[PATCH] net: atlantic: Avoid out-of-bounds indexing

Kai-Heng Feng posted 1 patch 4 years, 3 months ago
There is a newer version of this series
.../net/ethernet/aquantia/atlantic/aq_vec.c   | 24 +++++++++----------
1 file changed, 12 insertions(+), 12 deletions(-)
[PATCH] net: atlantic: Avoid out-of-bounds indexing
Posted by Kai-Heng Feng 4 years, 3 months ago
UBSAN warnings are observed on atlantic driver:
[ 294.432996] UBSAN: array-index-out-of-bounds in /build/linux-Qow4fL/linux-5.15.0/drivers/net/ethernet/aquantia/atlantic/aq_nic.c:484:48
[ 294.433695] index 8 is out of range for type 'aq_vec_s *[8]'

The index is assigned right before breaking out the loop, so there's no actual
deferencing happening.

So only use the index inside the loop to fix the issue.

BugLink: https://bugs.launchpad.net/bugs/1958770
Tested-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
---
 .../net/ethernet/aquantia/atlantic/aq_vec.c   | 24 +++++++++----------
 1 file changed, 12 insertions(+), 12 deletions(-)

diff --git a/drivers/net/ethernet/aquantia/atlantic/aq_vec.c b/drivers/net/ethernet/aquantia/atlantic/aq_vec.c
index f4774cf051c97..6ab1f3212d246 100644
--- a/drivers/net/ethernet/aquantia/atlantic/aq_vec.c
+++ b/drivers/net/ethernet/aquantia/atlantic/aq_vec.c
@@ -43,8 +43,8 @@ static int aq_vec_poll(struct napi_struct *napi, int budget)
 	if (!self) {
 		err = -EINVAL;
 	} else {
-		for (i = 0U, ring = self->ring[0];
-			self->tx_rings > i; ++i, ring = self->ring[i]) {
+		for (i = 0U; self->tx_rings > i; ++i) {
+			ring = self->ring[i];
 			u64_stats_update_begin(&ring[AQ_VEC_RX_ID].stats.rx.syncp);
 			ring[AQ_VEC_RX_ID].stats.rx.polls++;
 			u64_stats_update_end(&ring[AQ_VEC_RX_ID].stats.rx.syncp);
@@ -182,8 +182,8 @@ int aq_vec_init(struct aq_vec_s *self, const struct aq_hw_ops *aq_hw_ops,
 	self->aq_hw_ops = aq_hw_ops;
 	self->aq_hw = aq_hw;
 
-	for (i = 0U, ring = self->ring[0];
-		self->tx_rings > i; ++i, ring = self->ring[i]) {
+	for (i = 0U; self->tx_rings > i; ++i) {
+		ring = self->ring[i];
 		err = aq_ring_init(&ring[AQ_VEC_TX_ID], ATL_RING_TX);
 		if (err < 0)
 			goto err_exit;
@@ -224,8 +224,8 @@ int aq_vec_start(struct aq_vec_s *self)
 	unsigned int i = 0U;
 	int err = 0;
 
-	for (i = 0U, ring = self->ring[0];
-		self->tx_rings > i; ++i, ring = self->ring[i]) {
+	for (i = 0U; self->tx_rings > i; ++i) {
+		ring = self->ring[i];
 		err = self->aq_hw_ops->hw_ring_tx_start(self->aq_hw,
 							&ring[AQ_VEC_TX_ID]);
 		if (err < 0)
@@ -248,8 +248,8 @@ void aq_vec_stop(struct aq_vec_s *self)
 	struct aq_ring_s *ring = NULL;
 	unsigned int i = 0U;
 
-	for (i = 0U, ring = self->ring[0];
-		self->tx_rings > i; ++i, ring = self->ring[i]) {
+	for (i = 0U; self->tx_rings > i; ++i) {
+		ring = self->ring[i];
 		self->aq_hw_ops->hw_ring_tx_stop(self->aq_hw,
 						 &ring[AQ_VEC_TX_ID]);
 
@@ -268,8 +268,8 @@ void aq_vec_deinit(struct aq_vec_s *self)
 	if (!self)
 		goto err_exit;
 
-	for (i = 0U, ring = self->ring[0];
-		self->tx_rings > i; ++i, ring = self->ring[i]) {
+	for (i = 0U; self->tx_rings > i; ++i) {
+		ring = self->ring[i];
 		aq_ring_tx_clean(&ring[AQ_VEC_TX_ID]);
 		aq_ring_rx_deinit(&ring[AQ_VEC_RX_ID]);
 	}
@@ -297,8 +297,8 @@ void aq_vec_ring_free(struct aq_vec_s *self)
 	if (!self)
 		goto err_exit;
 
-	for (i = 0U, ring = self->ring[0];
-		self->tx_rings > i; ++i, ring = self->ring[i]) {
+	for (i = 0U; self->tx_rings > i; ++i) {
+		ring = self->ring[i];
 		aq_ring_free(&ring[AQ_VEC_TX_ID]);
 		if (i < self->rx_rings)
 			aq_ring_free(&ring[AQ_VEC_RX_ID]);
-- 
2.34.1
Re: [PATCH] net: atlantic: Avoid out-of-bounds indexing
Posted by Jakub Kicinski 4 years, 3 months ago
On Fri,  4 Mar 2022 13:08:12 +0800 Kai-Heng Feng wrote:
> UBSAN warnings are observed on atlantic driver:
> [ 294.432996] UBSAN: array-index-out-of-bounds in /build/linux-Qow4fL/linux-5.15.0/drivers/net/ethernet/aquantia/atlantic/aq_nic.c:484:48
> [ 294.433695] index 8 is out of range for type 'aq_vec_s *[8]'
> 
> The index is assigned right before breaking out the loop, so there's no actual
> deferencing happening.

I think you're underselling it. This codes does not compute the address
of the ring, it reads the ring pointer from an array. The description
reads like it was doing:

	ring = &self->ring[i];

which would indeed be valid. What the code actually does is not.

Please repost with the commit message improved and a Fixes: tag(s)
pointing to commit(s) where the buggy code was introduced.

> So only use the index inside the loop to fix the issue.
> 
> BugLink: https://bugs.launchpad.net/bugs/1958770
> Tested-by: Mario Limonciello <mario.limonciello@amd.com>
> Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
> ---
>  .../net/ethernet/aquantia/atlantic/aq_vec.c   | 24 +++++++++----------
>  1 file changed, 12 insertions(+), 12 deletions(-)
> 
> diff --git a/drivers/net/ethernet/aquantia/atlantic/aq_vec.c b/drivers/net/ethernet/aquantia/atlantic/aq_vec.c
> index f4774cf051c97..6ab1f3212d246 100644
> --- a/drivers/net/ethernet/aquantia/atlantic/aq_vec.c
> +++ b/drivers/net/ethernet/aquantia/atlantic/aq_vec.c
> @@ -43,8 +43,8 @@ static int aq_vec_poll(struct napi_struct *napi, int budget)
>  	if (!self) {
>  		err = -EINVAL;
>  	} else {
> -		for (i = 0U, ring = self->ring[0];
> -			self->tx_rings > i; ++i, ring = self->ring[i]) {
> +		for (i = 0U; self->tx_rings > i; ++i) {
> +			ring = self->ring[i];
>  			u64_stats_update_begin(&ring[AQ_VEC_RX_ID].stats.rx.syncp);
>  			ring[AQ_VEC_RX_ID].stats.rx.polls++;
>  			u64_stats_update_end(&ring[AQ_VEC_RX_ID].stats.rx.syncp);