* [PATCH] RDMA/mlx5: Check the DM type before mapping a MEMIC operation
@ 2026-11-15 10:43 Hungyu Lin
2026-11-16 13:22 ` Leon Romanovsky
2026-11-16 13:22 ` Leon Romanovsky
0 siblings, 2 replies; 3+ messages in thread
From: Hungyu Lin @ 2026-11-15 10:43 UTC (permalink / raw)
To: leon; +Cc: jgg, maorg, linux-rdma, linux-kernel, dennylin0707
MLX5_IB_METHOD_DM_MAP_OP_ADDR accepts a handle to any DM object and
converts it with to_memic() without checking the DM type. The handle can
refer to a SW ICM object, which is a struct mlx5_ib_dm_icm and is
smaller than struct mlx5_ib_dm_memic.
The handler then uses the ops_xa_lock, ops and ref members, which are
all located past the end of the mlx5_ib_dm_icm allocation, so it locks
and modifies memory that belongs to other slab objects.
Return -EOPNOTSUPP for DM objects that are not MEMIC, as
MLX5_IB_METHOD_DM_QUERY already does.
Fixes: cea85fa5dbc2 ("RDMA/mlx5: Add support in MEMIC operations")
Cc: stable@vger.kernel.org
Signed-off-by: Hungyu Lin <dennylin0707@gmail.com>
---
drivers/infiniband/hw/mlx5/dm.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/drivers/infiniband/hw/mlx5/dm.c b/drivers/infiniband/hw/mlx5/dm.c
index 7a6fe4fea3e2..16e42fd6c802 100644
--- a/drivers/infiniband/hw/mlx5/dm.c
+++ b/drivers/infiniband/hw/mlx5/dm.c
@@ -213,6 +213,9 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_DM_MAP_OP_ADDR)(
int err;
u8 op;
+ if (dm->base.type != MLX5_IB_UAPI_DM_TYPE_MEMIC)
+ return -EOPNOTSUPP;
+
err = uverbs_copy_from(&op, attrs, MLX5_IB_ATTR_DM_MAP_OP_ADDR_REQ_OP);
if (err)
return err;
--
2.43.0
^ permalink raw reply [flat|nested] 3+ messages in thread* Re: [PATCH] RDMA/mlx5: Check the DM type before mapping a MEMIC operation
2026-11-15 10:43 [PATCH] RDMA/mlx5: Check the DM type before mapping a MEMIC operation Hungyu Lin
@ 2026-11-16 13:22 ` Leon Romanovsky
2026-11-16 13:22 ` Leon Romanovsky
1 sibling, 0 replies; 3+ messages in thread
From: Leon Romanovsky @ 2026-11-16 13:22 UTC (permalink / raw)
To: Hungyu Lin; +Cc: jgg, maorg, linux-rdma, linux-kernel
On Sun, Nov 15, 2026 at 10:43:42AM +0000, Hungyu Lin wrote:
> MLX5_IB_METHOD_DM_MAP_OP_ADDR accepts a handle to any DM object and
> converts it with to_memic() without checking the DM type. The handle can
> refer to a SW ICM object, which is a struct mlx5_ib_dm_icm and is
> smaller than struct mlx5_ib_dm_memic.
>
> The handler then uses the ops_xa_lock, ops and ref members, which are
> all located past the end of the mlx5_ib_dm_icm allocation, so it locks
> and modifies memory that belongs to other slab objects.
>
> Return -EOPNOTSUPP for DM objects that are not MEMIC, as
> MLX5_IB_METHOD_DM_QUERY already does.
>
> Fixes: cea85fa5dbc2 ("RDMA/mlx5: Add support in MEMIC operations")
> Cc: stable@vger.kernel.org
> Signed-off-by: Hungyu Lin <dennylin0707@gmail.com>
> ---
> drivers/infiniband/hw/mlx5/dm.c | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/drivers/infiniband/hw/mlx5/dm.c b/drivers/infiniband/hw/mlx5/dm.c
> index 7a6fe4fea3e2..16e42fd6c802 100644
> --- a/drivers/infiniband/hw/mlx5/dm.c
> +++ b/drivers/infiniband/hw/mlx5/dm.c
> @@ -213,6 +213,9 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_DM_MAP_OP_ADDR)(
> int err;
> u8 op;
>
> + if (dm->base.type != MLX5_IB_UAPI_DM_TYPE_MEMIC)
> + return -EOPNOTSUPP;
> +
> err = uverbs_copy_from(&op, attrs, MLX5_IB_ATTR_DM_MAP_OP_ADDR_REQ_OP);
> if (err)
> return err;
I applied this patch with the following change:
diff --git a/drivers/infiniband/hw/mlx5/dm.c b/drivers/infiniband/hw/mlx5/dm.c
index 6a1971b9627b1..7d1e642a8d4c9 100644
--- a/drivers/infiniband/hw/mlx5/dm.c
+++ b/drivers/infiniband/hw/mlx5/dm.c
@@ -208,12 +208,13 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_DM_MAP_OP_ADDR)(
attrs, MLX5_IB_ATTR_DM_MAP_OP_ADDR_REQ_HANDLE);
struct mlx5_ib_dev *dev = to_mdev(uobj->context->device);
struct ib_dm *ibdm = uobj->object;
+ struct mlx5_ib_dm *mdm = to_mdm(ibdm);
struct mlx5_ib_dm_memic *dm = to_memic(ibdm);
struct mlx5_ib_dm_op_entry *op_entry;
int err;
u8 op;
- if (dm->base.type != MLX5_IB_UAPI_DM_TYPE_MEMIC)
+ if (mdm->base.type != MLX5_IB_UAPI_DM_TYPE_MEMIC)
return -EOPNOTSUPP;
err = uverbs_copy_from(&op, attrs, MLX5_IB_ATTR_DM_MAP_OP_ADDR_REQ_OP);
Thanks
> --
> 2.43.0
^ permalink raw reply [flat|nested] 3+ messages in thread* Re: [PATCH] RDMA/mlx5: Check the DM type before mapping a MEMIC operation
2026-11-15 10:43 [PATCH] RDMA/mlx5: Check the DM type before mapping a MEMIC operation Hungyu Lin
2026-11-16 13:22 ` Leon Romanovsky
@ 2026-11-16 13:22 ` Leon Romanovsky
1 sibling, 0 replies; 3+ messages in thread
From: Leon Romanovsky @ 2026-11-16 13:22 UTC (permalink / raw)
To: Hungyu Lin; +Cc: jgg, maorg, linux-rdma, linux-kernel
On Sun, 15 Nov 2026 10:43:42 +0000, Hungyu Lin wrote:
> MLX5_IB_METHOD_DM_MAP_OP_ADDR accepts a handle to any DM object and
> converts it with to_memic() without checking the DM type. The handle can
> refer to a SW ICM object, which is a struct mlx5_ib_dm_icm and is
> smaller than struct mlx5_ib_dm_memic.
>
> The handler then uses the ops_xa_lock, ops and ref members, which are
> all located past the end of the mlx5_ib_dm_icm allocation, so it locks
> and modifies memory that belongs to other slab objects.
>
> [...]
Applied, thanks!
[1/1] RDMA/mlx5: Check the DM type before mapping a MEMIC operation
https://git.kernel.org/rdma/rdma/c/ea97d0a83f4ed6
Best regards,
--
Leon Romanovsky <leon@kernel.org>
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2026-11-16 13:22 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-11-15 10:43 [PATCH] RDMA/mlx5: Check the DM type before mapping a MEMIC operation Hungyu Lin
2026-11-16 13:22 ` Leon Romanovsky
2026-11-16 13:22 ` Leon Romanovsky
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®