[0-9]*/yXXX/g'
> > +	test "${PIPESTATUS[0]}" -ne 0 && break
> > +done
> > +repquota -upn $SCRATCH_MNT >> $seqres.full
> > +
> > +# success, all done
> > +echo Silence is golden
> > +status=0
> > +exit
> > diff --git a/tests/generic/832.out b/tests/generic/832.out
> > new file mode 100644
> > index 00000000..593afe8b
> > --- /dev/null
> > +++ b/tests/generic/832.out
> > @@ -0,0 +1,3 @@
> > +QA output created by 832
> > +ln: failed to create hard link 'SCRATCH_MNT/dir/yXXX': Disk quota exceeded
> > +Silence is golden
> > 
>