@@ -758,6 +758,178 @@ int kvm_copy_ptes(void *ptr, unsigned long size)
}
EXPORT_SYMBOL(kvm_copy_ptes);
+static void kvm_clear_pte_range(pmd_t *pmd, unsigned long addr,
+ unsigned long end)
+{
+ pte_t *pte;
+
+ pte = kvm_pte_offset(pmd, addr);
+ if (IS_ERR(pte)) {
+ pr_debug("PTE not found, skip clearing\n");
+ return;
+ }
+ do {
+ pr_debug("PTE: %lx/%lx clear[%lx]\n", addr, end, (long)pte);
+ pte_clear(NULL, addr, pte);
+ } while (pte++, addr += PAGE_SIZE, addr < end);
+}
+
+static void kvm_clear_pmd_range(pud_t *pud, unsigned long addr,
+ unsigned long end, enum page_table_level level)
+{
+ pmd_t *pmd;
+ unsigned long next;
+
+ pmd = kvm_pmd_offset(pud, addr);
+ if (IS_ERR(pmd)) {
+ pr_debug("PMD not found, skip clearing\n");
+ return;
+ }
+ do {
+ next = pmd_addr_end(addr, end);
+ if (pmd_none(*pmd))
+ continue;
+ BUG_ON(!pmd_present(*pmd));
+ if (level == PGT_LEVEL_PMD || pmd_trans_huge(*pmd) ||
+ pmd_devmap(*pmd)) {
+ pr_debug("PMD: %lx/%lx clear[%lx]\n",
+ addr, end, (long)pmd);
+ pmd_clear(pmd);
+ continue;
+ }
+ kvm_clear_pte_range(pmd, addr, next);
+ } while (pmd++, addr = next, addr < end);
+}
+
+static void kvm_clear_pud_range(p4d_t *p4d, unsigned long addr,
+ unsigned long end, enum page_table_level level)
+{
+ pud_t *pud;
+ unsigned long next;
+
+ pud = kvm_pud_offset(p4d, addr);
+ if (IS_ERR(pud)) {
+ pr_debug("PUD not found, skip clearing\n");
+ return;
+ }
+ do {
+ next = pud_addr_end(addr, end);
+ if (pud_none(*pud))
+ continue;
+ if (level == PGT_LEVEL_PUD || pud_trans_huge(*pud) ||
+ pud_devmap(*pud)) {
+ pr_debug("PUD: %lx/%lx clear[%lx]\n",
+ addr, end, (long)pud);
+ pud_clear(pud);
+ continue;
+ }
+ kvm_clear_pmd_range(pud, addr, next, level);
+ } while (pud++, addr = next, addr < end);
+}
+
+static void kvm_clear_p4d_range(pgd_t *pgd, unsigned long addr,
+ unsigned long end, enum page_table_level level)
+{
+ p4d_t *p4d;
+ unsigned long next;
+
+ p4d = kvm_p4d_offset(pgd, addr);
+ if (IS_ERR(p4d)) {
+ pr_debug("P4D not found, skip clearing\n");
+ return;
+ }
+
+ do {
+ next = p4d_addr_end(addr, end);
+ if (p4d_none(*p4d))
+ continue;
+ if (level == PGT_LEVEL_P4D) {
+ pr_debug("P4D: %lx/%lx clear[%lx]\n",
+ addr, end, (long)p4d);
+ p4d_clear(p4d);
+ continue;
+ }
+ kvm_clear_pud_range(p4d, addr, next, level);
+ } while (p4d++, addr = next, addr < end);
+}
+
+static void kvm_clear_pgd_range(struct mm_struct *mm, unsigned long addr,
+ unsigned long end, enum page_table_level level)
+{
+ pgd_t *pgd;
+ unsigned long next;
+
+ pgd = pgd_offset(mm, addr);
+ do {
+ next = pgd_addr_end(addr, end);
+ if (pgd_none(*pgd))
+ continue;
+ if (level == PGT_LEVEL_PGD) {
+ pr_debug("PGD: %lx/%lx clear[%lx]\n",
+ addr, end, (long)pgd);
+ pgd_clear(pgd);
+ continue;
+ }
+ kvm_clear_p4d_range(pgd, addr, next, level);
+ } while (pgd++, addr = next, addr < end);
+}
+
+/*
+ * Clear page table entries in the KVM page table. The level parameter
+ * specifies the page table level (PGD, P4D, PUD PMD, PTE) at which the
+ * clear should be done.
+ *
+ * WARNING: The KVM page table can have direct references to the kernel
+ * page table, at different levels (PGD, P4D, PUD, PMD). When clearing
+ * such references, if the level is incorrect (for example, clear at the
+ * PTE level while the mapping was done at PMD level), then the clearing
+ * will occur in the kernel page table and the system will likely crash
+ * on an unhandled page fault.
+ */
+static void kvm_clear_mapping(void *ptr, size_t size,
+ enum page_table_level level)
+{
+ unsigned long start = (unsigned long)ptr;
+ unsigned long end = start + ((unsigned long)size);
+
+ pr_debug("CLEAR %px, %lx [%lx,%lx], level=%d\n",
+ ptr, size, start, end, level);
+ kvm_clear_pgd_range(&kvm_mm, start, end, level);
+}
+
+/*
+ * Clear a range mapping in the KVM page table.
+ */
+void kvm_clear_range_mapping(void *ptr)
+{
+ struct kvm_range_mapping *range_mapping;
+ bool subset;
+
+ mutex_lock(&kvm_range_mapping_lock);
+
+ range_mapping = kvm_get_range_mapping_locked(ptr, &subset);
+ if (!range_mapping) {
+ mutex_unlock(&kvm_range_mapping_lock);
+ pr_debug("CLEAR %px - range not found\n", ptr);
+ return;
+ }
+ if (subset) {
+ mutex_unlock(&kvm_range_mapping_lock);
+ pr_debug("CLEAR %px - ignored, subset of %px/%lx/%d\n",
+ ptr, range_mapping->ptr, range_mapping->size,
+ range_mapping->level);
+ return;
+ }
+
+ kvm_clear_mapping(range_mapping->ptr, range_mapping->size,
+ range_mapping->level);
+ list_del(&range_mapping->list);
+ mutex_unlock(&kvm_range_mapping_lock);
+
+ kfree(range_mapping);
+}
+EXPORT_SYMBOL(kvm_clear_range_mapping);
+
static int kvm_isolation_init_mm(void)
{
@@ -17,5 +17,6 @@ static inline bool kvm_isolation(void)
extern void kvm_isolation_exit(void);
extern void kvm_may_access_sensitive_data(struct kvm_vcpu *vcpu);
extern int kvm_copy_ptes(void *ptr, unsigned long size);
+extern void kvm_clear_range_mapping(void *ptr);
#endif
These functions will be used to unmapped memory from the KVM address space. When clearing mapping in the KVM page table, check that the clearing effectively happens in the KVM page table and there is no crossing of the KVM page table boundary (with references to the kernel page table), so that the kernel page table isn't mistakenly modified. Information (address, size, page table level) about address ranges mapped to the KVM page table is tracked, so mapping clearing is done with just specified the start address of the range. Signed-off-by: Alexandre Chartre <alexandre.chartre@oracle.com> --- arch/x86/kvm/isolation.c | 172 ++++++++++++++++++++++++++++++++++++++++++++++ arch/x86/kvm/isolation.h | 1 + 2 files changed, 173 insertions(+), 0 deletions(-)