Re: [PATCH] xfs: test mkfs.xfs sizing of internal logs that

[Dave Chinner <david@fromorbit.com>]

On Tue, May 24, 2022 at 04:49:30PM -0700, Darrick J. Wong wrote:
> On Wed, May 25, 2022 at 09:44:26AM +1000, Dave Chinner wrote:
> > On Tue, May 24, 2022 at 12:52:57PM -0700, Darrick J. Wong wrote:
> > > From: Darrick J. Wong <djwong@kernel.org>
> > > 
> > > This is a regression test that exercises the mkfs.xfs code that creates
> > > log sizes that are very close to the AG size when stripe units are in
> > > play and/or when the log is forced to be in AG 0.
> > > 
> > > Signed-off-by: Darrick J. Wong <djwong@kernel.org>
> > > ---
> > >  tests/xfs/843     |   56 +++++++++++++++++++++++++++++++++++++++++++++++++++++
> > >  tests/xfs/843.out |    2 ++
> > >  2 files changed, 58 insertions(+)
> > >  create mode 100755 tests/xfs/843
> > >  create mode 100644 tests/xfs/843.out
> > > 
> > > diff --git a/tests/xfs/843 b/tests/xfs/843
> > > new file mode 100755
> > > index 00000000..3384b1aa
> > > --- /dev/null
> > > +++ b/tests/xfs/843
> > > @@ -0,0 +1,56 @@
> > > +#! /bin/bash
> > > +# SPDX-License-Identifier: GPL-2.0
> > > +# Copyright (c) 2022 Oracle.  All Rights Reserved.
> > > +#
> > > +# FS QA Test 843
> > > +#
> > > +# Now that we've increased the default log size calculation, test mkfs with
> > > +# various stripe units and filesystem sizes to see if we can provoke mkfs into
> > > +# breaking.
> > > +#
> > > +. ./common/preamble
> > > +_begin_fstest auto mkfs
> > > +
> > > +_cleanup()
> > > +{
> > > +	cd /
> > > +	rm -r -f $tmp.* $testfile
> > > +}
> > 
> > I'd omit this completely.
> > 
> > > +# real QA test starts here
> > > +
> > > +# Modify as appropriate.
> > > +_supported_fs xfs
> > > +_require_test
> > > +
> > > +testfile=$TEST_DIR/a
> > > +rm -f $testfile
> > > +
> > > +test_format() {
> > > +	local tag="$1"
> > > +	shift
> > > +
> > > +	echo "$tag" >> $seqres.full
> > > +	$MKFS_XFS_PROG $@ -d file,name=$testfile &>> $seqres.full
> > > +	local res=$?
> > > +	test $res -eq 0 || echo "$tag FAIL $res" | tee -a $seqres.full
> > 
> > What breakage are you trying to provoke? Just the log size
> > calculation? If so, why do we need to actually write the filesystem
> > to disk? Won't "-N" still calculate everything and fail if it's
> > broken or quit with success without needing to write anything to
> > disk?
> 
> It will, but...
> 
> > > +}
> > > +
> > > +# First we try various small filesystems and stripe sizes.
> > > +for M in `seq 298 302` `seq 490 520`; do
> > > +	for S in `seq 32 4 64`; do
> > > +		test_format "M=$M S=$S" -dsu=${S}k,sw=1,size=${M}m
> > > +	done
> > > +done
> > > +
> > > +# log so large it pushes the root dir into AG 1
> > > +test_format "log pushes rootdir into AG 1" -d agcount=3200,size=6366g -lagnum=0
> 
> ...this particular check in mkfs only happens after we allocate the root
> directory, which an -N invocation doesn't do.

Ok, so for this test can we drop the -N? We don't need to do 30 IOs
and write 64MB logs for every config we test - I think there's ~35 *
8 invocations of test_format in the loop above before we get here...

Also, why do we need a 6.3TB filesystem with 2.1GB AGs and a 2GB log
to trigger this? That means we have to write 2GB to disk, plus
~20,000 write IOs for the AG headers and btree root blocks before we
get to the failure case, yes?

Why not just exercise the failure case with something like this:

# mkfs.xfs -d agcount=2,size=64M -l size=8180b,agnum=0 -d file,name=test.img
meta-data=test.img               isize=512    agcount=2, agsize=8192 blks
         =                       sectsz=512   attr=2, projid32bit=1
         =                       crc=1        finobt=1, sparse=1, rmapbt=0
         =                       reflink=1    bigtime=0 inobtcount=0
data     =                       bsize=4096   blocks=16384, imaxpct=25
         =                       sunit=0      swidth=0 blks
naming   =version 2              bsize=4096   ascii-ci=0, ftype=1
log      =internal log           bsize=4096   blocks=8180, version=2
         =                       sectsz=512   sunit=0 blks, lazy-count=1
realtime =none                   extsz=4096   blocks=0, rtextents=0
mkfs.xfs: root inode created in AG 1, not AG 0
# echo $?
1
#

Otherwise I don't exactly understand what this specific case is
supposed to be testing, so maybe some more explaining is necessary?

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com