On Wednesday, 18 August 1999 at 9:50:54 -0400, Mark J. Taylor wrote:
> On Wed, 18 Aug 1999, Sheldon Hearn wrote:
>
>>
>>
>> On Wed, 18 Aug 1999 09:19:20 -0400, "Mark J. Taylor" wrote:
>>
>>> There is a long as a parameter to ccdbuffer that needs to be a u_long.
>>> Otherwise, you'll get panics (can't remember where).
>>> Basically, bcount needs to be a u_long in all cases.
>>
>> Que? Are you sure? That means you want to change struct buf, where
>> b_bcount is declared as long, as well?
>
> We didn't go as far as modifying the structure definition, just the
> few "bcount" variables in the ccd.c code.
> The problem was that we were seeing bcount go "negative".
That's a bug. I'd guess that your modification took the bite of the
bug, but the bug remains.
b_bcount is, of course, the number of bytes that the request should
transfer. Both ccd and vinum play around with struct bufs, creating
their own headers, and it's conceivable that under some circumstances
ccd might end up with a negative count. But it's definitely a bug,
and a serious one, if it happens.
> I believe that
> "newfs" of the ccd would panic the kernel, reliably. Even on "smaller"
> ccds (1 Gbyte), I believe.
> I'm talking about ccds configured as in:
> ccdconfig -c ccd0 0 0 /dev/da0s1c
>
> I know, this is fairly worthless as it stands, but it is done so we can
> later "upgrade" the ccd to a mirror.
Right, but it's probably also a pointer to the bug.
> RSN, we will be moving to vinum. Hi Greg!
Hi, Mark! I suppose I should try the corresponding vinum config :-)
> Linux, BTW, does s/w RAID5. But, it seems you can't operate in degraded
> mode, and the RAID5 reconstruction happens at boot time. We've got a 130
> GB RAID5 volume under Linux that takes about 10 hours to "ckraid". I hope
> that this box never panics!
You want vinum on Linux?
Greg
--
See complete headers for address, home page and phone numbers
finger [EMAIL PROTECTED] for PGP public key
To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-current" in the body of the message
