[PULL,2/2] vfio/common: Fix vfio_iommu_type1_info use after free

Commit Message

Alex Williamson Sept. 27, 2022, 10:29 p.m. UTC

On error, vfio_get_iommu_info() frees and clears *info, but
vfio_connect_container() continues to use the pointer regardless
of the return value.  Restructure the code such that a failure
of this function triggers an error and clean up the remainder of
the function, including updating an outdated comment that had
drifted from its relevant line of code and using host page size
for a default for better compatibility on non-4KB systems.

Reported-by: Nicolin Chen <[e-mail]>
Link: https://lore.kernel.org/all/[e-mail]/
Signed-off-by: Alex Williamson <[e-mail]>
Reviewed-by: Philippe Mathieu-Daudé <[e-mail]>
Reviewed-by: Nicolin Chen <[e-mail]>
Tested-by: Nicolin Chen <[e-mail]>
Link: https://lore.kernel.org/r/166326219630.3388898.12882473157184946072.stgit@omen
Signed-off-by: Alex Williamson <[e-mail]>
---
 hw/vfio/common.c |   36 +++++++++++++++++++-----------------
 1 file changed, 19 insertions(+), 17 deletions(-)

Patch

diff --git a/hw/vfio/common.c b/hw/vfio/common.c
index ace9562a9ba1..6b5d8c0bf694 100644
--- a/hw/vfio/common.c
+++ b/hw/vfio/common.c
@@ -2111,29 +2111,31 @@  static int vfio_connect_container(VFIOGroup *group, AddressSpace *as,
     {
         struct vfio_iommu_type1_info *info;
 
-        /*
-         * FIXME: This assumes that a Type1 IOMMU can map any 64-bit
-         * IOVA whatsoever.  That's not actually true, but the current
-         * kernel interface doesn't tell us what it can map, and the
-         * existing Type1 IOMMUs generally support any IOVA we're
-         * going to actually try in practice.
-         */
         ret = vfio_get_iommu_info(container, &info);
+        if (ret) {
+            error_setg_errno(errp, -ret, "Failed to get VFIO IOMMU info");
+            goto enable_discards_exit;
+        }
 
-        if (ret || !(info->flags & VFIO_IOMMU_INFO_PGSIZES)) {
-            /* Assume 4k IOVA page size */
-            info->iova_pgsizes = 4096;
+        if (info->flags & VFIO_IOMMU_INFO_PGSIZES) {
+            container->pgsizes = info->iova_pgsizes;
+        } else {
+            container->pgsizes = qemu_real_host_page_size();
         }
-        vfio_host_win_add(container, 0, (hwaddr)-1, info->iova_pgsizes);
-        container->pgsizes = info->iova_pgsizes;
 
-        /* The default in the kernel ("dma_entry_limit") is 65535. */
-        container->dma_max_mappings = 65535;
-        if (!ret) {
-            vfio_get_info_dma_avail(info, &container->dma_max_mappings);
-            vfio_get_iommu_info_migration(container, info);
+        if (!vfio_get_info_dma_avail(info, &container->dma_max_mappings)) {
+            container->dma_max_mappings = 65535;
         }
+        vfio_get_iommu_info_migration(container, info);
         g_free(info);
+
+        /*
+         * FIXME: We should parse VFIO_IOMMU_TYPE1_INFO_CAP_IOVA_RANGE
+         * information to get the actual window extent rather than assume
+         * a 64-bit IOVA address space.
+         */
+        vfio_host_win_add(container, 0, (hwaddr)-1, container->pgsizes);
+
         break;
     }
     case VFIO_SPAPR_TCE_v2_IOMMU: