Re: [PATCH] RDMA/irdma: Restore original PD on rereg_user_mr failure

Re: [PATCH] RDMA/irdma: Restore original PD on rereg_user_mr failure

[Leon Romanovsky]

On Wed, Oct 28, 2026 at 11:07:01AM +0100, Mohamad Pishdar wrote:
> Hi,
> 
> While auditing the memory-region re-registration path in the irdma driver I
> found a case where irdma_rereg_user_mr() can leave a struct ib_mr pointing
> at a PD that the uverbs core no longer considers it to hold a reference on,
> if the reregistration itself fails partway through. I think this is worth a
> look before someone hits it in the field, so I'm sending the analysis and a
> proposed fix together rather than just the report.

Thanks for the bug report. Please submit a patch following the guidance in
Documentation/process/submitting-patches.rst. There is no need to CC the
security@ mailing list; this is a regular bug.

Thanks

Re: [PATCH] RDMA/irdma: Restore original PD on rereg_user_mr failure

[Jacob Moroni]

Hi,

Looks like we discovered the same thing at around the same time.

I think this patch which was just merged should fix it:

https://lore.kernel.org/linux-rdma/179371200572.1726055.11640512427251640287.b4-ty@kernel.org/T/#m8e7c4a912cb2539b08e2603cf751fcf16d13b291

Thanks,
Jake

Re: [PATCH] RDMA/irdma: Restore original PD on rereg_user_mr failure

[Leon Romanovsky]

Yes, i just reading and answering in FIFO mode.

On Tue, Nov 3, 2026, at 15:39, Jacob Moroni wrote:
> Hi,
>
> Looks like we discovered the same thing at around the same time.
>
> I think this patch which was just merged should fix it:
>
> https://lore.kernel.org/linux-rdma/179371200572.1726055.11640512427251640287.b4-ty@kernel.org/T/#m8e7c4a912cb2539b08e2603cf751fcf16d13b291
>
> Thanks,
> Jake