From: Colin Ian King <[email protected]>
Currently when the call to percpu_ref_init fails ctx->file_data is
set to null and because there is a missing return statement the
following statement dereferences this null pointer causing an oops.
Fix this by adding the missing -ENOMEM return to avoid the oops.
Addresses-Coverity: ("Explicit null dereference")
Fixes: cbb537634780 ("io_uring: avoid ring quiesce for fixed file set unregister and update")
Signed-off-by: Colin Ian King <[email protected]>
---
fs/io_uring.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/fs/io_uring.c b/fs/io_uring.c
index c756b8fc44c6..1d31294f5914 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -4937,6 +4937,7 @@ static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg,
kfree(ctx->file_data->table);
kfree(ctx->file_data);
ctx->file_data = NULL;
+ return -ENOMEM;
}
ctx->file_data->put_llist.first = NULL;
INIT_WORK(&ctx->file_data->ref_work, io_ring_file_ref_switch);
--
2.24.0
On 20/12/2019 23:48, Jens Axboe wrote:
> On 12/20/19 4:33 PM, Colin King wrote:
>> From: Colin Ian King <[email protected]>
>>
>> Currently when the call to percpu_ref_init fails ctx->file_data is
>> set to null and because there is a missing return statement the
>> following statement dereferences this null pointer causing an oops.
>> Fix this by adding the missing -ENOMEM return to avoid the oops.
>
> Nice, thanks! I'm guessing I didn't have the necessary magic debug
> options to allow failure injection for failing.
Fortunately we have Coverity to the rescue :-)
Colin
On 12/20/19 4:33 PM, Colin King wrote:
> From: Colin Ian King <[email protected]>
>
> Currently when the call to percpu_ref_init fails ctx->file_data is
> set to null and because there is a missing return statement the
> following statement dereferences this null pointer causing an oops.
> Fix this by adding the missing -ENOMEM return to avoid the oops.
Nice, thanks! I'm guessing I didn't have the necessary magic debug
options to allow failure injection for failing.
--
Jens Axboe
On 12/20/19 4:49 PM, Colin Ian King wrote:
> On 20/12/2019 23:48, Jens Axboe wrote:
>> On 12/20/19 4:33 PM, Colin King wrote:
>>> From: Colin Ian King <[email protected]>
>>>
>>> Currently when the call to percpu_ref_init fails ctx->file_data is
>>> set to null and because there is a missing return statement the
>>> following statement dereferences this null pointer causing an oops.
>>> Fix this by adding the missing -ENOMEM return to avoid the oops.
>>
>> Nice, thanks! I'm guessing I didn't have the necessary magic debug
>> options to allow failure injection for failing.
>
> Fortunately we have Coverity to the rescue :-)
Indeed!
--
Jens Axboe