On Mon, Apr 11, 2022 at 03:55:08PM -0700, Darrick J. Wong wrote: > From: Darrick J. Wong <djwong@xxxxxxxxxx> > > Add a new regression test for a stack corruption problem uncovered in > the mkfs config file parsing code. > > Signed-off-by: Darrick J. Wong <djwong@xxxxxxxxxx> > --- Good to me, Reviewed-by: Zorro Lang <zlang@xxxxxxxxxx> > tests/xfs/831 | 68 +++++++++++++++++++++++++++++++++++++++++++++++++++++ > tests/xfs/831.out | 2 ++ > 2 files changed, 70 insertions(+) > create mode 100755 tests/xfs/831 > create mode 100644 tests/xfs/831.out > > > diff --git a/tests/xfs/831 b/tests/xfs/831 > new file mode 100755 > index 00000000..a73f14ff > --- /dev/null > +++ b/tests/xfs/831 > @@ -0,0 +1,68 @@ > +#! /bin/bash > +# SPDX-License-Identifier: GPL-2.0 > +# Copyright (c) 2022 Oracle. All Rights Reserved. > +# > +# FS QA Test 831 > +# > +# Regression test for xfsprogs commit: > +# > +# 99c78777 ("mkfs: prevent corruption of passed-in suboption string values") > +# > +. ./common/preamble > +_begin_fstest auto quick mkfs > + > +_cleanup() > +{ > + rm -f $TEST_DIR/fubar.img > + cd / > + rm -r -f $tmp.* > +} > + > +# Import common functions. > +# . ./common/filter > + > +# real QA test starts here > + > +# Modify as appropriate. > +_supported_fs xfs > +_require_test > +_require_xfs_mkfs_cfgfile > + > +# Set up a configuration file with an exact block size and log stripe unit > +# so that mkfs won't complain about having to correct the log stripe unit > +# size that is implied by the provided data device stripe unit. > +cfgfile=$tmp.cfg > +cat << EOF >> $tmp.cfg > +[block] > +size=4096 > + > +[data] > +su=2097152 > +sw=1 > +EOF > + > +# Some mkfs options store the user's value string for processing after certain > +# geometry parameters (e.g. the fs block size) have been settled. This is how > +# the su= option can accept arguments such as "8b" to mean eight filesystem > +# blocks. > +# > +# Unfortunately, on Ubuntu 20.04, the libini parser uses an onstack char[] > +# array to store value that it parse, and it passes the address of this array > +# to the parse_cfgopt. The getstr function returns its argument, which is > +# stored in the cli_params structure by the D_SU parsing code. By the time we > +# get around to interpreting this string, of course, the stack array has long > +# since lost scope and is now full of garbage. If we're lucky, the value will > +# cause a number interpretation failure. If not, the fs is configured with > +# garbage geometry. > +# > +# Either way, set up a config file to exploit this vulnerability so that we > +# can prove that current mkfs works correctly. > +$XFS_IO_PROG -f -c "truncate 1g" $TEST_DIR/fubar.img > +options=(-c options=$cfgfile -l sunit=8 -f -N $TEST_DIR/fubar.img) > +$MKFS_XFS_PROG "${options[@]}" >> $seqres.full || > + echo "mkfs failed" > + > +# success, all done > +echo Silence is golden > +status=0 > +exit > diff --git a/tests/xfs/831.out b/tests/xfs/831.out > new file mode 100644 > index 00000000..abe137e3 > --- /dev/null > +++ b/tests/xfs/831.out > @@ -0,0 +1,2 @@ > +QA output created by 831 > +Silence is golden >