Hi,
here is the second version of the patch-set to implement
proper access checks into the io-page-fault handlers of the
AMD IOMMU and Intel VT-d drivers.
Two additional patches clean up the AMD part a bit further.
Since I can't test this this code myself due to lack of
hardware or software that utilizes it, I'd appreciate some
external testing.
It took me a while to get these out, mostly because I tried
to setup my own HSA test environment to at least test the
AMD changes myself. That failed, so I am sending this out
with another request for testing.
Oded, Jesse, would you two please test these patches and
report back? Thanks a lot!
Joerg
Changes since v1:
* Updated the access_error functions based on
Linus' feedback
* Rebased to v4.4-rc1
Joerg Roedel (4):
iommu/amd: Do proper access checking before calling handle_mm_fault()
iommu/amd: Correctly set flags for handle_mm_fault call
iommu/amd: Cleanup error handling in do_fault()
iommu/vt-d: Do access checks before calling handle_mm_fault()
drivers/iommu/amd_iommu_v2.c | 54 +++++++++++++++++++++++++++-----------------
drivers/iommu/intel-svm.c | 20 ++++++++++++++++
2 files changed, 53 insertions(+), 21 deletions(-)
--
1.9.1
From: Joerg Roedel <[email protected]>
The handle_mm_fault function expects the caller to do the
access checks. Not doing so and calling the function with
wrong permissions is a bug (catched by a BUG_ON).
So fix this bug by adding proper access checking to the io
page-fault code in the AMD IOMMUv2 driver.
Signed-off-by: Joerg Roedel <[email protected]>
---
drivers/iommu/amd_iommu_v2.c | 20 ++++++++++++++++++--
1 file changed, 18 insertions(+), 2 deletions(-)
diff --git a/drivers/iommu/amd_iommu_v2.c b/drivers/iommu/amd_iommu_v2.c
index d21d4ed..7caf2fa 100644
--- a/drivers/iommu/amd_iommu_v2.c
+++ b/drivers/iommu/amd_iommu_v2.c
@@ -494,6 +494,22 @@ static void handle_fault_error(struct fault *fault)
}
}
+static bool access_error(struct vm_area_struct *vma, struct fault *fault)
+{
+ unsigned long requested = 0;
+
+ if (fault->flags & PPR_FAULT_EXEC)
+ requested |= VM_EXEC;
+
+ if (fault->flags & PPR_FAULT_READ)
+ requested |= VM_READ;
+
+ if (fault->flags & PPR_FAULT_WRITE)
+ requested |= VM_WRITE;
+
+ return (requested & ~vma->vm_flags) != 0;
+}
+
static void do_fault(struct work_struct *work)
{
struct fault *fault = container_of(work, struct fault, work);
@@ -516,8 +532,8 @@ static void do_fault(struct work_struct *work)
goto out;
}
- if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))) {
- /* handle_mm_fault would BUG_ON() */
+ /* Check if we have the right permissions on the vma */
+ if (access_error(vma, fault)) {
up_read(&mm->mmap_sem);
handle_fault_error(fault);
goto out;
--
1.9.1
From: Joerg Roedel <[email protected]>
Instead of just checking for a write access, calculate the
flags that are passed to handle_mm_fault() more precisly and
use the pre-defined macros.
Signed-off-by: Joerg Roedel <[email protected]>
---
drivers/iommu/amd_iommu_v2.c | 14 +++++++++-----
1 file changed, 9 insertions(+), 5 deletions(-)
diff --git a/drivers/iommu/amd_iommu_v2.c b/drivers/iommu/amd_iommu_v2.c
index 7caf2fa..a7edbd6 100644
--- a/drivers/iommu/amd_iommu_v2.c
+++ b/drivers/iommu/amd_iommu_v2.c
@@ -513,16 +513,20 @@ static bool access_error(struct vm_area_struct *vma, struct fault *fault)
static void do_fault(struct work_struct *work)
{
struct fault *fault = container_of(work, struct fault, work);
- struct mm_struct *mm;
struct vm_area_struct *vma;
+ unsigned int flags = 0;
+ struct mm_struct *mm;
u64 address;
- int ret, write;
-
- write = !!(fault->flags & PPR_FAULT_WRITE);
+ int ret;
mm = fault->state->mm;
address = fault->address;
+ if (fault->flags & PPR_FAULT_USER)
+ flags |= FAULT_FLAG_USER;
+ if (fault->flags & PPR_FAULT_WRITE)
+ flags |= FAULT_FLAG_WRITE;
+
down_read(&mm->mmap_sem);
vma = find_extend_vma(mm, address);
if (!vma || address < vma->vm_start) {
@@ -539,7 +543,7 @@ static void do_fault(struct work_struct *work)
goto out;
}
- ret = handle_mm_fault(mm, vma, address, write);
+ ret = handle_mm_fault(mm, vma, address, flags);
if (ret & VM_FAULT_ERROR) {
/* failed to service fault */
up_read(&mm->mmap_sem);
--
1.9.1
From: Joerg Roedel <[email protected]>
Get rid of the three error paths that look the same and move
error handling to a single place.
Signed-off-by: Joerg Roedel <[email protected]>
---
drivers/iommu/amd_iommu_v2.c | 24 ++++++++----------------
1 file changed, 8 insertions(+), 16 deletions(-)
diff --git a/drivers/iommu/amd_iommu_v2.c b/drivers/iommu/amd_iommu_v2.c
index a7edbd6..6a28b74 100644
--- a/drivers/iommu/amd_iommu_v2.c
+++ b/drivers/iommu/amd_iommu_v2.c
@@ -514,10 +514,10 @@ static void do_fault(struct work_struct *work)
{
struct fault *fault = container_of(work, struct fault, work);
struct vm_area_struct *vma;
+ int ret = VM_FAULT_ERROR;
unsigned int flags = 0;
struct mm_struct *mm;
u64 address;
- int ret;
mm = fault->state->mm;
address = fault->address;
@@ -529,31 +529,23 @@ static void do_fault(struct work_struct *work)
down_read(&mm->mmap_sem);
vma = find_extend_vma(mm, address);
- if (!vma || address < vma->vm_start) {
+ if (!vma || address < vma->vm_start)
/* failed to get a vma in the right range */
- up_read(&mm->mmap_sem);
- handle_fault_error(fault);
goto out;
- }
/* Check if we have the right permissions on the vma */
- if (access_error(vma, fault)) {
- up_read(&mm->mmap_sem);
- handle_fault_error(fault);
+ if (access_error(vma, fault))
goto out;
- }
ret = handle_mm_fault(mm, vma, address, flags);
- if (ret & VM_FAULT_ERROR) {
- /* failed to service fault */
- up_read(&mm->mmap_sem);
- handle_fault_error(fault);
- goto out;
- }
+out:
up_read(&mm->mmap_sem);
-out:
+ if (ret & VM_FAULT_ERROR)
+ /* failed to service fault */
+ handle_fault_error(fault);
+
finish_pri_tag(fault->dev_state, fault->state, fault->tag);
put_pasid_state(fault->state);
--
1.9.1
From: Joerg Roedel <[email protected]>
Not doing so is a bug and might trigger a BUG_ON in
handle_mm_fault(). So add the proper permission checks
before calling into mm code.
Signed-off-by: Joerg Roedel <[email protected]>
---
drivers/iommu/intel-svm.c | 20 ++++++++++++++++++++
1 file changed, 20 insertions(+)
diff --git a/drivers/iommu/intel-svm.c b/drivers/iommu/intel-svm.c
index c69e3f9..5046483 100644
--- a/drivers/iommu/intel-svm.c
+++ b/drivers/iommu/intel-svm.c
@@ -484,6 +484,23 @@ struct page_req_dsc {
};
#define PRQ_RING_MASK ((0x1000 << PRQ_ORDER) - 0x10)
+
+static bool access_error(struct vm_area_struct *vma, struct page_req_dsc *req)
+{
+ unsigned long requested = 0;
+
+ if (req->exe_req)
+ requested |= VM_EXEC;
+
+ if (req->rd_req)
+ requested |= VM_READ;
+
+ if (req->wr_req)
+ requested |= VM_WRITE;
+
+ return (requested & ~vma->vm_flags) != 0;
+}
+
static irqreturn_t prq_event_thread(int irq, void *d)
{
struct intel_iommu *iommu = d;
@@ -539,6 +556,9 @@ static irqreturn_t prq_event_thread(int irq, void *d)
if (!vma || address < vma->vm_start)
goto invalid;
+ if (access_error(vma, req))
+ goto invalid;
+
ret = handle_mm_fault(svm->mm, vma, address,
req->wr_req ? FAULT_FLAG_WRITE : 0);
if (ret & VM_FAULT_ERROR)
--
1.9.1
On Tue, Nov 17, 2015 at 5:11 PM, Joerg Roedel <[email protected]> wrote:
> Hi,
>
> here is the second version of the patch-set to implement
> proper access checks into the io-page-fault handlers of the
> AMD IOMMU and Intel VT-d drivers.
>
> Two additional patches clean up the AMD part a bit further.
> Since I can't test this this code myself due to lack of
> hardware or software that utilizes it, I'd appreciate some
> external testing.
>
> It took me a while to get these out, mostly because I tried
> to setup my own HSA test environment to at least test the
> AMD changes myself. That failed, so I am sending this out
> with another request for testing.
>
> Oded, Jesse, would you two please test these patches and
> report back? Thanks a lot!
>
>
> Joerg
>
> Changes since v1:
>
> * Updated the access_error functions based on
> Linus' feedback
> * Rebased to v4.4-rc1
>
> Joerg Roedel (4):
> iommu/amd: Do proper access checking before calling handle_mm_fault()
> iommu/amd: Correctly set flags for handle_mm_fault call
> iommu/amd: Cleanup error handling in do_fault()
> iommu/vt-d: Do access checks before calling handle_mm_fault()
>
> drivers/iommu/amd_iommu_v2.c | 54 +++++++++++++++++++++++++++-----------------
> drivers/iommu/intel-svm.c | 20 ++++++++++++++++
> 2 files changed, 53 insertions(+), 21 deletions(-)
>
> --
> 1.9.1
>
Hi Joerg,
I have very limited testing capabilities since I left AMD - I only
have Kaveri (which is kind of obsolete) and I only have a very simple
test utility.
>From what I could check, I didn't see any problems, but this is really
such a very basic check that I don't even want to presume to give a
Tested-by tag. Sorry.
I suggest you talk to AMD people, as they have access to Carrizo H/W
and S/W testing framework.
Oded
On 11/17/2015 07:11 AM, Joerg Roedel wrote:
> Hi,
>
> here is the second version of the patch-set to implement
> proper access checks into the io-page-fault handlers of the
> AMD IOMMU and Intel VT-d drivers.
>
> Two additional patches clean up the AMD part a bit further.
> Since I can't test this this code myself due to lack of
> hardware or software that utilizes it, I'd appreciate some
> external testing.
>
> It took me a while to get these out, mostly because I tried
> to setup my own HSA test environment to at least test the
> AMD changes myself. That failed, so I am sending this out
> with another request for testing.
>
> Oded, Jesse, would you two please test these patches and
> report back? Thanks a lot!
I might be able to find time later this week or next; I have to update
my tree to David's latest bits and rebase my i915 code on top along with
these patches. Then I can run some of my tests and expand them to try
r/w tests w/o mapping the buffer with the right access mode.
Jesse
On Tue, 2015-11-17 at 16:11 +0100, Joerg Roedel wrote:
> From: Joerg Roedel <[email protected]>
>
> Not doing so is a bug and might trigger a BUG_ON in
> handle_mm_fault(). So add the proper permission checks
> before calling into mm code.
>
> Signed-off-by: Joerg Roedel <[email protected]>
Acked-By: David Woodhouse <[email protected]>
As noted, I can't test it right now until I've replaced my test
hardware. Certainly looks better than the last version though :)
Thanks.
--
David Woodhouse Open Source Technology Centre
[email protected] Intel Corporation
Hey Jesse,
On Wed, Nov 18, 2015 at 08:13:56AM -0800, Jesse Barnes wrote:
>
> I might be able to find time later this week or next; I have to update
> my tree to David's latest bits and rebase my i915 code on top along with
> these patches. Then I can run some of my tests and expand them to try
> r/w tests w/o mapping the buffer with the right access mode.
Have you found time to run some small tests on this?
Thanks,
Joerg
On Fri, Nov 20, 2015 at 03:35:53PM +0000, David Woodhouse wrote:
> Acked-By: David Woodhouse <[email protected]>
>
> As noted, I can't test it right now until I've replaced my test
> hardware. Certainly looks better than the last version though :)
Okay, thanks. I applied the series to the iommu tree. Patch 1 and 4 as
fixes, the others the the x86/amd branch.