quick question about write-intent bitmap

quick question about write-intent bitmap

[Marcus Sorensen]

I would appreciate being pointed to any sort of in-depth documentation
on how the write intent bitmap works. To make my question quick, I'm
wondering how the write-intent bitmap holds up to writeback caching.
It seems like md would assume that the cached writes had made it to
disk, and only rebuild writes that happened since the disk was
removed, leaving an inconsistent array.

Re: quick question about write-intent bitmap

[NeilBrown]

[-- Attachment #1: Type: text/plain, Size: 722 bytes --]

On Wed, 30 Nov 2011 23:34:13 -0700 Marcus Sorensen <shadowsor@gmail.com>
wrote:

> I would appreciate being pointed to any sort of in-depth documentation
> on how the write intent bitmap works. To make my question quick, I'm
> wondering how the write-intent bitmap holds up to writeback caching.
> It seems like md would assume that the cached writes had made it to
> disk, and only rebuild writes that happened since the disk was
> removed, leaving an inconsistent array.
>

md writes bitmaps with

  REQ_WRITE | REQ_SYNC | REQ_FLUSH | REQ_FUA

If that returns before the data is truly safe, then you have a bug somewhere.

i.e. md tries to write-through an cache that might be in the way.

NeilBrown

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 828 bytes --]