Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Subject: Re: [PATCH v3 0/4] forbid fix {SQ,IO}POLL
To:     Pavel Begunkov <asml.silence@gmail.com>, io-uring@vger.kernel.org,
        linux-kernel@vger.kernel.org
References: <cover.1591196426.git.asml.silence@gmail.com>
 <414b9a24-2e70-3637-0b98-10adf3636c37@kernel.dk>
 <f5370eb3-af80-5481-3589-675befa41009@kernel.dk>
 <d1d92d99-c6b1-fc6e-ea1d-6c2e5097d83f@gmail.com>
 <cc3197f9-e8b1-ac13-c121-291bb32646e3@kernel.dk>
 <947accf4-5ba1-cd39-2aeb-efb7065efb84@gmail.com>
From:   Jens Axboe <axboe@kernel.dk>
Message-ID: <8d817367-436b-073c-6a2c-e60830d7646e@kernel.dk>
Date:   Thu, 4 Jun 2020 14:17:27 -0600
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101
 Thunderbird/68.8.0
MIME-Version: 1.0
In-Reply-To: <947accf4-5ba1-cd39-2aeb-efb7065efb84@gmail.com>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 6/4/20 2:12 PM, Pavel Begunkov wrote:
> On 04/06/2020 22:52, Jens Axboe wrote:
>> On 6/4/20 1:22 PM, Pavel Begunkov wrote:
>>> On 04/06/2020 20:06, Jens Axboe wrote:
>>>> On 6/3/20 12:51 PM, Jens Axboe wrote:
>>>>> On 6/3/20 9:03 AM, Pavel Begunkov wrote:
>>>>>> The first one adds checks {SQPOLL,IOPOLL}. IOPOLL check can be
>>>>>> moved in the common path later, or rethinked entirely, e.g.
>>>>>> not io_iopoll_req_issued()'ed for unsupported opcodes.
>>>>>>
>>>>>> 3 others are just cleanups on top.
>>>>>>
>>>>>>
>>>>>> v2: add IOPOLL to the whole bunch of opcodes in [1/4].
>>>>>>     dirty and effective.
>>>>>> v3: sent wrong set in v2, re-sending right one 
>>>>>>
>>>>>> Pavel Begunkov (4):
>>>>>>   io_uring: fix {SQ,IO}POLL with unsupported opcodes
>>>>>>   io_uring: do build_open_how() only once
>>>>>>   io_uring: deduplicate io_openat{,2}_prep()
>>>>>>   io_uring: move send/recv IOPOLL check into prep
>>>>>>
>>>>>>  fs/io_uring.c | 94 ++++++++++++++++++++++++++-------------------------
>>>>>>  1 file changed, 48 insertions(+), 46 deletions(-)
>>>>>
>>>>> Thanks, applied.
>>>>
>>>> #1 goes too far, provide/remove buffers is fine with iopoll. I'll
>>>> going to edit the patch.
>>>
>>> Conceptually it should work, but from a quick look:
>>>
>>> - io_provide_buffers() drops a ref from req->refs, which should've
>>> been used by iopoll*. E.g. io_complete_rw_iopoll() doesn't do that.
>>>
>>> - it doesn't set REQ_F_IOPOLL_COMPLETED, thus iopoll* side will
>>> call req->file->iopoll().
>>
>> We don't poll for provide/remove buffers, or file update. The
>> completion is done inline. The REQ_F_IOPOLL_COMPLETED and friends
>> is only applicable on read/writes.
>>
> 
> 1. Let io_provide_buffers() succeeds, putting a ref and returning 0
> 
> 2. io_issue_sqe() on the way back do IORING_SETUP_IOPOLL check,
> where it calls io_iopoll_req_issued(req)

Only if req->file is valid, which it isn't for these non-file requests.

> 
> 3. io_iopoll_req_issued() unconditionally adds the req into ->poll_list
> 
> 4. io_do_iopoll() checks the req, doesn't find it flagged with
> REQ_F_IOPOLL_COMPLETED, and tries req->file->iopoll().
> 
> 
> Do I miss something? Just did a quick and dirty test, which segfaulted.
> Not certain about it though.
> 


-- 
Jens Axboe