GBDE + md() + ccd() = corruption?

[Poul-Henning Kamp]

In message <20030413174853.F57816-100000 at grogged.dyndns.org>, matt writes:

>And now the question, To verify the data I've placed on this rather
>unusual setup, I ran md5 on everything I stuck on it.  Some files seem to
>work fine(about half), and I get consistent output from the md5 utility,
>the other half of the files on the device are very inconsistent.  I can
>run md5 on the same file 5 times in a row and get 5 different output
>strings.  To double check things I ran the inconsistent files against
>known good copies sitting on a plain old ufs partition (with diff), and it
>turns out the files *are* different.
>
>Anyone have any idea what's going on here?  Is some portion of this setup
>corrupting the data?

Hmm, that is very bad.  Can you send me an example of the diff ?

How large are your md(4) devices ?

>Also, when writing to the gbde encrypted filesystem I sporadically get
>messages like:
>
>"ENOMEM 0xc26ced80 on 0xc2648b80(md1.bde)"

This is when we cannot sleeplessly allocate memory in the I/O path.
The request gets retried in that case.

-- 
Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe    
Never attribute to malice what can adequately be explained by incompetence.