Re: [PATCH] io_uring: fix page leak in io_sqe_buffer_register()

[Jens Axboe <axboe@kernel.dk>]

On 6/17/25 8:02 AM, Penglei Jiang wrote:
> On Tue, 17 Jun 2025 06:53:04 -0600, Jens Axboe wrote:
>> On 6/17/25 6:39 AM, Penglei Jiang wrote:
>>> Add missing unpin_user_pages() in the error path
>>>
>>> Fixes: d8c2237d0aa9 ("io_uring: add io_pin_pages() helper")
>>> Signed-off-by: Penglei Jiang <superman.xpt@gmail.com>
>>> ---
>>>  io_uring/rsrc.c | 4 +++-
>>>  1 file changed, 3 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/io_uring/rsrc.c b/io_uring/rsrc.c
>>> index c592ceace97d..f5ac1b530e21 100644
>>> --- a/io_uring/rsrc.c
>>> +++ b/io_uring/rsrc.c
>>> @@ -804,8 +804,10 @@ static struct io_rsrc_node *io_sqe_buffer_register(struct io_ring_ctx *ctx,
>>>  	}
>>>  
>>>  	imu = io_alloc_imu(ctx, nr_pages);
>>> -	if (!imu)
>>> +	if (!imu) {
>>> +		unpin_user_pages(pages, nr_pages);
>>>  		goto done;
>>> +	}
>>>  
>>>  	imu->nr_bvecs = nr_pages;
>>>  	ret = io_buffer_account_pin(ctx, pages, nr_pages, imu, last_hpage);
>>
>> Wouldn't it be better to have the unpin be part of the normal error
>> handling? Not sure why the pin accounting failure doesn't do that
>> already.
>>
>> Totally untested...
>>
>> diff --git a/io_uring/rsrc.c b/io_uring/rsrc.c
>> index 94a9db030e0e..a68f0cd677a3 100644
>> --- a/io_uring/rsrc.c
>> +++ b/io_uring/rsrc.c
>> @@ -809,10 +809,8 @@ static struct io_rsrc_node *io_sqe_buffer_register(struct io_ring_ctx *ctx,
>>  
>>  	imu->nr_bvecs = nr_pages;
>>  	ret = io_buffer_account_pin(ctx, pages, nr_pages, imu, last_hpage);
>> -	if (ret) {
>> -		unpin_user_pages(pages, nr_pages);
>> +	if (ret)
>>  		goto done;
>> -	}
>>  
>>  	size = iov->iov_len;
>>  	/* store original address for later verification */
>> @@ -840,6 +838,7 @@ static struct io_rsrc_node *io_sqe_buffer_register(struct io_ring_ctx *ctx,
>>  	}
>>  done:
>>  	if (ret) {
>> +		unpin_user_pages(pages, nr_pages);
>>  		if (imu)
>>  			io_free_imu(ctx, imu);
>>  		io_cache_free(&ctx->node_cache, node);
> 
> Thank you for taking the time to address this issue!
> 
> However, if io_pin_pages() fails, it will also jump to the done label,
> but at that point, the value of nr_pages is undefined because nr_pages
> is only assigned a value inside io_pin_pages() if it succeeds.
> 
> 	pages = io_pin_pages((unsigned long) iov->iov_base, iov->iov_len,
> 				&nr_pages);
> 	if (IS_ERR(pages)) {
> 		ret = PTR_ERR(pages);
> 		pages = NULL;
> 		goto done;
> 	}
> 
> 	...
> 
> 	done:
> 		if (ret) {
> 			unpin_user_pages(NULL, undefined-value);
> 			...
> 
> I'm not sure what the impact of calling unpin_user_pages() in this way would be.

We should just check for 'pages' being valid first. Updated below. If
you want to test and send a v2 based on that, I do think that's the
better approach as it keeps all the error handling consistent.


diff --git a/io_uring/rsrc.c b/io_uring/rsrc.c
index 94a9db030e0e..454cd8855c6c 100644
--- a/io_uring/rsrc.c
+++ b/io_uring/rsrc.c
@@ -809,10 +809,8 @@ static struct io_rsrc_node *io_sqe_buffer_register(struct io_ring_ctx *ctx,
 
 	imu->nr_bvecs = nr_pages;
 	ret = io_buffer_account_pin(ctx, pages, nr_pages, imu, last_hpage);
-	if (ret) {
-		unpin_user_pages(pages, nr_pages);
+	if (ret)
 		goto done;
-	}
 
 	size = iov->iov_len;
 	/* store original address for later verification */
@@ -840,6 +838,8 @@ static struct io_rsrc_node *io_sqe_buffer_register(struct io_ring_ctx *ctx,
 	}
 done:
 	if (ret) {
+		if (pages)
+			unpin_user_pages(pages, nr_pages);
 		if (imu)
 			io_free_imu(ctx, imu);
 		io_cache_free(&ctx->node_cache, node);

-- 
Jens Axboe