kvm: introduce manual dirty log reprotect

There are two problems with KVM_GET_DIRTY_LOG. First, and less important,
it can take kvm->mmu_lock for an extended period of time. Second, its user
can actually see many false positives in some cases. The latter is due
to a benign race like this:

1. KVM_GET_DIRTY_LOG returns a set of dirty pages and write protects
them.
2. The guest modifies the pages, causing them to be marked ditry.
3. Userspace actually copies the pages.
4. KVM_GET_DIRTY_LOG returns those pages as dirty again, even though
they were not written to since (3).

This is especially a problem for large guests, where the time between
(1) and (3) can be substantial. This patch introduces a new
capability which, when enabled, makes KVM_GET_DIRTY_LOG not
write-protect the pages it returns. Instead, userspace has to
explicitly clear the dirty log bits just before using the content
of the page. The new KVM_CLEAR_DIRTY_LOG ioctl can also operate on a
64-page granularity rather than requiring to sync a full memslot;
this way, the mmu_lock is taken for small amounts of time, and
only a small amount of time will pass between write protection
of pages and the sending of their content.

Signed-off-by: Paolo Bonzini <[e-mail]>

Documentation/virtual/kvm/api.txt

···
 They must be less than the value that KVM_CHECK_EXTENSION returns for
 the KVM_CAP_MULTI_ADDRESS_SPACE capability.
 
+The bits in the dirty bitmap are cleared before the ioctl returns, unless
+KVM_CAP_MANUAL_DIRTY_LOG_PROTECT is enabled.  For more information,
+see the description of the capability.
 
 4.9 KVM_SET_MEMORY_ALIAS
 
···
 between coalesced mmio and pio except that coalesced pio records accesses
 to I/O ports.
 
+4.117 KVM_CLEAR_DIRTY_LOG (vm ioctl)
+
+Capability: KVM_CAP_MANUAL_DIRTY_LOG_PROTECT
+Architectures: x86
+Type: vm ioctl
+Parameters: struct kvm_dirty_log (in)
+Returns: 0 on success, -1 on error
+
+/* for KVM_CLEAR_DIRTY_LOG */
+struct kvm_clear_dirty_log {
+	__u32 slot;
+	__u32 num_pages;
+	__u64 first_page;
+	union {
+		void __user *dirty_bitmap; /* one bit per page */
+		__u64 padding;
+	};
+};
+
+The ioctl clears the dirty status of pages in a memory slot, according to
+the bitmap that is passed in struct kvm_clear_dirty_log's dirty_bitmap
+field.  Bit 0 of the bitmap corresponds to page "first_page" in the
+memory slot, and num_pages is the size in bits of the input bitmap.
+Both first_page and num_pages must be a multiple of 64.  For each bit
+that is set in the input bitmap, the corresponding page is marked "clean"
+in KVM's dirty bitmap, and dirty tracking is re-enabled for that page
+(for example via write-protection, or by clearing the dirty bit in
+a page table entry).
+
+If KVM_CAP_MULTI_ADDRESS_SPACE is available, bits 16-31 specifies
+the address space for which you want to return the dirty bitmap.
+They must be less than the value that KVM_CHECK_EXTENSION returns for
+the KVM_CAP_MULTI_ADDRESS_SPACE capability.
+
+This ioctl is mostly useful when KVM_CAP_MANUAL_DIRTY_LOG_PROTECT
+is enabled; for more information, see the description of the capability.
+However, it can always be used as long as KVM_CHECK_EXTENSION confirms
+that KVM_CAP_MANUAL_DIRTY_LOG_PROTECT is present.
+
+
 5. The kvm_run structure
 ------------------------
 
···
 
 * For the new DR6 bits, note that bit 16 is set iff the #DB exception
   will clear DR6.RTM.
+
+7.18 KVM_CAP_MANUAL_DIRTY_LOG_PROTECT
+
+Architectures: all
+Parameters: args[0] whether feature should be enabled or not
+
+With this capability enabled, KVM_GET_DIRTY_LOG will not automatically
+clear and write-protect all pages that are returned as dirty.
+Rather, userspace will have to do this operation separately using
+KVM_CLEAR_DIRTY_LOG.
+
+At the cost of a slightly more complicated operation, this provides better
+scalability and responsiveness for two reasons.  First,
+KVM_CLEAR_DIRTY_LOG ioctl can operate on a 64-page granularity rather
+than requiring to sync a full memslot; this ensures that KVM does not
+take spinlocks for an extended period of time.  Second, in some cases a
+large amount of time can pass between a call to KVM_GET_DIRTY_LOG and
+userspace actually using the data in the page.  Pages can be modified
+during this time, which is inefficint for both the guest and userspace:
+the guest will incur a higher penalty due to write protection faults,
+while userspace can see false reports of dirty pages.  Manual reprotection
+helps reducing this time, improving guest performance and reducing the
+number of dirty log false positives.
+
 
 8. Other capabilities.
 ----------------------

arch/mips/kvm/mips.c

···
 	return r;
 }
 
+int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
+{
+	struct kvm_memslots *slots;
+	struct kvm_memory_slot *memslot;
+	bool flush = false;
+	int r;
+
+	mutex_lock(&kvm->slots_lock);
+
+	r = kvm_clear_dirty_log_protect(kvm, log, &flush);
+
+	if (flush) {
+		slots = kvm_memslots(kvm);
+		memslot = id_to_memslot(slots, log->slot);
+
+		/* Let implementation handle TLB/GVA invalidation */
+		kvm_mips_callbacks->flush_shadow_memslot(kvm, memslot);
+	}
+
+	mutex_unlock(&kvm->slots_lock);
+	return r;
+}
+
 long kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
 {
 	long r;

arch/x86/kvm/x86.c

···
 	return r;
 }
 
+int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
+{
+	bool flush = false;
+	int r;
+
+	mutex_lock(&kvm->slots_lock);
+
+	/*
+	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
+	 */
+	if (kvm_x86_ops->flush_log_dirty)
+		kvm_x86_ops->flush_log_dirty(kvm);
+
+	r = kvm_clear_dirty_log_protect(kvm, log, &flush);
+
+	/*
+	 * All the TLBs can be flushed out of mmu lock, see the comments in
+	 * kvm_mmu_slot_remove_write_access().
+	 */
+	lockdep_assert_held(&kvm->slots_lock);
+	if (flush)
+		kvm_flush_remote_tlbs(kvm);
+
+	mutex_unlock(&kvm->slots_lock);
+	return r;
+}
+
 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
 			bool line_status)
 {

include/linux/kvm_host.h

···
 #endif
 	long tlbs_dirty;
 	struct list_head devices;
+	bool manual_dirty_log_protect;
 	struct dentry *debugfs_dentry;
 	struct kvm_stat_data **debugfs_stat_data;
 	struct srcu_struct srcu;
···
 
 int kvm_get_dirty_log_protect(struct kvm *kvm,
 			      struct kvm_dirty_log *log, bool *flush);
+int kvm_clear_dirty_log_protect(struct kvm *kvm,
+				struct kvm_clear_dirty_log *log, bool *flush);
 
 void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm,
 					struct kvm_memory_slot *slot,
···
 
 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 				struct kvm_dirty_log *log);
+int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
+				  struct kvm_clear_dirty_log *log);
 
 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 			bool line_status);

include/uapi/linux/kvm.h

···
 	};
 };
 
+/* for KVM_CLEAR_DIRTY_LOG */
+struct kvm_clear_dirty_log {
+	__u32 slot;
+	__u32 num_pages;
+	__u64 first_page;
+	union {
+		void __user *dirty_bitmap; /* one bit per page */
+		__u64 padding2;
+	};
+};
+
 /* for KVM_SET_SIGNAL_MASK */
 struct kvm_signal_mask {
 	__u32 len;
···
 #define KVM_CAP_HYPERV_ENLIGHTENED_VMCS 163
 #define KVM_CAP_EXCEPTION_PAYLOAD 164
 #define KVM_CAP_ARM_VM_IPA_SIZE 165
+#define KVM_CAP_MANUAL_DIRTY_LOG_PROTECT 166
 
 #ifdef KVM_CAP_IRQ_ROUTING
 
···
 /* Available with KVM_CAP_NESTED_STATE */
 #define KVM_GET_NESTED_STATE         _IOWR(KVMIO, 0xbe, struct kvm_nested_state)
 #define KVM_SET_NESTED_STATE         _IOW(KVMIO,  0xbf, struct kvm_nested_state)
+
+/* Available with KVM_CAP_MANUAL_DIRTY_LOG_PROTECT */
+#define KVM_CLEAR_DIRTY_LOG          _IOWR(KVMIO, 0xc0, struct kvm_clear_dirty_log)
 
 /* Secure Encrypted Virtualization command */
 enum sev_cmd_id {

tools/testing/selftests/kvm/Makefile

···
 TEST_GEN_PROGS_x86_64 += x86_64/state_test
 TEST_GEN_PROGS_x86_64 += x86_64/evmcs_test
 TEST_GEN_PROGS_x86_64 += dirty_log_test
+TEST_GEN_PROGS_x86_64 += clear_dirty_log_test
 
 TEST_GEN_PROGS_aarch64 += dirty_log_test
+TEST_GEN_PROGS_aarch64 += clear_dirty_log_test
 
 TEST_GEN_PROGS += $(TEST_GEN_PROGS_$(UNAME_M))
 LIBKVM += $(LIBKVM_$(UNAME_M))

tools/testing/selftests/kvm/clear_dirty_log_test.c

···
+#define USE_CLEAR_DIRTY_LOG
+#include "dirty_log_test.c"

tools/testing/selftests/kvm/dirty_log_test.c

···
 
 	vm = create_vm(mode, VCPU_ID, guest_num_pages, guest_code);
 
+#ifdef USE_CLEAR_DIRTY_LOG
+	struct kvm_enable_cap cap = {};
+
+	cap.cap = KVM_CAP_MANUAL_DIRTY_LOG_PROTECT;
+	cap.args[0] = 1;
+	vm_enable_cap(vm, &cap);
+#endif
+
 	/* Add an extra memory slot for testing dirty logging */
 	vm_userspace_mem_region_add(vm, VM_MEM_SRC_ANONYMOUS,
 				    guest_test_mem,
···
 		/* Give the vcpu thread some time to dirty some pages */
 		usleep(interval * 1000);
 		kvm_vm_get_dirty_log(vm, TEST_MEM_SLOT_INDEX, bmap);
+#ifdef USE_CLEAR_DIRTY_LOG
+		kvm_vm_clear_dirty_log(vm, TEST_MEM_SLOT_INDEX, bmap, 0,
+				       DIV_ROUND_UP(host_num_pages, 64) * 64);
+#endif
 		vm_dirty_log_verify(bmap);
 		iteration++;
 		sync_global_to_guest(vm, iteration);
···
 	bool top_offset = false;
 	unsigned int mode;
 	int opt, i;
+
+#ifdef USE_CLEAR_DIRTY_LOG
+	if (!kvm_check_cap(KVM_CAP_MANUAL_DIRTY_LOG_PROTECT)) {
+		fprintf(stderr, "KVM_CLEAR_DIRTY_LOG not available, skipping tests\n");
+		exit(KSFT_SKIP);
+	}
+#endif
 
 	while ((opt = getopt(argc, argv, "hi:I:o:tm:")) != -1) {
 		switch (opt) {

tools/testing/selftests/kvm/include/kvm_util.h

···
 void kvm_vm_restart(struct kvm_vm *vmp, int perm);
 void kvm_vm_release(struct kvm_vm *vmp);
 void kvm_vm_get_dirty_log(struct kvm_vm *vm, int slot, void *log);
+void kvm_vm_clear_dirty_log(struct kvm_vm *vm, int slot, void *log,
+			    uint64_t first_page, uint32_t num_pages);
 
 int kvm_memcmp_hva_gva(void *hva, struct kvm_vm *vm, const vm_vaddr_t gva,
 		       size_t len);

tools/testing/selftests/kvm/lib/kvm_util.c

···
 		    strerror(-ret));
 }
 
+void kvm_vm_clear_dirty_log(struct kvm_vm *vm, int slot, void *log,
+			    uint64_t first_page, uint32_t num_pages)
+{
+	struct kvm_clear_dirty_log args = { .dirty_bitmap = log, .slot = slot,
+		                            .first_page = first_page,
+	                                    .num_pages = num_pages };
+	int ret;
+
+	ret = ioctl(vm->fd, KVM_CLEAR_DIRTY_LOG, &args);
+	TEST_ASSERT(ret == 0, "%s: KVM_CLEAR_DIRTY_LOG failed: %s",
+		    strerror(-ret));
+}
+
 /*
  * Userspace Memory Region Find
  *

virt/kvm/arm/arm.c

···
 	return r;
 }
 
+int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
+{
+	bool flush = false;
+	int r;
+
+	mutex_lock(&kvm->slots_lock);
+
+	r = kvm_clear_dirty_log_protect(kvm, log, &flush);
+
+	if (flush)
+		kvm_flush_remote_tlbs(kvm);
+
+	mutex_unlock(&kvm->slots_lock);
+	return r;
+}
+
 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
 					struct kvm_arm_device_addr *dev_addr)
 {

virt/kvm/kvm_main.c

···
 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
 /**
  * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
- *	are dirty write protect them for next write.
+ *	and reenable dirty page tracking for the corresponding pages.
  * @kvm:	pointer to kvm instance
  * @log:	slot id and address to which we copy the log
  * @is_dirty:	flag set if any page is dirty
···
 		return -ENOENT;
 
 	n = kvm_dirty_bitmap_bytes(memslot);
-
-	dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
-	memset(dirty_bitmap_buffer, 0, n);
-
-	spin_lock(&kvm->mmu_lock);
 	*flush = false;
-	for (i = 0; i < n / sizeof(long); i++) {
-		unsigned long mask;
-		gfn_t offset;
+	if (kvm->manual_dirty_log_protect) {
+		/*
+		 * Unlike kvm_get_dirty_log, we always return false in *flush,
+		 * because no flush is needed until KVM_CLEAR_DIRTY_LOG.  There
+		 * is some code duplication between this function and
+		 * kvm_get_dirty_log, but hopefully all architecture
+		 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
+		 * can be eliminated.
+		 */
+		dirty_bitmap_buffer = dirty_bitmap;
+	} else {
+		dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
+		memset(dirty_bitmap_buffer, 0, n);
 
-		if (!dirty_bitmap[i])
-			continue;
+		spin_lock(&kvm->mmu_lock);
+		for (i = 0; i < n / sizeof(long); i++) {
+			unsigned long mask;
+			gfn_t offset;
 
-		*flush = true;
+			if (!dirty_bitmap[i])
+				continue;
 
-		mask = xchg(&dirty_bitmap[i], 0);
-		dirty_bitmap_buffer[i] = mask;
+			*flush = true;
+			mask = xchg(&dirty_bitmap[i], 0);
+			dirty_bitmap_buffer[i] = mask;
 
-		if (mask) {
-			offset = i * BITS_PER_LONG;
-			kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
-								offset, mask);
+			if (mask) {
+				offset = i * BITS_PER_LONG;
+				kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
+									offset, mask);
+			}
 		}
+		spin_unlock(&kvm->mmu_lock);
 	}
 
-	spin_unlock(&kvm->mmu_lock);
 	if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
 		return -EFAULT;
 	return 0;
 }
 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
+
+/**
+ * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
+ *	and reenable dirty page tracking for the corresponding pages.
+ * @kvm:	pointer to kvm instance
+ * @log:	slot id and address from which to fetch the bitmap of dirty pages
+ */
+int kvm_clear_dirty_log_protect(struct kvm *kvm,
+				struct kvm_clear_dirty_log *log, bool *flush)
+{
+	struct kvm_memslots *slots;
+	struct kvm_memory_slot *memslot;
+	int as_id, id, n;
+	gfn_t offset;
+	unsigned long i;
+	unsigned long *dirty_bitmap;
+	unsigned long *dirty_bitmap_buffer;
+
+	as_id = log->slot >> 16;
+	id = (u16)log->slot;
+	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
+		return -EINVAL;
+
+	if ((log->first_page & 63) || (log->num_pages & 63))
+		return -EINVAL;
+
+	slots = __kvm_memslots(kvm, as_id);
+	memslot = id_to_memslot(slots, id);
+
+	dirty_bitmap = memslot->dirty_bitmap;
+	if (!dirty_bitmap)
+		return -ENOENT;
+
+	n = kvm_dirty_bitmap_bytes(memslot);
+	*flush = false;
+	dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
+	if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
+		return -EFAULT;
+
+	spin_lock(&kvm->mmu_lock);
+	for (offset = log->first_page,
+	     i = offset / BITS_PER_LONG, n = log->num_pages / BITS_PER_LONG; n--;
+	     i++, offset += BITS_PER_LONG) {
+		unsigned long mask = *dirty_bitmap_buffer++;
+		atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
+		if (!mask)
+			continue;
+
+		mask &= atomic_long_fetch_andnot(mask, p);
+
+		/*
+		 * mask contains the bits that really have been cleared.  This
+		 * never includes any bits beyond the length of the memslot (if
+		 * the length is not aligned to 64 pages), therefore it is not
+		 * a problem if userspace sets them in log->dirty_bitmap.
+		*/
+		if (mask) {
+			*flush = true;
+			kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
+								offset, mask);
+		}
+	}
+	spin_unlock(&kvm->mmu_lock);
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(kvm_clear_dirty_log_protect);
 #endif
 
 bool kvm_largepages_enabled(void)
···
 	case KVM_CAP_IOEVENTFD_ANY_LENGTH:
 	case KVM_CAP_CHECK_EXTENSION_VM:
 	case KVM_CAP_ENABLE_CAP_VM:
+#ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
+	case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT:
+#endif
 		return 1;
 #ifdef CONFIG_KVM_MMIO
 	case KVM_CAP_COALESCED_MMIO:
···
 					   struct kvm_enable_cap *cap)
 {
 	switch (cap->cap) {
+#ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
+	case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT:
+		if (cap->flags || (cap->args[0] & ~1))
+			return -EINVAL;
+		kvm->manual_dirty_log_protect = cap->args[0];
+		return 0;
+#endif
 	default:
 		return kvm_vm_ioctl_enable_cap(kvm, cap);
 	}
···
 		r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
 		break;
 	}
+#ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
+	case KVM_CLEAR_DIRTY_LOG: {
+		struct kvm_clear_dirty_log log;
+
+		r = -EFAULT;
+		if (copy_from_user(&log, argp, sizeof(log)))
+			goto out;
+		r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
+		break;
+	}
+#endif
 #ifdef CONFIG_KVM_MMIO
 	case KVM_REGISTER_COALESCED_MMIO: {
 		struct kvm_coalesced_mmio_zone zone;