epic_close() calls epic_rx() and uses dma buffer, but in epic_remove_one()
we already freed the dma buffer. To fix this issue, reorder function calls
like in the .probe function.
BUG: KASAN: use-after-free in epic_rx+0xa6/0x7e0 [epic100]
Call Trace:
epic_rx+0xa6/0x7e0 [epic100]
epic_close+0xec/0x2f0 [epic100]
unregister_netdev+0x18/0x20
epic_remove_one+0xaa/0xf0 [epic100]
Fixes: ae150435b59e ("smsc: Move the SMC (SMSC) drivers")
Reported-by: Yilun Wu <[email protected]>
Signed-off-by: Tong Zhang <[email protected]>
---
drivers/net/ethernet/smsc/epic100.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/net/ethernet/smsc/epic100.c b/drivers/net/ethernet/smsc/epic100.c
index a0654e88444c..0329caf63279 100644
--- a/drivers/net/ethernet/smsc/epic100.c
+++ b/drivers/net/ethernet/smsc/epic100.c
@@ -1515,14 +1515,14 @@ static void epic_remove_one(struct pci_dev *pdev)
struct net_device *dev = pci_get_drvdata(pdev);
struct epic_private *ep = netdev_priv(dev);
+ unregister_netdev(dev);
dma_free_coherent(&pdev->dev, TX_TOTAL_SIZE, ep->tx_ring,
ep->tx_ring_dma);
dma_free_coherent(&pdev->dev, RX_TOTAL_SIZE, ep->rx_ring,
ep->rx_ring_dma);
- unregister_netdev(dev);
pci_iounmap(pdev, ep->ioaddr);
- pci_release_regions(pdev);
free_netdev(dev);
+ pci_release_regions(pdev);
pci_disable_device(pdev);
/* pci_power_off(pdev, -1); */
}
--
2.25.1
Tong Zhang <[email protected]> :
> epic_close() calls epic_rx() and uses dma buffer, but in epic_remove_one()
> we already freed the dma buffer. To fix this issue, reorder function calls
> like in the .probe function.
>
> BUG: KASAN: use-after-free in epic_rx+0xa6/0x7e0 [epic100]
> Call Trace:
> epic_rx+0xa6/0x7e0 [epic100]
> epic_close+0xec/0x2f0 [epic100]
> unregister_netdev+0x18/0x20
> epic_remove_one+0xaa/0xf0 [epic100]
>
> Fixes: ae150435b59e ("smsc: Move the SMC (SMSC) drivers")
> Reported-by: Yilun Wu <[email protected]>
> Signed-off-by: Tong Zhang <[email protected]>
The "Fixes:" tag is a bit misleading: this code path predates the move
by several years. Ignoring pci_* vs dma_* API changes, this is pre-2005
material.
Reviewed-by: Francois Romieu <[email protected]>
--
Ueimor
On Thu, 23 Jun 2022 11:22:28 +0200 Francois Romieu wrote:
> Tong Zhang <[email protected]> :
> > epic_close() calls epic_rx() and uses dma buffer, but in epic_remove_one()
> > we already freed the dma buffer. To fix this issue, reorder function calls
> > like in the .probe function.
> >
> > BUG: KASAN: use-after-free in epic_rx+0xa6/0x7e0 [epic100]
> > Call Trace:
> > epic_rx+0xa6/0x7e0 [epic100]
> > epic_close+0xec/0x2f0 [epic100]
> > unregister_netdev+0x18/0x20
> > epic_remove_one+0xaa/0xf0 [epic100]
> >
> > Fixes: ae150435b59e ("smsc: Move the SMC (SMSC) drivers")
> > Reported-by: Yilun Wu <[email protected]>
> > Signed-off-by: Tong Zhang <[email protected]>
>
> The "Fixes:" tag is a bit misleading: this code path predates the move
> by several years. Ignoring pci_* vs dma_* API changes, this is pre-2005
> material.
Yeah, please find the correct Fixes tag.
> Reviewed-by: Francois Romieu <[email protected]>
Keep Francois' tag when reposting.
On Fri, Jun 24, 2022 at 11:41 AM Jakub Kicinski <[email protected]> wrote:
>
> On Thu, 23 Jun 2022 11:22:28 +0200 Francois Romieu wrote:
> > Tong Zhang <[email protected]> :
> > > epic_close() calls epic_rx() and uses dma buffer, but in epic_remove_one()
> > > we already freed the dma buffer. To fix this issue, reorder function calls
> > > like in the .probe function.
> > >
> > > BUG: KASAN: use-after-free in epic_rx+0xa6/0x7e0 [epic100]
> > > Call Trace:
> > > epic_rx+0xa6/0x7e0 [epic100]
> > > epic_close+0xec/0x2f0 [epic100]
> > > unregister_netdev+0x18/0x20
> > > epic_remove_one+0xaa/0xf0 [epic100]
> > >
> > > Fixes: ae150435b59e ("smsc: Move the SMC (SMSC) drivers")
> > > Reported-by: Yilun Wu <[email protected]>
> > > Signed-off-by: Tong Zhang <[email protected]>
> >
> > The "Fixes:" tag is a bit misleading: this code path predates the move
> > by several years. Ignoring pci_* vs dma_* API changes, this is pre-2005
> > material.
>
> Yeah, please find the correct Fixes tag.
>
> > Reviewed-by: Francois Romieu <[email protected]>
>
> Keep Francois' tag when reposting.
Looks like drivers/net/ethernet/smsc/epic100.c is renamed from
drivers/net/epic100.c
and this bug has been around since the very initial commit. What would
you suggest ?
Remove the fix tag or use
Fix: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Tong Zhang <[email protected]> :
[...]
> Looks like drivers/net/ethernet/smsc/epic100.c is renamed from
> drivers/net/epic100.c and this bug has been around since the very
> initial commit.
> What would you suggest ? Remove the fix tag or use
> Fix: 1da177e4c3f4 ("Linux-2.6.12-rc2")
'Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")' has already been signed-by or
directly used by maintainers, including netdev ones. See:
$ git log --grep='^Fixes: 1da177e4c3f4'
The patch will require some change for pre-5.9 stable kernels due to
63692803899b563f94bf1b4f821b574eb74316ae "epic100: switch from 'pci_' to 'dma_' API".
--
Ueimor
epic_close() calls epic_rx() and uses dma buffer, but in epic_remove_one()
we already freed the dma buffer. To fix this issue, reorder function calls
like in the .probe function.
BUG: KASAN: use-after-free in epic_rx+0xa6/0x7e0 [epic100]
Call Trace:
epic_rx+0xa6/0x7e0 [epic100]
epic_close+0xec/0x2f0 [epic100]
unregister_netdev+0x18/0x20
epic_remove_one+0xaa/0xf0 [epic100]
Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Reported-by: Yilun Wu <[email protected]>
Signed-off-by: Tong Zhang <[email protected]>
Reviewed-by: Francois Romieu <[email protected]>
---
v2: amend fix tag
drivers/net/ethernet/smsc/epic100.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/net/ethernet/smsc/epic100.c b/drivers/net/ethernet/smsc/epic100.c
index a0654e88444c..0329caf63279 100644
--- a/drivers/net/ethernet/smsc/epic100.c
+++ b/drivers/net/ethernet/smsc/epic100.c
@@ -1515,14 +1515,14 @@ static void epic_remove_one(struct pci_dev *pdev)
struct net_device *dev = pci_get_drvdata(pdev);
struct epic_private *ep = netdev_priv(dev);
+ unregister_netdev(dev);
dma_free_coherent(&pdev->dev, TX_TOTAL_SIZE, ep->tx_ring,
ep->tx_ring_dma);
dma_free_coherent(&pdev->dev, RX_TOTAL_SIZE, ep->rx_ring,
ep->rx_ring_dma);
- unregister_netdev(dev);
pci_iounmap(pdev, ep->ioaddr);
- pci_release_regions(pdev);
free_netdev(dev);
+ pci_release_regions(pdev);
pci_disable_device(pdev);
/* pci_power_off(pdev, -1); */
}
--
2.25.1
On Sun, Jun 26, 2022 at 8:07 AM Francois Romieu <[email protected]> wrote:
>
> Tong Zhang <[email protected]> :
> [...]
> > Looks like drivers/net/ethernet/smsc/epic100.c is renamed from
> > drivers/net/epic100.c and this bug has been around since the very
> > initial commit.
> > What would you suggest ? Remove the fix tag or use
> > Fix: 1da177e4c3f4 ("Linux-2.6.12-rc2")
>
> 'Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")' has already been signed-by or
> directly used by maintainers, including netdev ones. See:
>
> $ git log --grep='^Fixes: 1da177e4c3f4'
>
> The patch will require some change for pre-5.9 stable kernels due to
> 63692803899b563f94bf1b4f821b574eb74316ae "epic100: switch from 'pci_' to 'dma_' API".
>
> --
> Ueimor
Thanks! I amended the fix tag and sent it as v2.
- Tong
Hello:
This patch was applied to netdev/net.git (master)
by Jakub Kicinski <[email protected]>:
On Sun, 26 Jun 2022 21:33:48 -0700 you wrote:
> epic_close() calls epic_rx() and uses dma buffer, but in epic_remove_one()
> we already freed the dma buffer. To fix this issue, reorder function calls
> like in the .probe function.
>
> BUG: KASAN: use-after-free in epic_rx+0xa6/0x7e0 [epic100]
> Call Trace:
> epic_rx+0xa6/0x7e0 [epic100]
> epic_close+0xec/0x2f0 [epic100]
> unregister_netdev+0x18/0x20
> epic_remove_one+0xaa/0xf0 [epic100]
>
> [...]
Here is the summary with links:
- [v2] epic100: fix use after free on rmmod
https://git.kernel.org/netdev/net/c/8ee9d82cd0a4
You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html