Bug with O_DIRECT in 2.6.36-rc3?

Bug with O_DIRECT in 2.6.36-rc3?

[Ben Greear]

I was getting ENOMEM errors on write when using O_DIRECT in 2.6.36-rc3 plus the
patch I just posted for srcaddr= (my test harness needs that feature to work).

It seems to work fine w/out O_DIRECT enabled.

The same patch against 2.6.34.6 works fine with and without O_DIRECT.

After the ENOMEM, I cannot un-mount, even with -f (says it's busy).

But, fuser doesn't show any users.

Anyone tried O_DIRECT recently?

Thanks,
Ben

-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com

Re: Bug with O_DIRECT in 2.6.36-rc3?

[Ben Greear]

On 09/03/2010 11:58 AM, Ben Greear wrote:
> I was getting ENOMEM errors on write when using O_DIRECT in 2.6.36-rc3
> plus the
> patch I just posted for srcaddr= (my test harness needs that feature to
> work).
>
> It seems to work fine w/out O_DIRECT enabled.
>
> The same patch against 2.6.34.6 works fine with and without O_DIRECT.
>
> After the ENOMEM, I cannot un-mount, even with -f (says it's busy).
>
> But, fuser doesn't show any users.
>
> Anyone tried O_DIRECT recently?

I did some testing on un-patched kernels:

It seems this was broken after 2.6.35:

# good: [26845f585fad66dc23d87dad89d403cd64b48780] [SCSI] remove fake "address-of" expression

and before 2.6.36-rc2.

2.6.36-rc1 won't boot for me, and neither would the kernels
generated by my attempt to git-bisect this further.

Thanks,
Ben

-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com