Re: [PATCH v6 3/3] io_uring: allow disabling rings during the creation

From: Jens Axboe <axboe@kernel.dk>
To: Stefano Garzarella <sgarzare@redhat.com>
Cc: Kernel Hardening <kernel-hardening@lists.openwall.com>,
	Christian Brauner <christian.brauner@ubuntu.com>,
	linux-fsdevel@vger.kernel.org, io-uring@vger.kernel.org,
	Alexander Viro <viro@zeniv.linux.org.uk>,
	Stefan Hajnoczi <stefanha@redhat.com>,
	Jann Horn <jannh@google.com>, Jeff Moyer <jmoyer@redhat.com>,
	Aleksa Sarai <asarai@suse.de>, Sargun Dhillon <sargun@sargun.me>,
	linux-kernel@vger.kernel.org, Kees Cook <keescook@chromium.org>
Subject: Re: [PATCH v6 3/3] io_uring: allow disabling rings during the creation
Date: Tue, 8 Sep 2020 08:11:20 -0600	[thread overview]
Message-ID: <d1b18bdb-e718-3285-cfb3-0e2b73f03554@kernel.dk> (raw)
In-Reply-To: <20200908141003.wsm6pclfj6tsaffr@steredhat>

On 9/8/20 8:10 AM, Stefano Garzarella wrote:
> On Tue, Sep 08, 2020 at 07:57:08AM -0600, Jens Axboe wrote:
>> On 9/8/20 7:44 AM, Stefano Garzarella wrote:
>>> Hi Jens,
>>>
>>> On Thu, Aug 27, 2020 at 04:58:31PM +0200, Stefano Garzarella wrote:
>>>> This patch adds a new IORING_SETUP_R_DISABLED flag to start the
>>>> rings disabled, allowing the user to register restrictions,
>>>> buffers, files, before to start processing SQEs.
>>>>
>>>> When IORING_SETUP_R_DISABLED is set, SQE are not processed and
>>>> SQPOLL kthread is not started.
>>>>
>>>> The restrictions registration are allowed only when the rings
>>>> are disable to prevent concurrency issue while processing SQEs.
>>>>
>>>> The rings can be enabled using IORING_REGISTER_ENABLE_RINGS
>>>> opcode with io_uring_register(2).
>>>>
>>>> Suggested-by: Jens Axboe <axboe@kernel.dk>
>>>> Reviewed-by: Kees Cook <keescook@chromium.org>
>>>> Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
>>>> ---
>>>> v4:
>>>>  - fixed io_uring_enter() exit path when ring is disabled
>>>>
>>>> v3:
>>>>  - enabled restrictions only when the rings start
>>>>
>>>> RFC v2:
>>>>  - removed return value of io_sq_offload_start()
>>>> ---
>>>>  fs/io_uring.c                 | 52 ++++++++++++++++++++++++++++++-----
>>>>  include/uapi/linux/io_uring.h |  2 ++
>>>>  2 files changed, 47 insertions(+), 7 deletions(-)
>>>>
>>>> diff --git a/fs/io_uring.c b/fs/io_uring.c
>>>> index 5f62997c147b..b036f3373fbe 100644
>>>> --- a/fs/io_uring.c
>>>> +++ b/fs/io_uring.c
>>>> @@ -226,6 +226,7 @@ struct io_restriction {
>>>>  	DECLARE_BITMAP(sqe_op, IORING_OP_LAST);
>>>>  	u8 sqe_flags_allowed;
>>>>  	u8 sqe_flags_required;
>>>> +	bool registered;
>>>>  };
>>>>  
>>>>  struct io_ring_ctx {
>>>> @@ -7497,8 +7498,8 @@ static int io_init_wq_offload(struct io_ring_ctx *ctx,
>>>>  	return ret;
>>>>  }
>>>>  
>>>> -static int io_sq_offload_start(struct io_ring_ctx *ctx,
>>>> -			       struct io_uring_params *p)
>>>> +static int io_sq_offload_create(struct io_ring_ctx *ctx,
>>>> +				struct io_uring_params *p)
>>>>  {
>>>>  	int ret;
>>>>  
>>>> @@ -7532,7 +7533,6 @@ static int io_sq_offload_start(struct io_ring_ctx *ctx,
>>>>  			ctx->sqo_thread = NULL;
>>>>  			goto err;
>>>>  		}
>>>> -		wake_up_process(ctx->sqo_thread);
>>>>  	} else if (p->flags & IORING_SETUP_SQ_AFF) {
>>>>  		/* Can't have SQ_AFF without SQPOLL */
>>>>  		ret = -EINVAL;
>>>> @@ -7549,6 +7549,12 @@ static int io_sq_offload_start(struct io_ring_ctx *ctx,
>>>>  	return ret;
>>>>  }
>>>>  
>>>> +static void io_sq_offload_start(struct io_ring_ctx *ctx)
>>>> +{
>>>> +	if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sqo_thread)
>>>> +		wake_up_process(ctx->sqo_thread);
>>>> +}
>>>> +
>>>>  static inline void __io_unaccount_mem(struct user_struct *user,
>>>>  				      unsigned long nr_pages)
>>>>  {
>>>> @@ -8295,6 +8301,9 @@ SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
>>>>  	if (!percpu_ref_tryget(&ctx->refs))
>>>>  		goto out_fput;
>>>>  
>>>> +	if (ctx->flags & IORING_SETUP_R_DISABLED)
>>>> +		goto out_fput;
>>>> +
>>>
>>> While writing the man page paragraph, I discovered that if the rings are
>>> disabled I returned ENXIO error in io_uring_enter(), coming from the previous
>>> check.
>>>
>>> I'm not sure it is the best one, maybe I can return EBADFD or another
>>> error.
>>>
>>> What do you suggest?
>>
>> EBADFD seems indeed the most appropriate - the fd is valid, but not in the
>> right state to do this.
> 
> Yeah, the same interpretation as mine!
> 
> Also, in io_uring_register() I'm returning EINVAL if the rings are not
> disabled and the user wants to register restrictions.
> Maybe also in this case I can return EBADFD.

Yes let's do that, EINVAL is always way too overloaded, and it makes sense
to use EBADFD consistently for any operation related to that.

-- 
Jens Axboe