2019-04-17 09:15:35

by Pan Bian

Subject: iommu/vt-d: drop mm use count if address is not canonical

The use count of svm->mm is incremented by mmget_not_zero. However, it
is not dropped when the address is not canonical. This patch fixes the
bug.

Fixes: 9d8c3af31607("iommu/vt-d: IOMMU Page Request needs to check if
address is canonical.")

Signed-off-by: Pan Bian <[email protected]>
---
drivers/iommu/intel-svm.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/iommu/intel-svm.c b/drivers/iommu/intel-svm.c
index 3a4b09a..2630d2e 100644
--- a/drivers/iommu/intel-svm.c
+++ b/drivers/iommu/intel-svm.c
@@ -574,8 +574,10 @@ static irqreturn_t prq_event_thread(int irq, void *d)
goto bad_req;

/* If address is not canonical, return invalid response */
- if (!is_canonical_address(address))
+ if (!is_canonical_address(address)) {
+ mmput(svm->mm);
goto bad_req;
+ }

down_read(&svm->mm->mmap_sem);
vma = find_extend_vma(svm->mm, address);
--
2.7.4

2019-04-26 13:41:36

by Joerg Roedel

Subject: Re: iommu/vt-d: drop mm use count if address is not canonical

[ Adding more people. ]

On Wed, Apr 17, 2019 at 05:12:57PM +0800, Pan Bian wrote:
> The use count of svm->mm is incremented by mmget_not_zero. However, it
> is not dropped when the address is not canonical. This patch fixes the
> bug.
>
> Fixes: 9d8c3af31607("iommu/vt-d: IOMMU Page Request needs to check if
> address is canonical.")
>
> Signed-off-by: Pan Bian <[email protected]>
> ---
> drivers/iommu/intel-svm.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/iommu/intel-svm.c b/drivers/iommu/intel-svm.c
> index 3a4b09a..2630d2e 100644
> --- a/drivers/iommu/intel-svm.c
> +++ b/drivers/iommu/intel-svm.c
> @@ -574,8 +574,10 @@ static irqreturn_t prq_event_thread(int irq, void *d)
> goto bad_req;
>
> /* If address is not canonical, return invalid response */
> - if (!is_canonical_address(address))
> + if (!is_canonical_address(address)) {
> + mmput(svm->mm);
> goto bad_req;
> + }
>
> down_read(&svm->mm->mmap_sem);
> vma = find_extend_vma(svm->mm, address);

I think it is better to move the canonical address check before
mmget_not_zero() like in the diff below. But I let David and Jacob
comment on this.

Regards,

Joerg

diff --git a/drivers/iommu/intel-svm.c b/drivers/iommu/intel-svm.c
index 8f87304f915c..5e9e1eee8336 100644
--- a/drivers/iommu/intel-svm.c
+++ b/drivers/iommu/intel-svm.c
@@ -586,14 +586,15 @@ static irqreturn_t prq_event_thread(int irq, void *d)
* any faults on kernel addresses. */
if (!svm->mm)
goto bad_req;
- /* If the mm is already defunct, don't handle faults. */
- if (!mmget_not_zero(svm->mm))
- goto bad_req;

/* If address is not canonical, return invalid response */
if (!is_canonical_address(address))
goto bad_req;

+ /* If the mm is already defunct, don't handle faults. */
+ if (!mmget_not_zero(svm->mm))
+ goto bad_req;
+
down_read(&svm->mm->mmap_sem);
vma = find_extend_vma(svm->mm, address);
if (!vma || address < vma->vm_start)

2019-04-26 15:46:10

by Jacob Pan

Subject: Re: iommu/vt-d: drop mm use count if address is not canonical

On Fri, 26 Apr 2019 15:38:46 +0200
Joerg Roedel <[email protected]> wrote:

> [ Adding more people. ]
>
> On Wed, Apr 17, 2019 at 05:12:57PM +0800, Pan Bian wrote:
> > The use count of svm->mm is incremented by mmget_not_zero. However,
> > it is not dropped when the address is not canonical. This patch
> > fixes the bug.
> >
> > Fixes: 9d8c3af31607("iommu/vt-d: IOMMU Page Request needs to check
> > if address is canonical.")
> >
> > Signed-off-by: Pan Bian <[email protected]>
> > ---
> > drivers/iommu/intel-svm.c | 4 +++-
> > 1 file changed, 3 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/iommu/intel-svm.c b/drivers/iommu/intel-svm.c
> > index 3a4b09a..2630d2e 100644
> > --- a/drivers/iommu/intel-svm.c
> > +++ b/drivers/iommu/intel-svm.c
> > @@ -574,8 +574,10 @@ static irqreturn_t prq_event_thread(int irq,
> > void *d) goto bad_req;
> >
> > /* If address is not canonical, return invalid
> > response */
> > - if (!is_canonical_address(address))
> > + if (!is_canonical_address(address)) {
> > + mmput(svm->mm);
> > goto bad_req;
> > + }
> >
> > down_read(&svm->mm->mmap_sem);
> > vma = find_extend_vma(svm->mm, address);
>
> I think it is better to move the canonical address check before
> mmget_not_zero() like in the diff below. But I let David and Jacob
> comment on this.
>
Looks good to me. Tested with PRQ and fake non canonical address then
send an invalid page response.

Tested-by: Jacob Pan <[email protected]>


> Regards,
>
> Joerg
>
> diff --git a/drivers/iommu/intel-svm.c b/drivers/iommu/intel-svm.c
> index 8f87304f915c..5e9e1eee8336 100644
> --- a/drivers/iommu/intel-svm.c
> +++ b/drivers/iommu/intel-svm.c
> @@ -586,14 +586,15 @@ static irqreturn_t prq_event_thread(int irq,
> void *d)
> * any faults on kernel addresses. */
> if (!svm->mm)
> goto bad_req;
> - /* If the mm is already defunct, don't handle
> faults. */
> - if (!mmget_not_zero(svm->mm))
> - goto bad_req;
>
> /* If address is not canonical, return invalid
> response */ if (!is_canonical_address(address))
> goto bad_req;
>
> + /* If the mm is already defunct, don't handle
> faults. */
> + if (!mmget_not_zero(svm->mm))
> + goto bad_req;
> +
> down_read(&svm->mm->mmap_sem);
> vma = find_extend_vma(svm->mm, address);
> if (!vma || address < vma->vm_start)

[Jacob Pan]