On 16/11/2022 08:10, Yanjun Zhu wrote:
>>
>> index d4f10c2d1aa7..7c99d1591580 100644
>> --- a/drivers/infiniband/sw/rxe/rxe_mr.c
>> +++ b/drivers/infiniband/sw/rxe/rxe_mr.c
>> @@ -99,6 +99,7 @@ static int rxe_mr_alloc(struct rxe_mr *mr, int num_buf)
>> kfree(mr->map[i]);
>> kfree(mr->map);
>> + mr->map = NULL;
>> err1:
>> return -ENOMEM;
>> }
>> @@ -122,7 +123,6 @@ int rxe_mr_init_user(struct rxe_dev *rxe, u64
>> start, u64 length, u64 iova,
>> int num_buf;
>> void *vaddr;
>> int err;
>> - int i;
>> umem = ib_umem_get(&rxe->ib_dev, start, length, access);
>> if (IS_ERR(umem)) {
>> @@ -163,9 +163,8 @@ int rxe_mr_init_user(struct rxe_dev *rxe, u64
>> start, u64 length, u64 iova,
>> pr_warn("%s: Unable to get virtual address\n",
>> __func__);
>> err = -ENOMEM;
>> - goto err_cleanup_map;
>> + goto err_release_umem;
>
> This call trace results from page_address's returning NULL, then goto
> err_cleanup_map where mr->map[i] and mr->map are freed.
>
> And finally rxe_reg_user_mr gets an error from rxe_mr_init_user, the
> function rxe_mr_cleanup is called to handle mr to free mr->map[i] and
> mr->map again.
>
> So mr->map[i] and mr->map are double freed.
>
> As such, this commit is reasonable.
>
> But why page_address will return NULL?
ENOMEM? but I don't think we need taking too much care upon the reason.
this patch is most likely porting the reverted back, commit:
8ff5f5d9d8cf ("RDMA/rxe: Prevent double freeing rxe_map_set()")
Actually, the double free can be triggered by below error path too.
149 err = rxe_mr_alloc(mr, num_buf);
150 if (err) {
151 pr_warn("%s: Unable to allocate memory for map\n",
152 __func__);
153 goto err_release_umem;
154 }
where rxe_mr_alloc() freed the memory but don't set 'mr->map = NULL'
Thanks
Zhijian
>
> Zhu Yanjun
>
