[1/2] kvm: fix potentially corrupt mmio cache

Commit Message

David Matlack Aug. 18, 2014, 10:46 p.m. UTC

vcpu exits and memslot mutations can run concurrently as long as the
vcpu does not aquire the slots mutex. Thus it is theoretically possible
for memslots to change underneath a vcpu that is handling an exit.

If we increment the memslot generation number again after
synchronize_srcu_expedited(), vcpus can safely cache memslot generation
without maintaining a single rcu_dereference through an entire vm exit.
And much of the x86/kvm code does not maintain a single rcu_dereference
of the current memslots during each exit.

We can prevent the following case:

   vcpu (CPU 0)                             | thread (CPU 1)
--------------------------------------------+--------------------------
1  vm exit                                  |
2  decide to cache something based on       |
     old memslots                           |
3                                           | change memslots
4                                           | increment generation
5  tag cache with new memslot generation    |
...                                         |
   <action based on cache occurs even       |
    though the caching decision was based   |
    on the old memslots>                    |
...                                         |
   <action *continues* to occur until next  |
    memslot generation change, which may    |
    be never>                               |

By incrementing the generation again after synchronizing kvm->srcu
readers, we guarantee the generation cached in (5) will very soon
become invalid.

Cc: [e-mail]
Cc: Xiao Guangrong <[e-mail]>
Signed-off-by: David Matlack <[e-mail]>
---
 virt/kvm/kvm_main.c | 18 +++++++++++-------
 1 file changed, 11 insertions(+), 7 deletions(-)

Patch

diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index 4b6c01b..86d3697 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -95,8 +95,6 @@  static int hardware_enable_all(void);
 static void hardware_disable_all(void);
 
 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
-static void update_memslots(struct kvm_memslots *slots,
-			    struct kvm_memory_slot *new, u64 last_generation);
 
 static void kvm_release_pfn_dirty(pfn_t pfn);
 static void mark_page_dirty_in_slot(struct kvm *kvm,
@@ -685,8 +683,7 @@  static void sort_memslots(struct kvm_memslots *slots)
 }
 
 static void update_memslots(struct kvm_memslots *slots,
-			    struct kvm_memory_slot *new,
-			    u64 last_generation)
+			    struct kvm_memory_slot *new)
 {
 	if (new) {
 		int id = new->id;
@@ -697,8 +694,6 @@  static void update_memslots(struct kvm_memslots *slots,
 		if (new->npages != npages)
 			sort_memslots(slots);
 	}
-
-	slots->generation = last_generation + 1;
 }
 
 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
@@ -720,10 +715,19 @@  static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 {
 	struct kvm_memslots *old_memslots = kvm->memslots;
 
-	update_memslots(slots, new, kvm->memslots->generation);
+	slots->generation = old_memslots->generation + 1;
+
+	update_memslots(slots, new);
 	rcu_assign_pointer(kvm->memslots, slots);
 	synchronize_srcu_expedited(&kvm->srcu);
 
+	/*
+	 * Increment the new memslot generation a second time. This prevents
+	 * vm exits that race with memslot updates from caching a memslot
+	 * generation that will (potentially) be valid forever.
+	 */
+	slots->generation++;
+
 	kvm_arch_memslots_updated(kvm);
 
 	return old_memslots;