diff mbox series

io_uring: don't allow duplicate registrations

Message ID x497ef4hzuj.fsf@segfault.boston.devel.redhat.com (mailing list archive)
State New, archived
Headers show
Series io_uring: don't allow duplicate registrations | expand

Commit Message

Jeff Moyer Jan. 17, 2019, 12:50 a.m. UTC
Hi, Jens,

It looks to me like calling io_uring_register more than once (for either
IORING_REGISTER_BUFFERS or IORING_REGISTER_FILES) will leak the
references taken in previous calls.

Signed-off-by: Jeff Moyer <jmoyer@redhat.com>

---

If this makes sense to you, feel free to just fold this into your
patches w/o any attribution.

Comments

Jens Axboe Jan. 17, 2019, 1:23 a.m. UTC | #1
On 1/16/19 5:50 PM, Jeff Moyer wrote:
> Hi, Jens,
> 
> It looks to me like calling io_uring_register more than once (for either
> IORING_REGISTER_BUFFERS or IORING_REGISTER_FILES) will leak the
> references taken in previous calls.

Oops, thanks for that. Let's make it -EBUSY though, everything ends up
being -EINVAL and that sometimes makes it hard for an application to
figure out wtf is going on.

I also added -EINVAL for unregister when not already setup.
Jeff Moyer Jan. 17, 2019, 1:31 a.m. UTC | #2
Jens Axboe <axboe@kernel.dk> writes:

> On 1/16/19 5:50 PM, Jeff Moyer wrote:
>> Hi, Jens,
>> 
>> It looks to me like calling io_uring_register more than once (for either
>> IORING_REGISTER_BUFFERS or IORING_REGISTER_FILES) will leak the
>> references taken in previous calls.
>
> Oops, thanks for that. Let's make it -EBUSY though, everything ends up
> being -EINVAL and that sometimes makes it hard for an application to
> figure out wtf is going on.

Sounds good.

> I also added -EINVAL for unregister when not already setup.

Yep, I saw that.

-Jeff
Jens Axboe Jan. 17, 2019, 1:33 a.m. UTC | #3
On 1/16/19 6:31 PM, Jeff Moyer wrote:
> Jens Axboe <axboe@kernel.dk> writes:
> 
>> On 1/16/19 5:50 PM, Jeff Moyer wrote:
>>> Hi, Jens,
>>>
>>> It looks to me like calling io_uring_register more than once (for either
>>> IORING_REGISTER_BUFFERS or IORING_REGISTER_FILES) will leak the
>>> references taken in previous calls.
>>
>> Oops, thanks for that. Let's make it -EBUSY though, everything ends up
>> being -EINVAL and that sometimes makes it hard for an application to
>> figure out wtf is going on.
> 
> Sounds good.
> 
>> I also added -EINVAL for unregister when not already setup.
> 
> Yep, I saw that.

Changed it to -ENXIO, that's more descriptive.
diff mbox series

Patch

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 3650e8b63a32..f0492b0e23a0 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -1399,6 +1399,9 @@  static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg,
 	if (!nr_args)
 		return -EINVAL;
 
+	if (ctx->user_files)
+		return -EINVAL;
+
 	ctx->user_files = kcalloc(nr_args, sizeof(struct file *), GFP_KERNEL);
 	if (!ctx->user_files)
 		return -ENOMEM;
@@ -1580,6 +1583,9 @@  static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg,
 	if (!nr_args || nr_args > UIO_MAXIOV)
 		return -EINVAL;
 
+	if (ctx->user_bufs)
+		return -EINVAL;
+
 	ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf),
 					GFP_KERNEL);
 	if (!ctx->user_bufs)