Re: [PATCH for-next] io_uring: fix CQE reordering

[Jens Axboe <axboe@kernel.dk>]

On 9/23/22 8:26 AM, Pavel Begunkov wrote:
> On 9/23/22 15:19, Jens Axboe wrote:
>> On 9/23/22 7:53 AM, Pavel Begunkov wrote:
>>> Overflowing CQEs may result in reordeing, which is buggy in case of
>>> links, F_MORE and so.
>>>
>>> Reported-by: Dylan Yudaken <dylany@fb.com>
>>> Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
>>> ---
>>>   io_uring/io_uring.c | 12 ++++++++++--
>>>   io_uring/io_uring.h | 12 +++++++++---
>>>   2 files changed, 19 insertions(+), 5 deletions(-)
>>>
>>> diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
>>> index f359e24b46c3..62d1f55fde55 100644
>>> --- a/io_uring/io_uring.c
>>> +++ b/io_uring/io_uring.c
>>> @@ -609,7 +609,7 @@ static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force)
>>>         io_cq_lock(ctx);
>>>       while (!list_empty(&ctx->cq_overflow_list)) {
>>> -        struct io_uring_cqe *cqe = io_get_cqe(ctx);
>>> +        struct io_uring_cqe *cqe = io_get_cqe_overflow(ctx, true);
>>>           struct io_overflow_cqe *ocqe;
>>>             if (!cqe && !force)
>>> @@ -736,12 +736,19 @@ bool io_req_cqe_overflow(struct io_kiocb *req)
>>>    * control dependency is enough as we're using WRITE_ONCE to
>>>    * fill the cq entry
>>>    */
>>> -struct io_uring_cqe *__io_get_cqe(struct io_ring_ctx *ctx)
>>> +struct io_uring_cqe *__io_get_cqe(struct io_ring_ctx *ctx, bool overflow)
>>>   {
>>>       struct io_rings *rings = ctx->rings;
>>>       unsigned int off = ctx->cached_cq_tail & (ctx->cq_entries - 1);
>>>       unsigned int free, queued, len;
>>>   +    /*
>>> +     * Posting into the CQ when there are pending overflowed CQEs may break
>>> +     * ordering guarantees, which will affect links, F_MORE users and more.
>>> +     * Force overflow the completion.
>>> +     */
>>> +    if (!overflow && (ctx->check_cq & BIT(IO_CHECK_CQ_OVERFLOW_BIT)))
>>> +        return NULL;
>>
>> Rather than pass this bool around for the hot path, why not add a helper
>> for the case where 'overflow' isn't known? That can leave the regular
>> io_get_cqe() avoiding this altogether.
> 
> Was choosing from two ugly-ish solutions, but io_get_cqe() should be
> inline and shouldn't really matter, but that's only the case in theory
> though. If someone cleans up the CQE32 part and puts it into a separate
> non-inline function, it'll be actually inlined.

Yes, in theory the current one will be fine as it's known at compile
time. In theory... Didn't check if practice agrees with that, would
prefer if we didn't leave this to the compiler. Fiddling some other
bits, will check in a bit if I have a better idea.

-- 
Jens Axboe