On 05-Mar-19 03:22, Jason Gunthorpe wrote:
> On Wed, Feb 27, 2019 at 10:13:46PM +0000, Saleem, Shiraz wrote:
>
>>> + pbl->phys.indirect.pbl_buf_size_in_pages = size_in_pages;
>>> + pbl->phys.indirect.sgl = sgl;
>>> + pbl->phys.indirect.sg_dma_cnt = sg_dma_cnt;
>>> + err = pbl_chunk_list_create(dev, pbl);
>>> + if (err) {
>>> + efa_err(&dev->ibdev.dev,
>>> + "chunk_list creation failed[%d]\n", err);
>>> + goto err_chunk;
>>> + }
>>
>> Seems like you have many error prints in this driver. Do you really want this to show up
>> in dmesg or just the critical ones? Consider dev or pr_dbg?
>
> As a general rule, no driver should have a path where unprivileged
> user space can cause unbounded dmesg prints.
>
> This is sort of a security problem.
As I said, auditing prints to debug.
>
> We have such a common request in drivers to do this though, I wonder
> if we should have a special core printer function for it that is
> automatically limited in some way.
>
> Jason
>
