Re: [fuse-devel] fusexmp, O_DIRECT, pwrite, EINVAL

[Maxim V. P. <mpa...@pa...>]

Hi Lyndon,

I think fusexmp and fusexmp_fh are just simple examples, something like 
'hello world!'. IMHO, they are not supposed to support all possible 
use-cases.

As for fusexmp, it should work 'as is' because it re-opens file on each 
write w/o O_DIRECT. See its xmp_write. It's strange that it doesn't work 
for you.

As for fusexmp_fh, you can either patch fusexmp_fh to realloc data 
buffer by posix_memalign, fill it with data and pass it to pwrite. Or 
locate where userspace fuse library allocates data-buffer for write and 
substitute malloc with posix_memalign there.

Thanks,
Maxim

09/28/2012 12:44 PM, Lyndon Clarke пишет:
> Hi Mi Jinlong,
>
> Thanks - the kernel is 3.4.6 (latest and greatest open suse).
>
> So, is this the expected behaviour? Is there anything I can do to get these buffers aligned? Or perhaps is there some known reason I am should not be thinking about O_DIRECT through FUSE in a recent kernel?
>
> Best Wishes,
> Lyndon
>
> -----Original Message-----
> From: Mi Jinlong [mailto:kin...@gm...]
> Sent: 28 September 2012 02:54
> To: fus...@li...
> Subject: Re: [fuse-devel] fusexmp, O_DIRECT, pwrite, EINVAL
>
> 于 2012-9-27 21:50, Lyndon Clarke 写道:
>> Hello,
>>
>> I've been experimenting with fusexmp and fusexmp_fh as a learning exercise. I have a small query about writes to files opened with O_DIRECT ...
>>
>> I have a 4KB aligned buffer in a simple little program which opens with O_DIRECT and does some writes (or reads). It runs fine against ext4. When I bring fusexmp[_fh] into the loop, the writes fail with EINVAL.
>>
>> I can see that xmp_write_buf in fusexmp_fh is being invoked, and fuse_buf_copy is returning -EINVAL. For a little more information, if I remove the registration of xmp_write_buf, then I can see that xmp_write is being invoked, and the down-call of pwrite sets errno to EINVAL. I think this is happening because the buf parameter is not sector aligned.
>>
>> Is this the expected behaviour? Is there anything I can do to get these buffers aligned? Or perhaps is there some known reason I am should not be thinking about O_DIRECT through FUSE?
> As I known, FUSE supports O_DIRECT at the following patch, http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff;h=4273b793ec68753cc3fcf5be7cbfd88c2be2058d
>
> If your kernel is older, O_DIRECT isnot supported.
>
> thanks,
> Mi Jinlong
>
>
> ------------------------------------------------------------------------------
> Got visibility?
> Most devs has no idea what their production app looks like.
> Find out how fast your code is with AppDynamics Lite.
> http://ad.doubleclick.net/clk;262219671;13503038;y?
> http://info.appdynamics.com/FreeJavaPerformanceDownload.html
> _______________________________________________
> fuse-devel mailing list
> fus...@li...
> https://lists.sourceforge.net/lists/listinfo/fuse-devel
>
> This message has been scanned by MailController - portal1.mailcontroller.co.uk
>
> This message has been scanned by MailController - portal1.mailcontroller.co.uk
> ------------------------------------------------------------------------------
> Got visibility?
> Most devs has no idea what their production app looks like.
> Find out how fast your code is with AppDynamics Lite.
> http://ad.doubleclick.net/clk;262219671;13503038;y?
> http://info.appdynamics.com/FreeJavaPerformanceDownload.html
> _______________________________________________
> fuse-devel mailing list
> fus...@li...
> https://lists.sourceforge.net/lists/listinfo/fuse-devel