On Wed, Feb 13, 2013 at 04:52:14PM -0500, Benjamin LaHaise wrote:
> The io_cancel() syscall allows for cancellation of iocbs in flight to
> generate a completion event. The current behaviour of batch_complete_aio()
> is to suppress all completion events. Some types of asynchronous operations
> cannot be cancelled synchronously, and must generate a completion event at
> some point after the io_cancel() syscall. Instead, only suppress
> completion events during kioctx teardown by free_ioctx().
>
> Signed-off-by: Benjamin LaHaise <bcrl@xxxxxxxxx>
> ---
> fs/aio.c | 6 +++++-
> 1 files changed, 5 insertions(+), 1 deletions(-)
>
> diff --git a/fs/aio.c b/fs/aio.c
> index 46f9dd0..1bcb818 100644
> --- a/fs/aio.c
> +++ b/fs/aio.c
> @@ -81,6 +81,7 @@ struct kioctx {
>
> /* sys_io_setup currently limits this to an unsigned int */
> unsigned max_reqs;
> + unsigned dead;
Can use percpu_ref_dead() for this - it returns true after
percpu_ref_kill() has been called (it's what I converted the old
ctx->dead uses to).
>
> unsigned long mmap_base;
> unsigned long mmap_size;
> @@ -311,6 +312,7 @@ static void free_ioctx(struct kioctx *ctx)
> struct kiocb *req;
> unsigned cpu, head, avail;
>
> + ctx->dead = 1;
> spin_lock_irq(&ctx->ctx_lock);
>
> while (!list_empty(&ctx->active_reqs)) {
> @@ -749,7 +751,9 @@ void batch_complete_aio(struct batch_complete *batch)
> n = rb_parent(n);
> }
>
> - if (unlikely(xchg(&req->ki_cancel,
> + /* Suppress cancelled events if free_ioctx() is in progress. */
> + if (unlikely(req->ki_ctx->dead &&
> + xchg(&req->ki_cancel,
> KIOCB_CANCELLED) == KIOCB_CANCELLED)) {
I'm not seeing why we need to supress events during teardown - if we
just let it be delivered to the ringbuffer like normal, the free_ioctx()
code will find it there.
I think the event supressing code can be deleted entirely.
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
