Re: [fuse-devel] Caching per-request stat() results

[David S. <dav...@gm...>]

On 3/21/07, Miklos Szeredi <mi...@sz...> wrote:
> > During a single fuse request, I stat() one or more paths (that are not
> > handled by my filesystem) to determine information that will allow the
> > fuse request to proceed, return -EACCES, etc. Depending on what's
> > going on, different functions may stat the same path more than once.
> >
> > I'd like to cache the results of stat()ing a few paths on a
> > per-fuse-request basis. The private_data member of the fuse_context
> > seems to be the logical place for this, as follows:
> >
> > - At the beginning of each function that handles a fuse operation
> > (getattr, create, etc.) allocate memory and initialize private_data to
> > my struct which will hold the stat info.
> >
> > - As request processing proceeds, that private_data pointer can be
> > obtained from the context (by any function) and read from / written to
> >
> > - Before the function that is handling the operation returns, it frees
> > up the memory allocated for the private_data struct.
>
> You shouln't use the private_data field for this.
>
> If you really want some per-request context, and your filesystem is
> multithreaded, then just create a thread local storage for yourself
> with pthread_key_create().  If the filesystem is single threaded, just
> use a global variable.

Ah, OK. Thanks for the clarification. Other members of the
fuse_context (e.g. pid) are thread/request specific, though, right? (I
am assuming that one request is handled by one and only 1 thread -- is
that incorrect?)

Thanks,
David