Re: [PULL 01/19] util/hbitmap: strict hbitmap_reset

[Kevin Wolf <kwolf@redhat.com>]

Am 14.10.2019 um 20:10 hat John Snow geschrieben:
> 
> 
> On 10/11/19 7:18 PM, John Snow wrote:
> > 
> > 
> > On 10/11/19 5:48 PM, Eric Blake wrote:
> >> On 10/11/19 4:25 PM, John Snow wrote:
> >>> From: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
> >>>
> >>> hbitmap_reset has an unobvious property: it rounds requested region up.
> >>> It may provoke bugs, like in recently fixed write-blocking mode of
> >>> mirror: user calls reset on unaligned region, not keeping in mind that
> >>> there are possible unrelated dirty bytes, covered by rounded-up region
> >>> and information of this unrelated "dirtiness" will be lost.
> >>>
> >>> Make hbitmap_reset strict: assert that arguments are aligned, allowing
> >>> only one exception when @start + @count == hb->orig_size. It's needed
> >>> to comfort users of hbitmap_next_dirty_area, which cares about
> >>> hb->orig_size.
> >>>
> >>> Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
> >>> Reviewed-by: Max Reitz <mreitz@redhat.com>
> >>> Message-Id: <20190806152611.280389-1-vsementsov@virtuozzo.com>
> >>> [Maintainer edit: Max's suggestions from on-list. --js]
> >>> Signed-off-by: John Snow <jsnow@redhat.com>
> >>> ---
> >>>   include/qemu/hbitmap.h | 5 +++++
> >>>   tests/test-hbitmap.c   | 2 +-
> >>>   util/hbitmap.c         | 4 ++++
> >>>   3 files changed, 10 insertions(+), 1 deletion(-)
> >>>
> >>
> >>> +++ b/util/hbitmap.c
> >>> @@ -476,6 +476,10 @@ void hbitmap_reset(HBitmap *hb, uint64_t start,
> >>> uint64_t count)
> >>>       /* Compute range in the last layer.  */
> >>>       uint64_t first;
> >>>       uint64_t last = start + count - 1;
> >>> +    uint64_t gran = 1ULL << hb->granularity;
> >>> +
> >>> +    assert(!(start & (gran - 1)));
> >>> +    assert(!(count & (gran - 1)) || (start + count == hb->orig_size));
> >>
> >> I know I'm replying a bit late (since this is now a pull request), but
> >> would it be worth using the dedicated macro:
> >>
> >> assert(QEMU_IS_ALIGNED(start, gran));
> >> assert(QEMU_IS_ALIGNED(count, gran) || start + count == hb->orig_size);
> >>
> >> instead of open-coding it?  (I would also drop the extra () around the
> >> right half of ||). If we want it, that would now be a followup patch.
> 
> I've noticed that seasoned C programmers hate extra parentheses a lot.
> I've noticed that I cannot remember operator precedence enough to ever
> feel like this is actually an improvement.
> 
> Something about a nice weighted tree of ((expr1) || (expr2)) feels
> soothing to my weary eyes. So, if it's not terribly important, I'd
> prefer to leave it as-is.

I don't mind the parentheses, but I do prefer QEMU_IS_ALIGNED() to the
open-coded version. Would that be a viable compromise?

Kevin