xen/gntdev: mark userspace PTEs as special on x86 PV guests

In an x86 PV guest, get_user_pages_fast() on a userspace address range
containing foreign mappings does not work correctly because the M2P
lookup of the MFN from a userspace PTE may return the wrong page.

Force get_user_pages_fast() to fail on such addresses by marking the PTEs
as special.

If Xen has XENFEAT_gnttab_map_avail_bits (available since at least
4.0), we can do so efficiently in the grant map hypercall.  Otherwise,
it needs to be done afterwards.  This is both inefficient and racy
(the mapping is visible to the task before we fixup the PTEs), but
will be fine for well-behaved applications that do not use the mapping
until after the mmap() system call returns.

Guests with XENFEAT_auto_translated_physmap (ARM and x86 HVM or PVH)
do not need this since get_user_pages() has always worked correctly
for them.
Signed-off-by: NDavid Vrabel <david.vrabel@citrix.com>
Reviewed-by: NStefano Stabellini <stefano.stabellini@eu.citrix.com>

drivers/xen/gntdev.c

@@ -244,6 +244,14 @@ static int find_grant_ptes(pte_t *pte, pgtable_t token,
 	BUG_ON(pgnr >= map->count);
 	pte_maddr = arbitrary_virt_to_machine(pte).maddr;
 
+	/*
+	 * Set the PTE as special to force get_user_pages_fast() fall
+	 * back to the slow path.  If this is not supported as part of
+	 * the grant map, it will be done afterwards.
+	 */
+	if (xen_feature(XENFEAT_gnttab_map_avail_bits))
+		flags |= (1 << _GNTMAP_guest_avail0);
+
 	gnttab_set_map_op(&map->map_ops[pgnr], pte_maddr, flags,
 			  map->grants[pgnr].ref,
 			  map->grants[pgnr].domid);
@@ -252,6 +260,15 @@ static int find_grant_ptes(pte_t *pte, pgtable_t token,
 	return 0;
 }
 
+#ifdef CONFIG_X86
+static int set_grant_ptes_as_special(pte_t *pte, pgtable_t token,
+				     unsigned long addr, void *data)
+{
+	set_pte_at(current->mm, addr, pte, pte_mkspecial(*pte));
+	return 0;
+}
+#endif
+
 static int map_grant_pages(struct grant_map *map)
 {
 	int i, err = 0;
@@ -840,6 +857,23 @@ static int gntdev_mmap(struct file *flip, struct vm_area_struct *vma)
 			if (err)
 				goto out_put_map;
 		}
+	} else {
+#ifdef CONFIG_X86
+		/*
+		 * If the PTEs were not made special by the grant map
+		 * hypercall, do so here.
+		 *
+		 * This is racy since the mapping is already visible
+		 * to userspace but userspace should be well-behaved
+		 * enough to not touch it until the mmap() call
+		 * returns.
+		 */
+		if (!xen_feature(XENFEAT_gnttab_map_avail_bits)) {
+			apply_to_page_range(vma->vm_mm, vma->vm_start,
+					    vma->vm_end - vma->vm_start,
+					    set_grant_ptes_as_special, NULL);
+		}
+#endif
 	}
 
 	return 0;

include/xen/interface/features.h

@@ -41,6 +41,12 @@
 /* x86: Does this Xen host support the MMU_PT_UPDATE_PRESERVE_AD hypercall? */
 #define XENFEAT_mmu_pt_update_preserve_ad  5
 
+/*
+ * If set, GNTTABOP_map_grant_ref honors flags to be placed into guest kernel
+ * available pte bits.
+ */
+#define XENFEAT_gnttab_map_avail_bits      7
+
 /* x86: Does this Xen host support the HVM callback vector type? */
 #define XENFEAT_hvm_callback_vector        8
 

include/xen/interface/grant_table.h

@@ -525,6 +525,13 @@ DEFINE_GUEST_HANDLE_STRUCT(gnttab_cache_flush);
 #define _GNTMAP_contains_pte    (4)
 #define GNTMAP_contains_pte     (1<<_GNTMAP_contains_pte)
 
+/*
+ * Bits to be placed in guest kernel available PTE bits (architecture
+ * dependent; only supported when XENFEAT_gnttab_map_avail_bits is set).
+ */
+#define _GNTMAP_guest_avail0    (16)
+#define GNTMAP_guest_avail_mask ((uint32_t)~0 << _GNTMAP_guest_avail0)
+
 /*
  * Values for error status returns. All errors are -ve.
  */