On 3/20/13 5:47 AM, Jan Kara wrote:
> On Tue 19-03-13 12:05:18, Eric Sandeen wrote:
>> On 3/19/13 11:21 AM, Jan Kara wrote:
>>> Test 289 ignored the fact that historically journal is not accounted as
>>> fs overhead. For larger filesystems it is hidden in 1% tolerance but for
>>> filesystems smaller than 12G the test fails. So make the counting
>>> precise to work everywhere.
>>
>> Thanks.
>>
>> 0875a2b448fcaba67010850cf9649293a5ef653d ext4: include journal blocks in df
>> overhead calcs
> Bah, I missed that. But ext3 definitely needs this (and I'm not sure we
> want to change that behavior after so many years - it's mostly cosmetic
> anyway so the possibility of breaking some userspace seems a bad tradeoff).
Sure, that's fine. Ext4 has munged df reporting a few times recently
anyway, so figured it was worth sneaking in something to make it more
accurate.
>> changed this again, right - so will this change work in both cases?
> No, it will now fail for small ext4 filesystems instead of small ext3
> filesystems :). So we will count journal blocks only for ext3, ok?
Sounds like a plan.
>> Also: is using "Journal length: " any simpler?
> It would be, but older versions of dumpe2fs don't have that (it happened
> somewhere between 1.41.9 and 1.41.11) so I figured I will use a more
> generic approach (too lazy to fix my test machine ;).
Makes sense :)
Thanks,
-Eric
> Honza
>
>>> Signed-off-by: Jan Kara <jack@xxxxxxx>
>>> ---
>>> 289 | 12 ++++++++----
>>> common.filter | 15 +++++++++++++++
>>> 2 files changed, 23 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/289 b/289
>>> index b057c20..9bba144 100755
>>> --- a/289
>>> +++ b/289
>>> @@ -59,10 +59,14 @@ TOTAL_BLOCKS=`dumpe2fs -h $SCRATCH_DEV 2>/dev/null \
>>> FREE_BLOCKS=`dumpe2fs -h $SCRATCH_DEV 2>/dev/null \
>>> | awk '/Free blocks:/{print $3}'`
>>>
>>> -# nb: kernels today don't count journal blocks as overhead, but should.
>>> -# For most filesystems this will still be within tolerance.
>>> -# Overhead is all the blocks (already) used by the fs itself:
>>> -OVERHEAD=$(($TOTAL_BLOCKS-$FREE_BLOCKS))
>>> +JOURNAL_SIZE=`dumpe2fs -h $SCRATCH_DEV 2>/dev/null \
>>> + | awk '/Journal size:/{print $3}' | _filter_size_to_bytes`
>>> +BLOCK_SIZE=`dumpe2fs -h $SCRATCH_DEV 2>/dev/null \
>>> + | awk '/Block size:/{print $3}'`
>>> +JOURNAL_BLOCKS=$(($JOURNAL_SIZE/$BLOCK_SIZE))
>>> +
>>> +# kernels today don't count journal blocks as overhead, but should.
>>> +OVERHEAD=$(($TOTAL_BLOCKS-$FREE_BLOCKS-$JOURNAL_BLOCKS))
>>>
>>> # bsddf|minixdf
>>> # Set the behaviour for the statfs system call. The minixdf
>>> diff --git a/common.filter b/common.filter
>>> index f0f6076..fcd7589 100644
>>> --- a/common.filter
>>> +++ b/common.filter
>>> @@ -229,5 +229,20 @@ _filter_spaces()
>>> sed -e 's/ [ ]*/ /g'
>>> }
>>>
>>> +# Convert string read from stdin like 128K to bytes and print it to stdout
>>> +_filter_size_to_bytes()
>>> +{
>>> + read size
>>> + suffix=${size:${#size}-1}
>>> + mul=1
>>> + case $suffix in
>>> + K) mul=1024 ;;
>>> + M) mul=$((1024*1024)) ;;
>>> + G) mul=$((1024*1024*1024)) ;;
>>> + T) mul=$((1024*1024*1024*1024)) ;;
>>> + esac
>>> + echo $((${size:0:${#size}-1}*$mul))
>>> +}
>>> +
>>> # make sure this script returns success
>>> /bin/true
>>>
>>
|