2023-09-25 06:23:49

by Dinghao Liu

Subject: [PATCH] PCI: xgene-msi: Fix a potential UAF in xgene_msi_probe

xgene_allocate_domains() will call irq_domain_remove() to free
msi->inner_domain on failure. However, its caller, xgene_msi_probe(),
will also call irq_domain_remove() through xgene_msi_remove() on the
same failure, which may lead to a use-after-free. Set the freed pointer
to NULL to fix this issue.

Fixes: dcd19de36775 ("PCI: xgene: Add APM X-Gene v1 PCIe MSI/MSIX termination driver")
Signed-off-by: Dinghao Liu <[email protected]>
---
drivers/pci/controller/pci-xgene-msi.c | 1 +
1 file changed, 1 insertion(+)

diff --git a/drivers/pci/controller/pci-xgene-msi.c b/drivers/pci/controller/pci-xgene-msi.c
index 3ce38dfd0d29..c0192c5ff0f3 100644
--- a/drivers/pci/controller/pci-xgene-msi.c
+++ b/drivers/pci/controller/pci-xgene-msi.c
@@ -253,6 +253,7 @@ static int xgene_allocate_domains(struct xgene_msi *msi)

if (!msi->msi_domain) {
irq_domain_remove(msi->inner_domain);
+ msi->inner_domain = NULL;
return -ENOMEM;
}

--
2.17.1

2023-09-25 10:26:51

by Marc Zyngier

Subject: Re: [PATCH] PCI: xgene-msi: Fix a potential UAF in xgene_msi_probe

On Mon, 25 Sep 2023 07:21:32 +0100,
Dinghao Liu <[email protected]> wrote:
>
> xgene_allocate_domains() will call irq_domain_remove() to free
> msi->inner_domain on failure. However, its caller, xgene_msi_probe(),
> will also call irq_domain_remove() through xgene_msi_remove() on the
> same failure, which may lead to a use-after-free. Set the freed pointer
> to NULL to fix this issue.
>
> Fixes: dcd19de36775 ("PCI: xgene: Add APM X-Gene v1 PCIe MSI/MSIX termination driver")
> Signed-off-by: Dinghao Liu <[email protected]>
> ---
> drivers/pci/controller/pci-xgene-msi.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/drivers/pci/controller/pci-xgene-msi.c b/drivers/pci/controller/pci-xgene-msi.c
> index 3ce38dfd0d29..c0192c5ff0f3 100644
> --- a/drivers/pci/controller/pci-xgene-msi.c
> +++ b/drivers/pci/controller/pci-xgene-msi.c
> @@ -253,6 +253,7 @@ static int xgene_allocate_domains(struct xgene_msi *msi)
>
> if (!msi->msi_domain) {
> irq_domain_remove(msi->inner_domain);
> + msi->inner_domain = NULL;
> return -ENOMEM;
> }

Why can't we just drop the irq_domain_remove() call here instead, and
simply rely on xgene_msi_remove() to do the right thing? Something
like the untested patch below.

Thanks,

M.

diff --git a/drivers/pci/controller/pci-xgene-msi.c b/drivers/pci/controller/pci-xgene-msi.c
index 0234e528b9a5..f98c9eb7bebf 100644
--- a/drivers/pci/controller/pci-xgene-msi.c
+++ b/drivers/pci/controller/pci-xgene-msi.c
@@ -251,10 +251,8 @@ static int xgene_allocate_domains(struct xgene_msi *msi)
&xgene_msi_domain_info,
msi->inner_domain);

- if (!msi->msi_domain) {
- irq_domain_remove(msi->inner_domain);
+ if (!msi->msi_domain)
return -ENOMEM;
- }

return 0;
}

--
Without deviation from the norm, progress is not possible.

2023-09-26 02:25:11

by Dinghao Liu

Subject: Re: [PATCH] PCI: xgene-msi: Fix a potential UAF in xgene_msi_probe

> On Mon, 25 Sep 2023 07:21:32 +0100,
> Dinghao Liu <[email protected]> wrote:
> >
> > xgene_allocate_domains() will call irq_domain_remove() to free
> > msi->inner_domain on failure. However, its caller, xgene_msi_probe(),
> > will also call irq_domain_remove() through xgene_msi_remove() on the
> > same failure, which may lead to a use-after-free. Set the freed pointer
> > to NULL to fix this issue.
> >
> > Fixes: dcd19de36775 ("PCI: xgene: Add APM X-Gene v1 PCIe MSI/MSIX termination driver")
> > Signed-off-by: Dinghao Liu <[email protected]>
> > ---
> > drivers/pci/controller/pci-xgene-msi.c | 1 +
> > 1 file changed, 1 insertion(+)
> >
> > diff --git a/drivers/pci/controller/pci-xgene-msi.c b/drivers/pci/controller/pci-xgene-msi.c
> > index 3ce38dfd0d29..c0192c5ff0f3 100644
> > --- a/drivers/pci/controller/pci-xgene-msi.c
> > +++ b/drivers/pci/controller/pci-xgene-msi.c
> > @@ -253,6 +253,7 @@ static int xgene_allocate_domains(struct xgene_msi *msi)
> >
> > if (!msi->msi_domain) {
> > irq_domain_remove(msi->inner_domain);
> > + msi->inner_domain = NULL;
> > return -ENOMEM;
> > }
>
> Why can't we just drop the irq_domain_remove() call here instead, and
> simply rely on xgene_msi_remove() to do the right thing? Something
> like the untested patch below.
>
> Thanks,
>
> M.
>
> diff --git a/drivers/pci/controller/pci-xgene-msi.c b/drivers/pci/controller/pci-xgene-msi.c
> index 0234e528b9a5..f98c9eb7bebf 100644
> --- a/drivers/pci/controller/pci-xgene-msi.c
> +++ b/drivers/pci/controller/pci-xgene-msi.c
> @@ -251,10 +251,8 @@ static int xgene_allocate_domains(struct xgene_msi *msi)
> &xgene_msi_domain_info,
> msi->inner_domain);
>
> - if (!msi->msi_domain) {
> - irq_domain_remove(msi->inner_domain);
> + if (!msi->msi_domain)
> return -ENOMEM;
> - }
>
> return 0;
> }

Thanks for your advice! This patch is more concise. I will resend a new patch soon.

Regards,
Dinghao