On 9/1/24 7:37 AM, Bernd Schubert wrote:
> This is to allow copying into the buffer from the application
> without the need to copy in ring context (and with that,
> the need that the ring task is active in kernel space).
>
> Also absolutely needed for now to avoid this teardown issue
I'm fine using these helpers, but they are absolutely not needed to
avoid that teardown issue - well they may help because it's already
mapped, but it's really the fault of your handler from attempting to map
in user pages from when it's teardown/fallback task_work. If invoked and
the ring is dying or not in the right task (as per the patch from
Pavel), then just cleanup and return -ECANCELED.
> +/*
> + * Copy from memmap.c, should be exported
> + */
> +static void io_pages_free(struct page ***pages, int npages)
> +{
> + struct page **page_array = *pages;
> +
> + if (!page_array)
> + return;
> +
> + unpin_user_pages(page_array, npages);
> + kvfree(page_array);
> + *pages = NULL;
> +}
I noticed this and the mapping helper being copied before seeing the
comments - just export them from memmap.c and use those rather than
copying in the code. Add that as a prep patch.
> @@ -417,6 +437,7 @@ static int fuse_uring_out_header_has_err(struct fuse_out_header *oh,
> goto seterr;
> }
>
> + /* FIXME copied from dev.c, check what 512 means */
> if (oh->error <= -512 || oh->error > 0) {
> err = -EINVAL;
> goto seterr;
-512 is -ERESTARTSYS
--
Jens Axboe
