Skip to contentSkip to conversation

Two buffer overruns in device assignment

PATCH In review Markus Armbruster 3 participants 8 messages Feb 21, 2014 – Feb 24, 2014

lore
Share & tools
lore
#0 lore raw

cover Two buffer overruns in device assignment

Markus Armbruster (2):
vfio: Fix overrun after readlink() fills buffer completely
pci-assign: Fix potential read beyond buffer on -EBUSY

 hw/i386/kvm/pci-assign.c | 1 +
 hw/misc/vfio.c           | 2 +-
 2 files changed, 2 insertions(+), 1 deletion(-)
Signature · 1 line
1.8.1.4
#1 lore raw

PATCH 1/2 vfio: Fix overrun after readlink() fills buffer completely

readlink() returns the number of bytes written to the buffer, and it doesn't write a terminating null byte. vfio_init() writes it itself. Overruns the buffer when readlink() filled it completely.

Fix by reserving space for the null byte when calling readlink(), like we do elsewhere.

Spotted by Coverity.

---
 hw/misc/vfio.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
hw/misc/vfio.c +1 -1
index 8db182f..8e56785 100644
@@ -3681,7 +3681,7 @@ static int vfio_initfn(PCIDevice *pdev)
 
     strncat(path, "iommu_group", sizeof(path) - strlen(path) - 1);
 
-    len = readlink(path, iommu_group_path, PATH_MAX);
+    len = readlink(path, iommu_group_path, PATH_MAX - 1);
     if (len <= 0) {
         error_report("vfio: error no iommu_group for device");
         return -errno;
Signature · 1 line
1.8.1.4
Reviewed-by on v1 1/2
#2 lore raw

Re: [Qemu-devel] [PATCH 1/2] vfio: Fix overrun after readlink() fills buffer completely

Quoted 19 lines · Markus Armbruster

readlink() returns the number of bytes written to the buffer, and it doesn't write a terminating null byte. vfio_init() writes it itself. Overruns the buffer when readlink() filled it completely.

Fix by reserving space for the null byte when calling readlink(), like we do elsewhere.

Spotted by Coverity.

---
 hw/misc/vfio.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
hw/misc/vfio.c +0 -0
index 8db182f..8e56785 100644

hw/misc/vfio.c static int vfio_initfn(PCIDevice *pdev) · line 3687

Quoted 8 lines
hw/misc/vfio.c +1 -1
@@ -3681,7 +3681,7 @@ static int vfio_initfn(PCIDevice *pdev)
 
     strncat(path, "iommu_group", sizeof(path) - strlen(path) - 1);
 
-    len = readlink(path, iommu_group_path, PATH_MAX);
+    len = readlink(path, iommu_group_path, PATH_MAX - 1);
     if (len <= 0) {
         error_report("vfio: error no iommu_group for device");
         return -errno;

"sizeof(iommu_group_path) - 1" would be slightly better, I think, but PATH_MAX - 1 works too.

(cc stable?)

thanks
-- PMM

#3 lore raw

Re: [Qemu-devel] [PATCH 1/2] vfio: Fix overrun after readlink() fills buffer completely

In reply to Markus Armbruster
Quoted 19 lines · Markus Armbruster

readlink() returns the number of bytes written to the buffer, and it doesn't write a terminating null byte. vfio_init() writes it itself. Overruns the buffer when readlink() filled it completely.

Fix by reserving space for the null byte when calling readlink(), like we do elsewhere.

Spotted by Coverity.

---
 hw/misc/vfio.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
hw/misc/vfio.c +0 -0
index 8db182f..8e56785 100644

hw/misc/vfio.c static int vfio_initfn(PCIDevice *pdev) · line 3687

Quoted 8 lines
hw/misc/vfio.c +1 -1
@@ -3681,7 +3681,7 @@ static int vfio_initfn(PCIDevice *pdev)
 
     strncat(path, "iommu_group", sizeof(path) - strlen(path) - 1);
 
-    len = readlink(path, iommu_group_path, PATH_MAX);
+    len = readlink(path, iommu_group_path, PATH_MAX - 1);
     if (len <= 0) {
         error_report("vfio: error no iommu_group for device");
         return -errno;

I'm not sure why we wouldn't follow the same logic as pci-assign here. If we fill to the length provided we have no idea if we have the full path. We certainly expect it to fit well within PATH_MAX, so let's handle it as an error if we reach PATH_MAX. Thanks,

Alex

#4 lore raw

Re: [Qemu-devel] [PATCH 1/2] vfio: Fix overrun after readlink() fills buffer completely

Quoted 23 lines · Alex Williamson
Quoted 22 lines · Markus Armbruster

readlink() returns the number of bytes written to the buffer, and it doesn't write a terminating null byte. vfio_init() writes it itself. Overruns the buffer when readlink() filled it completely.

Fix by reserving space for the null byte when calling readlink(), like we do elsewhere.

Spotted by Coverity.

---
 hw/misc/vfio.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
hw/misc/vfio.c +0 -0
index 8db182f..8e56785 100644
@@ -3681,7 +3681,7 @@ static int vfio_initfn(PCIDevice *pdev)
 
     strncat(path, "iommu_group", sizeof(path) - strlen(path) - 1);
 
Quoted 10 lines

hw/misc/vfio.c static int vfio_initfn(PCIDevice *pdev) · line 3687

Quoted 5 lines
hw/misc/vfio.c +1 -1
-    len = readlink(path, iommu_group_path, PATH_MAX);
+    len = readlink(path, iommu_group_path, PATH_MAX - 1);
     if (len <= 0) {
         error_report("vfio: error no iommu_group for device");
         return -errno;

I'm not sure why we wouldn't follow the same logic as pci-assign here. If we fill to the length provided we have no idea if we have the full path. We certainly expect it to fit well within PATH_MAX, so let's handle it as an error if we reach PATH_MAX. Thanks,

Fair point. v2 sent.

#5 lore raw

PATCH 2/2 pci-assign: Fix potential read beyond buffer on -EBUSY

In reply to Markus Armbruster

readlink() doesn't write a terminating null byte. assign_failed_examine() passes the unterminated string to strrchr(). Oops. Terminate it.

Spotted by Coverity.

---
 hw/i386/kvm/pci-assign.c | 1 +
 1 file changed, 1 insertion(+)
hw/i386/kvm/pci-assign.c +1 -0
index 9686801..a825871 100644
@@ -743,6 +743,7 @@ static void assign_failed_examine(AssignedDevice *dev)
         goto fail;
     }
 
+    driver[r] = 0;
     ns = strrchr(driver, '/');
     if (!ns) {
         goto fail;
Signature · 1 line
1.8.1.4
#6 lore raw

Re: [Qemu-devel] [PATCH 2/2] pci-assign: Fix potential read beyond buffer on -EBUSY

Quoted 16 lines · Markus Armbruster

readlink() doesn't write a terminating null byte. assign_failed_examine() passes the unterminated string to strrchr(). Oops. Terminate it.

Spotted by Coverity.

---
 hw/i386/kvm/pci-assign.c | 1 +
 1 file changed, 1 insertion(+)
hw/i386/kvm/pci-assign.c +0 -0
index 9686801..a825871 100644

hw/i386/kvm/pci-assign.c static void assign_failed_examine(AssignedDevice *dev) · line 746

Quoted 4 lines
hw/i386/kvm/pci-assign.c +1 -0
@@ -743,6 +743,7 @@ static void assign_failed_examine(AssignedDevice *dev)
         goto fail;
     }
 
+    driver[r] = 0;

This will write off the end of the buffer if readlink() filled it completely, won't it? I think you also need to change the readlink() 3rd argument to "sizeof(driver) - 1".

thanks
-- PMM

Reviewed-by
#7 lore raw

Re: [Qemu-devel] [PATCH 2/2] pci-assign: Fix potential read beyond buffer on -EBUSY

Quoted 20 lines · Peter Maydell
Quoted 19 lines · Markus Armbruster

readlink() doesn't write a terminating null byte. assign_failed_examine() passes the unterminated string to strrchr(). Oops. Terminate it.

Spotted by Coverity.

---
 hw/i386/kvm/pci-assign.c | 1 +
 1 file changed, 1 insertion(+)
hw/i386/kvm/pci-assign.c +0 -0
index 9686801..a825871 100644
@@ -743,6 +743,7 @@ static void assign_failed_examine(AssignedDevice *dev)
         goto fail;
     }
 
Quoted 5 lines

hw/i386/kvm/pci-assign.c static void assign_failed_examine(AssignedDevice *dev) · line 746

Quoted 1 line
hw/i386/kvm/pci-assign.c +1 -0
+    driver[r] = 0;

This will write off the end of the buffer if readlink() filled it completely, won't it? I think you also need to change the readlink() 3rd argument to "sizeof(driver) - 1".

Apologies for this aspersion -- we check for "r == sizeof(driver)" and bail out before we get here, so the patch is ok.

thanks
-- PMM