2020-06-14 17:01:51

by Zekun Shen

Subject: [PATCH] net: alx: fix race condition in alx_remove

There is a race condition exist during termination. The path is
alx_stop and then alx_remove. An alx_schedule_link_check could be called
before alx_stop and invoke alx_link_check later. Alx_stop frees the
napis, and alx_remove cancels any pending works. If any of the work is
scheduled before termination and invoked before alx_remove, a
null-ptr-deref occurs because both expect alx->napis[i].

This patch fix the race condition by moving cancel_work_sync functions
before alx_free_napis inside alx_stop.

Signed-off-by: Zekun Shen <[email protected]>
---
drivers/net/ethernet/atheros/alx/main.c | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/net/ethernet/atheros/alx/main.c b/drivers/net/ethernet/atheros/alx/main.c
index b9b4edb91..1487d6765 100644
--- a/drivers/net/ethernet/atheros/alx/main.c
+++ b/drivers/net/ethernet/atheros/alx/main.c
@@ -1249,6 +1249,9 @@ static int __alx_open(struct alx_priv *alx, bool resume)

static void __alx_stop(struct alx_priv *alx)
{
+ cancel_work_sync(&alx->link_check_wk);
+ cancel_work_sync(&alx->reset_wk);
+
alx_halt(alx);
alx_free_irq(alx);
alx_free_rings(alx);
@@ -1855,9 +1858,6 @@ static void alx_remove(struct pci_dev *pdev)
struct alx_priv *alx = pci_get_drvdata(pdev);
struct alx_hw *hw = &alx->hw;

- cancel_work_sync(&alx->link_check_wk);
- cancel_work_sync(&alx->reset_wk);
-
/* restore permanent mac address */
alx_set_macaddr(hw, hw->perm_addr);

--
2.17.1

2020-06-15 15:53:24

by Zekun Shen

Subject: [PATCH v2] net: alx: fix race condition in alx_remove

There is a race condition exist during termination. The path is
alx_stop and then alx_remove. An alx_schedule_link_check could be called
before alx_stop by interrupt handler and invoke alx_link_check later.
Alx_stop frees the napis, and alx_remove cancels any pending works.
If any of the work is scheduled before termination and invoked before
alx_remove, a null-ptr-deref occurs because both expect alx->napis[i].

This patch fix the race condition by moving cancel_work_sync functions
before alx_free_napis inside alx_stop. Because interrupt handler can call
alx_schedule_link_check again, alx_free_irq is moved before
cancel_work_sync calls too.

Signed-off-by: Zekun Shen <[email protected]>
---
Changes in v2:
- move alx_free_irq before cancel_work_sync calls
- prevent another interrupt to cause race condition

drivers/net/ethernet/atheros/alx/main.c | 9 +++++----
1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/drivers/net/ethernet/atheros/alx/main.c b/drivers/net/ethernet/atheros/alx/main.c
index b9b4edb91..9b7f1af5f 100644
--- a/drivers/net/ethernet/atheros/alx/main.c
+++ b/drivers/net/ethernet/atheros/alx/main.c
@@ -1249,8 +1249,12 @@ static int __alx_open(struct alx_priv *alx, bool resume)

static void __alx_stop(struct alx_priv *alx)
{
- alx_halt(alx);
alx_free_irq(alx);
+
+ cancel_work_sync(&alx->link_check_wk);
+ cancel_work_sync(&alx->reset_wk);
+
+ alx_halt(alx);
alx_free_rings(alx);
alx_free_napis(alx);
}
@@ -1855,9 +1859,6 @@ static void alx_remove(struct pci_dev *pdev)
struct alx_priv *alx = pci_get_drvdata(pdev);
struct alx_hw *hw = &alx->hw;

- cancel_work_sync(&alx->link_check_wk);
- cancel_work_sync(&alx->reset_wk);
-
/* restore permanent mac address */
alx_set_macaddr(hw, hw->perm_addr);

--
2.17.1

2020-06-15 20:24:33

by David Miller

Subject: Re: [PATCH v2] net: alx: fix race condition in alx_remove

From: Zekun Shen <[email protected]>
Date: Mon, 15 Jun 2020 11:50:29 -0400

> There is a race condition exist during termination. The path is
> alx_stop and then alx_remove. An alx_schedule_link_check could be called
> before alx_stop by interrupt handler and invoke alx_link_check later.
> Alx_stop frees the napis, and alx_remove cancels any pending works.
> If any of the work is scheduled before termination and invoked before
> alx_remove, a null-ptr-deref occurs because both expect alx->napis[i].
>
> This patch fix the race condition by moving cancel_work_sync functions
> before alx_free_napis inside alx_stop. Because interrupt handler can call
> alx_schedule_link_check again, alx_free_irq is moved before
> cancel_work_sync calls too.
>
> Signed-off-by: Zekun Shen <[email protected]>

Applied, thank you.