Re: weird error check for aio_cancel

Re: weird error check for aio_cancel

[Jens Axboe]

On 2010-06-11 06:33, YAMAMOTO Takashi wrote:
> hi,
> 
> please look at the following patch.
> 
> YAMAMOTO Takashi
> 
> Signed-off-by: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
> 
> commit e7aa7a65700932fe68d96015c06e6e1b1ecea0da
> Author: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
> Date:   Fri Jun 11 12:32:42 2010 +0900
> 
>     fix a weird check of the return value of aio_cancel.
>     
>     i don't know what this 1 means in the first place but at least
>     it isn't portable.  instead, check if the i/o has already
>     completed. (AIO_ALLDONE)

You are right, that looks like a genuine bug. The last case of
AIO_NOTCANCELLED should be ok, the caller should use normal completions
to reap the remainders.

I'll apply this.

-- 
Jens Axboe

Re: weird error check for aio_cancel

[Jens Axboe]

On 2010-06-11 09:19, Jens Axboe wrote:
> On 2010-06-11 06:33, YAMAMOTO Takashi wrote:
>> hi,
>>
>> please look at the following patch.
>>
>> YAMAMOTO Takashi
>>
>> Signed-off-by: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
>>
>> commit e7aa7a65700932fe68d96015c06e6e1b1ecea0da
>> Author: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
>> Date:   Fri Jun 11 12:32:42 2010 +0900
>>
>>     fix a weird check of the return value of aio_cancel.
>>     
>>     i don't know what this 1 means in the first place but at least
>>     it isn't portable.  instead, check if the i/o has already
>>     completed. (AIO_ALLDONE)
> 
> You are right, that looks like a genuine bug. The last case of
> AIO_NOTCANCELLED should be ok, the caller should use normal completions
> to reap the remainders.
> 
> I'll apply this.

BTW, since you are on NetBSD, have you noticed any problems with the
performance of aio with posixaio there? I have a report that it doesn't
work particularly fast on FreeBSD, at least it isn't able to saturate a
fast device like a small test app using posixaio would. That sounds like
a bug and I'm going to look into it, just wondering if you have seen
anything to that extent in your testing.

-- 
Jens Axboe