Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:18 as permitted sender) client-ip=2620:137:e000::1:18;
Message-ID: <8b934240-4388-d0ff-0d36-69acb253a2a6@gmail.com>
Date:   Sat, 7 May 2022 14:32:01 +0800
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.15; rv:91.0)
 Gecko/20100101 Thunderbird/91.9.0
Subject: Re: [PATCH 3/5] io_uring: let fast poll support multishot
To:     Jens Axboe <axboe@kernel.dk>,
        Pavel Begunkov <asml.silence@gmail.com>,
        io-uring@vger.kernel.org
Cc:     linux-kernel@vger.kernel.org
References: <20220506070102.26032-1-haoxu.linux@gmail.com>
 <20220506070102.26032-4-haoxu.linux@gmail.com>
 <d68381cf-a9fc-33b8-8a9c-ff8485ba8d19@gmail.com>
 <8e81111d-398c-3810-50b4-e1475e956b6f@kernel.dk>
From:   Hao Xu <haoxu.linux@gmail.com>
In-Reply-To: <8e81111d-398c-3810-50b4-e1475e956b6f@kernel.dk>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk

在 2022/5/7 上午6:02, Jens Axboe 写道:
> On 5/6/22 11:19 AM, Pavel Begunkov wrote:
>> On 5/6/22 08:01, Hao Xu wrote:
>>> From: Hao Xu <howeyxu@tencent.com>
>>>
>>> For operations like accept, multishot is a useful feature, since we can
>>> reduce a number of accept sqe. Let's integrate it to fast poll, it may
>>> be good for other operations in the future.
>>>
>>> Signed-off-by: Hao Xu <howeyxu@tencent.com>
>>> ---
>>>    fs/io_uring.c | 41 ++++++++++++++++++++++++++---------------
>>>    1 file changed, 26 insertions(+), 15 deletions(-)
>>>
>>> diff --git a/fs/io_uring.c b/fs/io_uring.c
>>> index 8ebb1a794e36..d33777575faf 100644
>>> --- a/fs/io_uring.c
>>> +++ b/fs/io_uring.c
>>> @@ -5952,7 +5952,7 @@ static void io_poll_remove_entries(struct io_kiocb *req)
>>>     * either spurious wakeup or multishot CQE is served. 0 when it's done with
>>>     * the request, then the mask is stored in req->cqe.res.
>>>     */
>>> -static int io_poll_check_events(struct io_kiocb *req, bool locked)
>>> +static int io_poll_check_events(struct io_kiocb *req, bool *locked)
>>>    {
>>>        struct io_ring_ctx *ctx = req->ctx;
>>>        int v;
>>> @@ -5981,17 +5981,26 @@ static int io_poll_check_events(struct io_kiocb *req, bool locked)
>>>              /* multishot, just fill an CQE and proceed */
>>>            if (req->cqe.res && !(req->apoll_events & EPOLLONESHOT)) {
>>> -            __poll_t mask = mangle_poll(req->cqe.res & req->apoll_events);
>>> -            bool filled;
>>> -
>>> -            spin_lock(&ctx->completion_lock);
>>> -            filled = io_fill_cqe_aux(ctx, req->cqe.user_data, mask,
>>> -                         IORING_CQE_F_MORE);
>>> -            io_commit_cqring(ctx);
>>> -            spin_unlock(&ctx->completion_lock);
>>> -            if (unlikely(!filled))
>>> -                return -ECANCELED;
>>> -            io_cqring_ev_posted(ctx);
>>> +            if (req->flags & REQ_F_APOLL_MULTISHOT) {
>>> +                io_tw_lock(req->ctx, locked);
>>> +                if (likely(!(req->task->flags & PF_EXITING)))
>>> +                    io_queue_sqe(req);
>>
>> That looks dangerous, io_queue_sqe() usually takes the request
>> ownership and doesn't expect that someone, i.e.
>> io_poll_check_events(), may still be actively using it.
> 
> I took a look at this, too. We do own the request at this point, but
> it's still on the poll list. If io_accept() fails, then we do run the
> poll_clean.
> 
>> E.g. io_accept() fails on fd < 0, return an error, io_queue_sqe() ->
>> io_queue_async() -> io_req_complete_failed() kills it. Then
>> io_poll_check_events() and polling in general carry on using the freed
>> request => UAF. Didn't look at it too carefully, but there might other
>> similar cases.
> 
> But we better have done poll_clean() before returning the error. What am
> I missing here?

Sorry, I don't get it. I've done the poll_clean() before returnning
error in io_accept()
>