Re: [PATCH for 3.2] fs/direct-io.c: Calculate fs_count correctlyin get_more_blocks.

[Tao Ma]

On 11/01/2011 02:12 AM, Jeff Moyer wrote:
> Tao Ma <tm@xxxxxx> writes:
> 
>> From: Tao Ma <boyu.mt@xxxxxxxxxx>
>>
>> In get_more_blocks, we use dio_count to calculate fs_count and do some
>> tricky things to increase fs_count if dio_count isn't aligned. But
>> actually it still has some cornor case that can't be coverd. See the
>> following example:
>> ./dio_write foo -s 1024 -w 4096(direct write 4096 bytes at offset 1024).
>> The same goes if the offset isn't aligned to fs_blocksize.
>>
>> In this case, the old calculation counts fs_count to be 1, but actually
>> we will write into 2 different blocks(if fs_blocksize=4096). The old code
>> just works, since it will call get_block twice(and may have to allocate
>> and create extent twice for file systems like ext4). So we'd better call
>> get_block just once with the proper fs_count.
> 
> This description was *really* hard for me to understand.  It seems to me
> that right now there's an inefficiency in the code.  It's not clear
> whether you're claiming that it was introduced recently, though.  Was
> it, or has this problem been around for a while?
Actually it is there a long time ago. And the good thing is that it
isn't a bug, only some performance overhead.
> 
> How did you notice this?  Was there any evidence of a problem, such as
> performance overhead or less than ideal file layout?
I found it when I dig into some ext4 issues. The ext4 can't create the
whole 8K(in the above case) and ext4 has to create the blocks 2 times
for just one direct i/o write. In some of our test, it costs.

> 
> Anyway, I agree that the code does not correctly calculate the number of
> file system blocks in a request.  I also agree that your patch fixes
> that issue.
> 
> Please ammend the description and then you can add my:
So how about the following commit log(please feel free to modify it if I
still don't describe it correctly).

In get_more_blocks, we use dio_count to calculate fs_count to let the
file system map(maybe also create) blocks. And some tricky things are
done to increase fs_count if dio_count isn't aligned.

But actually it still has some cornor case that can't be coverd. See the
following example:
./dio_write foo -s 1024 -w 4096(direct write 4096 bytes at offset 1024).

In this case, the old calculation counts fs_count to be 1, but actually
we will write into 2 different blocks(if fs_blocksize=4096). So the
underlying file system is called twice and leads to some performance
overhead. So fix it by calculating fs_count correctly and let the file
system knows what we really want to write.

Thanks
Tao

> 
> Acked-by: Jeff Moyer <jmoyer@xxxxxxxxxx>
> 
> Cheers,
> Jeff
> --
> To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
> the body of a message to majordomo@xxxxxxxxxxxxxxx
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/