2018-07-02 21:04:06

by Tarick Bedeir

[permalink] [raw]
Subject: [PATCH] IB/mlx4: Test port number before querying type.

rdma_ah_find_type() can reach into ib_device->port_immutable with a
potentially out-of-bounds port number, so check that the port number is
valid first.

Fixes: 44c58487d51a ("IB/core: Define 'ib' and 'roce' rdma_ah_attr types")
Signed-off-by: Tarick Bedeir <[email protected]>
---
drivers/infiniband/hw/mlx4/qp.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index f045491f2c14..4f1dabecf9a5 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -4034,9 +4034,9 @@ static void to_rdma_ah_attr(struct mlx4_ib_dev *ibdev,
u8 port_num = path->sched_queue & 0x40 ? 2 : 1;

memset(ah_attr, 0, sizeof(*ah_attr));
- ah_attr->type = rdma_ah_find_type(&ibdev->ib_dev, port_num);
if (port_num == 0 || port_num > dev->caps.num_ports)
return;
+ ah_attr->type = rdma_ah_find_type(&ibdev->ib_dev, port_num);

if (ah_attr->type == RDMA_AH_ATTR_TYPE_ROCE)
rdma_ah_set_sl(ah_attr, ((path->sched_queue >> 3) & 0x7) |
--
2.18.0.399.gad0ab374a1-goog



2018-07-03 06:22:28

by Leon Romanovsky

[permalink] [raw]
Subject: Re: [PATCH] IB/mlx4: Test port number before querying type.

On Mon, Jul 02, 2018 at 02:02:34PM -0700, Tarick Bedeir wrote:
> rdma_ah_find_type() can reach into ib_device->port_immutable with a
> potentially out-of-bounds port number, so check that the port number is
> valid first.
>
> Fixes: 44c58487d51a ("IB/core: Define 'ib' and 'roce' rdma_ah_attr types")
> Signed-off-by: Tarick Bedeir <[email protected]>
> ---
> drivers/infiniband/hw/mlx4/qp.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
> index f045491f2c14..4f1dabecf9a5 100644
> --- a/drivers/infiniband/hw/mlx4/qp.c
> +++ b/drivers/infiniband/hw/mlx4/qp.c
> @@ -4034,9 +4034,9 @@ static void to_rdma_ah_attr(struct mlx4_ib_dev *ibdev,
> u8 port_num = path->sched_queue & 0x40 ? 2 : 1;
>
> memset(ah_attr, 0, sizeof(*ah_attr));
> - ah_attr->type = rdma_ah_find_type(&ibdev->ib_dev, port_num);
> if (port_num == 0 || port_num > dev->caps.num_ports)

Did you get any warning from any checker about out-of-bounds access?

According to the line above, port_num can be 2 or 1 which are valid ports for mlx4
and the check above doesn't do much.

> return;
> + ah_attr->type = rdma_ah_find_type(&ibdev->ib_dev, port_num);
>
> if (ah_attr->type == RDMA_AH_ATTR_TYPE_ROCE)
> rdma_ah_set_sl(ah_attr, ((path->sched_queue >> 3) & 0x7) |
> --
> 2.18.0.399.gad0ab374a1-goog
>


Attachments:
(No filename) (1.38 kB)
signature.asc (817.00 B)
Download all attachments

2018-07-03 14:45:45

by Tarick Bedeir

[permalink] [raw]
Subject: Re: [PATCH] IB/mlx4: Test port number before querying type.

On Mon, Jul 2, 2018 at 11:21 PM Leon Romanovsky <[email protected]> wrote:
>
> On Mon, Jul 02, 2018 at 02:02:34PM -0700, Tarick Bedeir wrote:
> > rdma_ah_find_type() can reach into ib_device->port_immutable with a
> > potentially out-of-bounds port number, so check that the port number is
> > valid first.
> >
> > Fixes: 44c58487d51a ("IB/core: Define 'ib' and 'roce' rdma_ah_attr types")
> > Signed-off-by: Tarick Bedeir <[email protected]>
> > ---
> > drivers/infiniband/hw/mlx4/qp.c | 2 +-
> > 1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
> > index f045491f2c14..4f1dabecf9a5 100644
> > --- a/drivers/infiniband/hw/mlx4/qp.c
> > +++ b/drivers/infiniband/hw/mlx4/qp.c
> > @@ -4034,9 +4034,9 @@ static void to_rdma_ah_attr(struct mlx4_ib_dev *ibdev,
> > u8 port_num = path->sched_queue & 0x40 ? 2 : 1;
> >
> > memset(ah_attr, 0, sizeof(*ah_attr));
> > - ah_attr->type = rdma_ah_find_type(&ibdev->ib_dev, port_num);
> > if (port_num == 0 || port_num > dev->caps.num_ports)
>
> Did you get any warning from any checker about out-of-bounds access?

Indeed, I observed a KASAN warning for an out-of-bounds read in
to_rdma_ah_attr().

> According to the line above, port_num can be 2 or 1 which are valid ports for mlx4
> and the check above doesn't do much.

There are mlx4 devices with only one port.

>
> > return;
> > + ah_attr->type = rdma_ah_find_type(&ibdev->ib_dev, port_num);
> >
> > if (ah_attr->type == RDMA_AH_ATTR_TYPE_ROCE)
> > rdma_ah_set_sl(ah_attr, ((path->sched_queue >> 3) & 0x7) |
> > --
> > 2.18.0.399.gad0ab374a1-goog
> >

2018-07-04 11:26:41

by Leon Romanovsky

[permalink] [raw]
Subject: Re: [PATCH] IB/mlx4: Test port number before querying type.

On Tue, Jul 03, 2018 at 07:44:25AM -0700, Tarick Bedeir wrote:
> On Mon, Jul 2, 2018 at 11:21 PM Leon Romanovsky <[email protected]> wrote:
> >
> > On Mon, Jul 02, 2018 at 02:02:34PM -0700, Tarick Bedeir wrote:
> > > rdma_ah_find_type() can reach into ib_device->port_immutable with a
> > > potentially out-of-bounds port number, so check that the port number is
> > > valid first.
> > >
> > > Fixes: 44c58487d51a ("IB/core: Define 'ib' and 'roce' rdma_ah_attr types")
> > > Signed-off-by: Tarick Bedeir <[email protected]>
> > > ---
> > > drivers/infiniband/hw/mlx4/qp.c | 2 +-
> > > 1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
> > > index f045491f2c14..4f1dabecf9a5 100644
> > > --- a/drivers/infiniband/hw/mlx4/qp.c
> > > +++ b/drivers/infiniband/hw/mlx4/qp.c
> > > @@ -4034,9 +4034,9 @@ static void to_rdma_ah_attr(struct mlx4_ib_dev *ibdev,
> > > u8 port_num = path->sched_queue & 0x40 ? 2 : 1;
> > >
> > > memset(ah_attr, 0, sizeof(*ah_attr));
> > > - ah_attr->type = rdma_ah_find_type(&ibdev->ib_dev, port_num);
> > > if (port_num == 0 || port_num > dev->caps.num_ports)
> >
> > Did you get any warning from any checker about out-of-bounds access?
>
> Indeed, I observed a KASAN warning for an out-of-bounds read in
> to_rdma_ah_attr().
>
> > According to the line above, port_num can be 2 or 1 which are valid ports for mlx4
> > and the check above doesn't do much.
>
> There are mlx4 devices with only one port.

Thanks, I never saw such device, but found one on ebay :).

Reviewed-by: Leon Romanovsky <[email protected]>


Attachments:
(No filename) (1.65 kB)
signature.asc (817.00 B)
Download all attachments