On Wed, Jul 24, 2019 at 05:43:17PM +0800, Murphy Zhou wrote:
> Hi,
>
> As subject.
>
> -d su=2m,sw=1 && -o dax fail
> -d su=2m,sw=1 && NO dax pass
> no su mkfs option && -o dax pass
> no su mkfs option && NO dax pass
>
> On latest Linus tree. Reproduce every time.
>
> Testing on older kernels are going on to see if it's a regression.
>
> Is this failure expected ?
I'm not sure it's actually a failure at all. DAX does not do delayed
allocation, so if the write is aligned to sunit and at EOF it will
round the allocation up to a full stripe unit. IOWs, for this test
once the file size gets beyond sunit on DAX, writes will allocate in
sunit chunks.
And, well, xfs/050 has checks in it for extent size hints, and
notruns if:
[ $extsize -ge 512000 ] && \
_notrun "Extent size hint is too large ($extsize bytes)"
Because EDQUOT occurs when:
> + URK 99: 2097152 is out of range! [3481600,4096000]
the file has less than 3.5MB or > 4MB allocated to it, and for a
stripe unit of > 512k, EDQUOT will occur at <3.5MB. That's what
we are seeing here - a 2MB allocation at offset 2MB is > 4096000
bytes, and so it gets EDQUOT at that point....
IOWs, this looks like a test problem, not a code failure...
Cheers,
Dave.
--
Dave Chinner
david@xxxxxxxxxxxxx
