[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-block] [PATCH v4 12/21] mirror: Switch mirror_do_read() to byt
From: |
Kevin Wolf |
Subject: |
Re: [Qemu-block] [PATCH v4 12/21] mirror: Switch mirror_do_read() to byte-based |
Date: |
Thu, 6 Jul 2017 16:55:03 +0200 |
User-agent: |
Mutt/1.5.21 (2010-09-15) |
Am 06.07.2017 um 16:25 hat Eric Blake geschrieben:
> On 07/06/2017 08:30 AM, Kevin Wolf wrote:
> > Am 05.07.2017 um 23:08 hat Eric Blake geschrieben:
> >> We are gradually converting to byte-based interfaces, as they are
> >> easier to reason about than sector-based. Convert another internal
> >> function (no semantic change).
> >>
> >> Signed-off-by: Eric Blake <address@hidden>
> >> Reviewed-by: John Snow <address@hidden>
> >> Reviewed-by: Jeff Cody <address@hidden>
> >
> >> - /* The sector range must meet granularity because:
> >> + assert(bytes <= s->buf_size);
> >> + /* The range will be sector-aligned because:
> >> * 1) Caller passes in aligned values;
> >> - * 2) mirror_cow_align is used only when target cluster is larger. */
> >> - assert(!(sector_num % sectors_per_chunk));
>
> So the strict translation would be assert(!(offset % s->granularity)),
> or rewritten to be assert(QEMU_IS_ALIGNED(offset, s->granularity)).
Right.
> >> - nb_chunks = DIV_ROUND_UP(nb_sectors, sectors_per_chunk);
> >> + * 2) mirror_cow_align is used only when target cluster is larger.
> >> + * But it might not be cluster-aligned at end-of-file. */
> >> + assert(QEMU_IS_ALIGNED(bytes, BDRV_SECTOR_SIZE));
>
> and you're right that this appears to be a new assertion.
>
> >> + nb_chunks = DIV_ROUND_UP(bytes, s->granularity);
> >
> > The assertion in the old code was about sector_num (i.e. that the start
> > of the range is cluster-aligned), not about nb_sectors, so while you add
> > a new assertion, it is asserting something different. This explains
> > why you had to switch to sector aligned even though the semantics
> > shouldn't be changed by this patch.
> >
> > Is this intentional or did you confuse sector_num and nb_sectors? I
> > think we can just have both assertions.
>
> At this point, I'm not sure if I confused things, or if I hit an actual
> iotest failure later in the series that I traced back to this point.
But if you did, wouldn't this indicate a real bug in the patch?
> If I have a reason to respin, I'll see if both assertions still hold
> (the rewritten alignment check on offset, and the new check on bytes
> being sector-aligned), through all the tests.
Actually, I think this one, while seemingly minor, is a reason to
respin. This kind of assertions is exactly for preventing bugs when
doing changes like this patch does, so removing the assertion in such a
patch looks really suspicious.
Also, with the new assertion, the comment doesn't really make that much
sense any more either.
> Also, while asserting that bytes is sector-aligned is good in the
> short term, it may be overkill by the time dirty-bitmap is changed to
> byte alignment (right now, bdrv_getlength() is always sector-aligned,
> because it rounds up; but if we ever make it byte-accurate, then the
> trailing cluster might not be a sector-aligned bytes length).
This is true, but I think for the time being the assertion is worth it.
Kevin
pgpik0pgVTKiT.pgp
Description: PGP signature
- [Qemu-block] [PATCH v4 08/21] mirror: Switch MirrorBlockJob to byte-based, (continued)
- [Qemu-block] [PATCH v4 08/21] mirror: Switch MirrorBlockJob to byte-based, Eric Blake, 2017/07/05
- [Qemu-block] [PATCH v4 09/21] mirror: Switch mirror_do_zero_or_discard() to byte-based, Eric Blake, 2017/07/05
- [Qemu-block] [PATCH v4 10/21] mirror: Update signature of mirror_clip_sectors(), Eric Blake, 2017/07/05
- [Qemu-block] [PATCH v4 11/21] mirror: Switch mirror_cow_align() to byte-based, Eric Blake, 2017/07/05
- [Qemu-block] [PATCH v4 12/21] mirror: Switch mirror_do_read() to byte-based, Eric Blake, 2017/07/05
- [Qemu-block] [PATCH v4 13/21] mirror: Switch mirror_iteration() to byte-based, Eric Blake, 2017/07/05
- [Qemu-block] [PATCH v4 14/21] block: Drop unused bdrv_round_sectors_to_clusters(), Eric Blake, 2017/07/05
- [Qemu-block] [PATCH v4 16/21] backup: Switch block_backup.h to byte-based, Eric Blake, 2017/07/05
- [Qemu-block] [PATCH v4 15/21] backup: Switch BackupBlockJob to byte-based, Eric Blake, 2017/07/05
- [Qemu-block] [PATCH v4 17/21] backup: Switch backup_do_cow() to byte-based, Eric Blake, 2017/07/05