[PATCH v1 01/13] KVM: Add KVM_MEM_USERFAULT memslot flag and bitmap

[James Houghton <jthoughton@google.com>]

Use one of the 14 reserved u64s in struct kvm_userspace_memory_region2
for the user to provide `userfault_bitmap`.

The memslot flag indicates if KVM should be reading from the
`userfault_bitmap` field from the memslot. The user is permitted to
provide a bogus pointer. If the pointer cannot be read from, we will
return -EFAULT (with no other information) back to the user.

Signed-off-by: James Houghton <jthoughton@google.com>
---
 include/linux/kvm_host.h | 14 ++++++++++++++
 include/uapi/linux/kvm.h |  4 +++-
 virt/kvm/Kconfig         |  3 +++
 virt/kvm/kvm_main.c      | 28 ++++++++++++++++++++++++++++
 4 files changed, 48 insertions(+), 1 deletion(-)

diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
index 401439bb21e3..f7a3dfd5e224 100644
--- a/include/linux/kvm_host.h
+++ b/include/linux/kvm_host.h
@@ -590,6 +590,7 @@ struct kvm_memory_slot {
 	unsigned long *dirty_bitmap;
 	struct kvm_arch_memory_slot arch;
 	unsigned long userspace_addr;
+	unsigned long __user *userfault_bitmap;
 	u32 flags;
 	short id;
 	u16 as_id;
@@ -724,6 +725,11 @@ static inline bool kvm_arch_has_readonly_mem(struct kvm *kvm)
 }
 #endif
 
+static inline bool kvm_has_userfault(struct kvm *kvm)
+{
+	return IS_ENABLED(CONFIG_HAVE_KVM_USERFAULT);
+}
+
 struct kvm_memslots {
 	u64 generation;
 	atomic_long_t last_used_slot;
@@ -2553,4 +2559,12 @@ long kvm_arch_vcpu_pre_fault_memory(struct kvm_vcpu *vcpu,
 				    struct kvm_pre_fault_memory *range);
 #endif
 
+int kvm_gfn_userfault(struct kvm *kvm, struct kvm_memory_slot *memslot,
+		      gfn_t gfn);
+
+static inline bool kvm_memslot_userfault(struct kvm_memory_slot *memslot)
+{
+	return memslot->flags & KVM_MEM_USERFAULT;
+}
+
 #endif
diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
index 502ea63b5d2e..94be7e8b46a4 100644
--- a/include/uapi/linux/kvm.h
+++ b/include/uapi/linux/kvm.h
@@ -40,7 +40,8 @@ struct kvm_userspace_memory_region2 {
 	__u64 guest_memfd_offset;
 	__u32 guest_memfd;
 	__u32 pad1;
-	__u64 pad2[14];
+	__u64 userfault_bitmap;
+	__u64 pad2[13];
 };
 
 /*
@@ -51,6 +52,7 @@ struct kvm_userspace_memory_region2 {
 #define KVM_MEM_LOG_DIRTY_PAGES	(1UL << 0)
 #define KVM_MEM_READONLY	(1UL << 1)
 #define KVM_MEM_GUEST_MEMFD	(1UL << 2)
+#define KVM_MEM_USERFAULT	(1UL << 3)
 
 /* for KVM_IRQ_LINE */
 struct kvm_irq_level {
diff --git a/virt/kvm/Kconfig b/virt/kvm/Kconfig
index 54e959e7d68f..9eb1fae238b1 100644
--- a/virt/kvm/Kconfig
+++ b/virt/kvm/Kconfig
@@ -124,3 +124,6 @@ config HAVE_KVM_ARCH_GMEM_PREPARE
 config HAVE_KVM_ARCH_GMEM_INVALIDATE
        bool
        depends on KVM_PRIVATE_MEM
+
+config HAVE_KVM_USERFAULT
+       bool
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index de2c11dae231..23fa3e911c4e 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -1541,6 +1541,9 @@ static int check_memory_region_flags(struct kvm *kvm,
 	    !(mem->flags & KVM_MEM_GUEST_MEMFD))
 		valid_flags |= KVM_MEM_READONLY;
 
+	if (kvm_has_userfault(kvm))
+		valid_flags |= KVM_MEM_USERFAULT;
+
 	if (mem->flags & ~valid_flags)
 		return -EINVAL;
 
@@ -2042,6 +2045,8 @@ int __kvm_set_memory_region(struct kvm *kvm,
 		if (r)
 			goto out;
 	}
+	if (mem->flags & KVM_MEM_USERFAULT)
+		new->userfault_bitmap = (unsigned long *)mem->userfault_bitmap;
 
 	r = kvm_set_memslot(kvm, old, new, change);
 	if (r)
@@ -6426,3 +6431,26 @@ void kvm_exit(void)
 	kvm_irqfd_exit();
 }
 EXPORT_SYMBOL_GPL(kvm_exit);
+
+int kvm_gfn_userfault(struct kvm *kvm, struct kvm_memory_slot *memslot,
+		       gfn_t gfn)
+{
+	unsigned long bitmap_chunk = 0;
+	off_t offset;
+
+	if (!kvm_memslot_userfault(memslot))
+		return 0;
+
+	if (WARN_ON_ONCE(!memslot->userfault_bitmap))
+		return 0;
+
+	offset = gfn - memslot->base_gfn;
+
+	if (copy_from_user(&bitmap_chunk,
+			   memslot->userfault_bitmap + offset / BITS_PER_LONG,
+			   sizeof(bitmap_chunk)))
+		return -EFAULT;
+
+	/* Set in the bitmap means that the gfn is userfault */
+	return !!(bitmap_chunk & (1ul << (offset % BITS_PER_LONG)));
+}
-- 
2.47.0.338.g60cca15819-goog