On Sun, 28 Nov 1999, Manfred Spraul wrote:
> coda_file_write() in fs/coda/file.c calls down(&i_sem) before calling
> f_op->write(). This will lock-up because generic_file_write() calls
> down(&i_sem) again.
Yes.
> AFAICS, these lines are superflous, below is an untested patch.
>
> Btw, where can I find informations about the file synchronization [ie
> who should call down(&i_sem), which level of concurrency the various
> standard allow,...]?
>
> I think it's a Bad Thing (tm) that file write operations are single
> threaded [generic_file_write() calls down(&i_sem)], and I'd like to
> change that.
Danger: clashing patches ahead. Seriously, there's some work going in that
area (some parts went into -pre3, but there will be more to that) and
we'ld better sync the stuff here. In particular, any stuff related to
areas, NFS atomicity, etc. would better live on address_space level. And
keep in mind that locks need redesign too - when acquired they should turn
into (and block on) the areas in file address space.
