KVM: arm64: Retry fault if vma_lookup() results become invalid

Read mmu_invalidate_seq before dropping the mmap_lock so that KVM can
detect if the results of vma_lookup() (e.g. vma_shift) become stale
before it acquires kvm->mmu_lock. This fixes a theoretical bug where a
VMA could be changed by userspace after vma_lookup() and before KVM
reads the mmu_invalidate_seq, causing KVM to install page table entries
based on a (possibly) no-longer-valid vma_shift.

Re-order the MMU cache top-up to earlier in user_mem_abort() so that it
is not done after KVM has read mmu_invalidate_seq (i.e. so as to avoid
inducing spurious fault retries).

This bug has existed since KVM/ARM's inception. It's unlikely that any
sane userspace currently modifies VMAs in such a way as to trigger this
race. And even with directed testing I was unable to reproduce it. But a
sufficiently motivated host userspace might be able to exploit this
race.

Fixes: 94f8e641 ("KVM: ARM: Handle guest faults in KVM")
Cc: stable@vger.kernel.org
Reported-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: David Matlack <dmatlack@google.com>
Reviewed-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20230313235454.2964067-1-dmatlack@google.comSigned-off-by: Oliver Upton <oliver.upton@linux.dev>

arch/arm64/kvm/mmu.c

@@ -1217,6 +1217,20 @@ static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,
 		return -EFAULT;
 	}
 
+	/*
+	 * Permission faults just need to update the existing leaf entry,
+	 * and so normally don't require allocations from the memcache. The
+	 * only exception to this is when dirty logging is enabled at runtime
+	 * and a write fault needs to collapse a block entry into a table.
+	 */
+	if (fault_status != ESR_ELx_FSC_PERM ||
+	    (logging_active && write_fault)) {
+		ret = kvm_mmu_topup_memory_cache(memcache,
+						 kvm_mmu_cache_min_pages(kvm));
+		if (ret)
+			return ret;
+	}
+
 	/*
 	 * Let's check if we will get back a huge page backed by hugetlbfs, or
 	 * get block mapping for device MMIO region.
@@ -1269,37 +1283,17 @@ static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,
 		fault_ipa &= ~(vma_pagesize - 1);
 
 	gfn = fault_ipa >> PAGE_SHIFT;
-	mmap_read_unlock(current->mm);
-
-	/*
-	 * Permission faults just need to update the existing leaf entry,
-	 * and so normally don't require allocations from the memcache. The
-	 * only exception to this is when dirty logging is enabled at runtime
-	 * and a write fault needs to collapse a block entry into a table.
-	 */
-	if (fault_status != ESR_ELx_FSC_PERM ||
-	    (logging_active && write_fault)) {
-		ret = kvm_mmu_topup_memory_cache(memcache,
-						 kvm_mmu_cache_min_pages(kvm));
-		if (ret)
-			return ret;
-	}
 
-	mmu_seq = vcpu->kvm->mmu_invalidate_seq;
 	/*
-	 * Ensure the read of mmu_invalidate_seq happens before we call
-	 * gfn_to_pfn_prot (which calls get_user_pages), so that we don't risk
-	 * the page we just got a reference to gets unmapped before we have a
-	 * chance to grab the mmu_lock, which ensure that if the page gets
-	 * unmapped afterwards, the call to kvm_unmap_gfn will take it away
-	 * from us again properly. This smp_rmb() interacts with the smp_wmb()
-	 * in kvm_mmu_notifier_invalidate_<page|range_end>.
+	 * Read mmu_invalidate_seq so that KVM can detect if the results of
+	 * vma_lookup() or __gfn_to_pfn_memslot() become stale prior to
+	 * acquiring kvm->mmu_lock.
 	 *
-	 * Besides, __gfn_to_pfn_memslot() instead of gfn_to_pfn_prot() is
-	 * used to avoid unnecessary overhead introduced to locate the memory
-	 * slot because it's always fixed even @gfn is adjusted for huge pages.
+	 * Rely on mmap_read_unlock() for an implicit smp_rmb(), which pairs
+	 * with the smp_wmb() in kvm_mmu_invalidate_end().
 	 */
-	smp_rmb();
+	mmu_seq = vcpu->kvm->mmu_invalidate_seq;
+	mmap_read_unlock(current->mm);
 
 	pfn = __gfn_to_pfn_memslot(memslot, gfn, false, false, NULL,
 				   write_fault, &writable, NULL);