Re: [fuse-devel] fuse do a truncate instead of ftruncate

[Alain S. <asp...@gm...>]

On Thu, Mar 10, 2011 at 1:22 PM, Miklos Szeredi <mi...@sz...> wrote:
> On Thu, 10 Mar 2011, Alain Spineux wrote:
>> > doing a
>> > # echo  hello > /mnt2/tmp/hello
>> > when /mnt2/tmp/hello already exists
>> >
>> > I don't understand why FUSE do a :
>> >
>> > open
>> > truncate
>> >
>> > instead of
>> >
>> > truncate
>> > open
>> >
>> > or at least
>> >
>> > open
>> > ftruncate (with a F)
>> >
>> > because in the first case, all initializations done in fi->fh in
>> > invalidate by the truncate.
>> > If ftruncate wall called instead, I could update fi->fh myself.
>> > I thing I read somewhere FUSE should do a truncate before the open.
>
> The open+ftruncate doesn't work in the open(..., O_TRUNC|O_RDONLY)
> case.
>
>> My probleme is identical to the one described here.
>> My fedora 14 is working as the Ubuntu :
>>
>> http://marc.info/?l=fuse-devel&m=129540380518880&w=1
>>
>> Anyways the answere is incomplete and the post tell me the problem is real.
>
> The problem is because different kernels call truncate/open in
> different orders.

This doesn't help us !

>
> This can be solved by setting FUSE_CAP_ATOMIC_O_TRUNC.  I'll fix
> libfuse to set this flag if the kernel supprts it and order the
> truncate and the open in the way it was done previously.

This will help.

>
> In the mean time you can set FUSE_CAP_ATOMIC_O_TRUNC yourself and
> handle O_TRUNC in the open() method, which should solve your problem.

Yes I saw that, this is in my mind.

Thanks for answering.


>
> Thanks,
> Miklos
>



-- 
Alain Spineux                   |  aspineux gmail com
Your email 100% available |  http://www.emailgency.com
Send backup mail report    | http://www.magikmon.com/mksbackup