On 9/14/23 7:02 AM, Pavel Begunkov wrote:
> On 9/14/23 09:34, Jiri Slaby wrote:
>> On 11. 08. 23, 14:53, Pavel Begunkov wrote:
>>> Don't allow overflowing multishot recv CQEs, it might get out of
>>> hand, hurt performanece, and in the worst case scenario OOM the task.
>>>
>>> Cc: stable@xxxxxxxxxxxxxxx
>>> Fixes: b3fdea6ecb55c ("io_uring: multishot recv")
>>> Signed-off-by: Pavel Begunkov <asml.silence@xxxxxxxxx>
>>> ---
>>> io_uring/net.c | 2 +-
>>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/io_uring/net.c b/io_uring/net.c
>>> index 1599493544a5..8c419c01a5db 100644
>>> --- a/io_uring/net.c
>>> +++ b/io_uring/net.c
>>> @@ -642,7 +642,7 @@ static inline bool io_recv_finish(struct io_kiocb *req, int *ret,
>>> if (!mshot_finished) {
>>> if (io_aux_cqe(req, issue_flags & IO_URING_F_COMPLETE_DEFER,
>>> - *ret, cflags | IORING_CQE_F_MORE, true)) {
>>> + *ret, cflags | IORING_CQE_F_MORE, false)) {
>>
>> This one breaks iouring's recv-multishot.t test:
>> Running test recv-multishot.t MORE flag not set
>> test stream=0 wait_each=0 recvmsg=0 early_error=4 defer=0 failed
>> Test recv-multishot.t failed with ret 1
>>
>> Is the commit or the test broken ;)?
>
> The test is not right. I fixed it up while sending the kernel patch,
> but a new liburing version hasn't been cut yet and I assume you're
> not using upstream version?
I'll probably cut a new release soonish, we're after the merge window
now so whatever went into 6.6-rc should be solid/sane API wise.
--
Jens Axboe
