[patch 0/4] set_memory_region locking fixes / vcpu->arch.cr3 + removal of memslots

[patch 0/4] set_memory_region locking fixes / vcpu->arch.cr3 + removal of memslots

[mtosatti]

[patch 1/4] KVM: MMU: protect kvm_mmu_change_mmu_pages with mmu_lock

[mtosatti]

[-- Attachment #1: set-mem-lock --]
[-- Type: text/plain, Size: 2139 bytes --]

kvm_handle_hva, called by MMU notifiers, manipulates mmu data only with
the protection of mmu_lock.

Update kvm_mmu_change_mmu_pages callers to take mmu_lock, thus protecting
against kvm_handle_hva.

CC: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

Index: kvm/arch/x86/kvm/mmu.c
===================================================================
--- kvm.orig/arch/x86/kvm/mmu.c
+++ kvm/arch/x86/kvm/mmu.c
@@ -2732,7 +2732,6 @@ void kvm_mmu_slot_remove_write_access(st
 {
 	struct kvm_mmu_page *sp;
 
-	spin_lock(&kvm->mmu_lock);
 	list_for_each_entry(sp, &kvm->arch.active_mmu_pages, link) {
 		int i;
 		u64 *pt;
@@ -2747,7 +2746,6 @@ void kvm_mmu_slot_remove_write_access(st
 				pt[i] &= ~PT_WRITABLE_MASK;
 	}
 	kvm_flush_remote_tlbs(kvm);
-	spin_unlock(&kvm->mmu_lock);
 }
 
 void kvm_mmu_zap_all(struct kvm *kvm)
Index: kvm/arch/x86/kvm/x86.c
===================================================================
--- kvm.orig/arch/x86/kvm/x86.c
+++ kvm/arch/x86/kvm/x86.c
@@ -1647,10 +1647,12 @@ static int kvm_vm_ioctl_set_nr_mmu_pages
 		return -EINVAL;
 
 	down_write(&kvm->slots_lock);
+	spin_lock(&kvm->mmu_lock);
 
 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
 
+	spin_unlock(&kvm->mmu_lock);
 	up_write(&kvm->slots_lock);
 	return 0;
 }
@@ -1826,7 +1828,9 @@ int kvm_vm_ioctl_get_dirty_log(struct kv
 
 	/* If nothing is dirty, don't bother messing with page tables. */
 	if (is_dirty) {
+		spin_lock(&kvm->mmu_lock);
 		kvm_mmu_slot_remove_write_access(kvm, log->slot);
+		spin_unlock(&kvm->mmu_lock);
 		kvm_flush_remote_tlbs(kvm);
 		memslot = &kvm->memslots[log->slot];
 		n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
@@ -4510,12 +4514,14 @@ int kvm_arch_set_memory_region(struct kv
 		}
 	}
 
+	spin_lock(&kvm->mmu_lock);
 	if (!kvm->arch.n_requested_mmu_pages) {
 		unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
 	}
 
 	kvm_mmu_slot_remove_write_access(kvm, mem->slot);
+	spin_unlock(&kvm->mmu_lock);
 	kvm_flush_remote_tlbs(kvm);
 
 	return 0;

[patch 2/4] KVM: take mmu_lock when updating a deleted slot

[mtosatti]

[-- Attachment #1: slot-delete-update-mmu-lock --]
[-- Type: text/plain, Size: 692 bytes --]

kvm_handle_hva relies on mmu_lock protection to safely access 
the memslot structures.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

Index: kvm/virt/kvm/kvm_main.c
===================================================================
--- kvm.orig/virt/kvm/kvm_main.c
+++ kvm/virt/kvm/kvm_main.c
@@ -1199,8 +1199,10 @@ int __kvm_set_memory_region(struct kvm *
 
 	kvm_free_physmem_slot(&old, npages ? &new : NULL);
 	/* Slot deletion case: we have to update the current slot */
+	spin_lock(&kvm->mmu_lock);
 	if (!npages)
 		*memslot = old;
+	spin_unlock(&kvm->mmu_lock);
 #ifdef CONFIG_DMAR
 	/* map the pages in iommu page table */
 	r = kvm_iommu_map_pages(kvm, base_gfn, npages);

[patch 3/4] KVM: introduce kvm_arch_can_free_memslot, disallow slot deletion if cached cr3

[mtosatti]

[-- Attachment #1: verify-cr3 --]
[-- Type: text/plain, Size: 4650 bytes --]

Disallow the deletion of memory slots (and aliases, for x86 case), if a
vcpu contains a cr3 that points to such slot/alias.

This complements commit 6c20e1442bb1c62914bb85b7f4a38973d2a423ba.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

Index: kvm/arch/ia64/kvm/kvm-ia64.c
===================================================================
--- kvm.orig/arch/ia64/kvm/kvm-ia64.c
+++ kvm/arch/ia64/kvm/kvm-ia64.c
@@ -1633,6 +1633,11 @@ void kvm_arch_flush_shadow(struct kvm *k
 	kvm_flush_remote_tlbs(kvm);
 }
 
+int kvm_arch_can_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
+{
+	return 1;
+}
+
 long kvm_arch_dev_ioctl(struct file *filp,
 			unsigned int ioctl, unsigned long arg)
 {
Index: kvm/arch/powerpc/kvm/powerpc.c
===================================================================
--- kvm.orig/arch/powerpc/kvm/powerpc.c
+++ kvm/arch/powerpc/kvm/powerpc.c
@@ -176,6 +176,11 @@ void kvm_arch_flush_shadow(struct kvm *k
 {
 }
 
+int kvm_arch_can_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
+{
+	return 1;
+}
+
 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 {
 	struct kvm_vcpu *vcpu;
Index: kvm/arch/s390/kvm/kvm-s390.c
===================================================================
--- kvm.orig/arch/s390/kvm/kvm-s390.c
+++ kvm/arch/s390/kvm/kvm-s390.c
@@ -691,6 +691,11 @@ void kvm_arch_flush_shadow(struct kvm *k
 {
 }
 
+int kvm_arch_can_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
+{
+	return 1;
+}
+
 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 {
 	return gfn;
Index: kvm/arch/x86/kvm/x86.c
===================================================================
--- kvm.orig/arch/x86/kvm/x86.c
+++ kvm/arch/x86/kvm/x86.c
@@ -1676,6 +1676,27 @@ gfn_t unalias_gfn(struct kvm *kvm, gfn_t
 	return gfn;
 }
 
+static int kvm_root_gfn_in_range(struct kvm *kvm, gfn_t base_gfn,
+				 gfn_t end_gfn, bool unalias)
+{
+	struct kvm_vcpu *vcpu;
+	gfn_t root_gfn;
+	int i;
+
+	for (i = 0; i < KVM_MAX_VCPUS; ++i) {
+		vcpu = kvm->vcpus[i];
+		if (!vcpu)
+			continue;
+		root_gfn = vcpu->arch.cr3 >> PAGE_SHIFT;
+		if (unalias)
+			root_gfn = unalias_gfn(kvm, root_gfn);
+		if (root_gfn >= base_gfn && root_gfn <= end_gfn)
+			return 1;
+	}
+
+	return 0;
+}
+
 /*
  * Set a new alias region.  Aliases map a portion of physical memory into
  * another portion.  This is useful for memory windows, for example the PC
@@ -1706,6 +1727,19 @@ static int kvm_vm_ioctl_set_memory_alias
 	spin_lock(&kvm->mmu_lock);
 
 	p = &kvm->arch.aliases[alias->slot];
+
+	/* FIXME: either disallow shrinking alias slots or disable
+ 	 * size changes as done with memslots
+ 	 */
+	if (!alias->memory_size) {
+		r = -EBUSY;
+		if (kvm_root_gfn_in_range(kvm, p->base_gfn,
+					       p->base_gfn + p->npages - 1,
+					       false))
+			goto out_unlock;
+	}
+
+
 	p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
 	p->npages = alias->memory_size >> PAGE_SHIFT;
 	p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
@@ -1722,6 +1756,9 @@ static int kvm_vm_ioctl_set_memory_alias
 
 	return 0;
 
+out_unlock:
+	spin_unlock(&kvm->mmu_lock);
+	up_write(&kvm->slots_lock);
 out:
 	return r;
 }
@@ -4532,6 +4569,15 @@ void kvm_arch_flush_shadow(struct kvm *k
 	kvm_mmu_zap_all(kvm);
 }
 
+int kvm_arch_can_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
+{
+	int ret;
+
+	ret = kvm_root_gfn_in_range(kvm, slot->base_gfn,
+				    slot->base_gfn + slot->npages - 1, true);
+	return !ret;
+}
+
 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
 {
 	return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
Index: kvm/include/linux/kvm_host.h
===================================================================
--- kvm.orig/include/linux/kvm_host.h
+++ kvm/include/linux/kvm_host.h
@@ -200,6 +200,7 @@ int kvm_arch_set_memory_region(struct kv
 				struct kvm_memory_slot old,
 				int user_alloc);
 void kvm_arch_flush_shadow(struct kvm *kvm);
+int kvm_arch_can_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot);
 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn);
 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn);
 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn);
Index: kvm/virt/kvm/kvm_main.c
===================================================================
--- kvm.orig/virt/kvm/kvm_main.c
+++ kvm/virt/kvm/kvm_main.c
@@ -1179,8 +1179,13 @@ int __kvm_set_memory_region(struct kvm *
 	}
 #endif /* not defined CONFIG_S390 */
 
-	if (!npages)
+	if (!npages) {
 		kvm_arch_flush_shadow(kvm);
+		if (!kvm_arch_can_free_memslot(kvm, memslot)) {
+			r = -EBUSY;
+			goto out_free;
+		}
+	}
 
 	spin_lock(&kvm->mmu_lock);
 	if (mem->slot >= kvm->nmemslots)

[patch 4/4] KVM: x86: disallow changing a slots size

[mtosatti]

[-- Attachment #1: disallow-alias-size-change --]
[-- Type: text/plain, Size: 1039 bytes --]

Support to shrinking aliases complicates kernel code unnecessarily,
while userspace can do the same with two operations, delete an alias,
and create a new alias.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

Index: kvm/arch/x86/kvm/x86.c
===================================================================
--- kvm.orig/arch/x86/kvm/x86.c
+++ kvm/arch/x86/kvm/x86.c
@@ -1707,6 +1707,7 @@ static int kvm_vm_ioctl_set_memory_alias
 {
 	int r, n;
 	struct kvm_mem_alias *p;
+	unsigned long npages;
 
 	r = -EINVAL;
 	/* General sanity checks */
@@ -1728,9 +1729,12 @@ static int kvm_vm_ioctl_set_memory_alias
 
 	p = &kvm->arch.aliases[alias->slot];
 
-	/* FIXME: either disallow shrinking alias slots or disable
- 	 * size changes as done with memslots
- 	 */
+	/* Disallow changing an alias slot's size. */
+	npages = alias->memory_size >> PAGE_SHIFT;
+	r = -EINVAL;
+	if (npages && p->npages && npages != p->npages)
+		goto out_unlock;
+
 	if (!alias->memory_size) {
 		r = -EBUSY;
 		if (kvm_root_gfn_in_range(kvm, p->base_gfn,

Re: [patch 3/4] KVM: introduce kvm_arch_can_free_memslot, disallow slot deletion if cached cr3

[Avi Kivity]

mtosatti@redhat.com wrote:
> Disallow the deletion of memory slots (and aliases, for x86 case), if a
> vcpu contains a cr3 that points to such slot/alias.
>   

That allows the guest to induce failures in the host.  Better to 
triple-fault the guest instead.

>  
> +int kvm_arch_can_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
> +{
> +	return 1;
> +}
> +
>   

In general, instead of stubs in every arch, have x86 say 
KVM_HAVE_ARCH_CAN_FREE_MEMSLOT and define the stub in generic code when 
that define is not present.


-- 
error compiling committee.c: too many arguments to function

Re: [patch 3/4] KVM: introduce kvm_arch_can_free_memslot, disallow slot deletion if cached cr3

[Marcelo Tosatti]

On Thu, May 07, 2009 at 05:16:35PM +0300, Avi Kivity wrote:
> mtosatti@redhat.com wrote:
>> Disallow the deletion of memory slots (and aliases, for x86 case), if a
>> vcpu contains a cr3 that points to such slot/alias.
>>   
>
> That allows the guest to induce failures in the host.  

I don't understand what you mean. What is the problem with returning
errors in the ioctl handlers?

The guest can cause an overflow in qemu, overwrite the parameters to
KVM_GET_MSR_INDEX_LIST in an attempt to read kernel data, and get
-E2BIG. Or pick your combination.

> Better to triple-fault the guest instead.

Sure can additionally triple fault it, but the kernel might attempt to
access the non-existant slot which cr3 points to before TRIPLE_FAULT is
processed. So you have to avoid that possibility in the first place,
thats why the patch modifies the ioctls to fail.

>>  +int kvm_arch_can_free_memslot(struct kvm *kvm, struct kvm_memory_slot 
>> *slot)
>> +{
>> +	return 1;
>> +}
>> +
>>   
>
> In general, instead of stubs in every arch, have x86 say  
> KVM_HAVE_ARCH_CAN_FREE_MEMSLOT and define the stub in generic code when  
> that define is not present.

Will fix that.

[patch 0/4] set_memory_region locking fixes / cr3 vs removal of memslots v2

[mtosatti]

Addressing comments.

[patch 1/4] KVM: MMU: protect kvm_mmu_change_mmu_pages with mmu_lock

[mtosatti]

[-- Attachment #1: set-mem-lock --]
[-- Type: text/plain, Size: 2187 bytes --]

kvm_handle_hva, called by MMU notifiers, manipulates mmu data only with
the protection of mmu_lock.

Update kvm_mmu_change_mmu_pages callers to take mmu_lock, thus protecting
against kvm_handle_hva.

CC: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

Index: kvm-pending/arch/x86/kvm/mmu.c
===================================================================
--- kvm-pending.orig/arch/x86/kvm/mmu.c
+++ kvm-pending/arch/x86/kvm/mmu.c
@@ -2723,7 +2723,6 @@ void kvm_mmu_slot_remove_write_access(st
 {
 	struct kvm_mmu_page *sp;
 
-	spin_lock(&kvm->mmu_lock);
 	list_for_each_entry(sp, &kvm->arch.active_mmu_pages, link) {
 		int i;
 		u64 *pt;
@@ -2738,7 +2737,6 @@ void kvm_mmu_slot_remove_write_access(st
 				pt[i] &= ~PT_WRITABLE_MASK;
 	}
 	kvm_flush_remote_tlbs(kvm);
-	spin_unlock(&kvm->mmu_lock);
 }
 
 void kvm_mmu_zap_all(struct kvm *kvm)
Index: kvm-pending/arch/x86/kvm/x86.c
===================================================================
--- kvm-pending.orig/arch/x86/kvm/x86.c
+++ kvm-pending/arch/x86/kvm/x86.c
@@ -1607,10 +1607,12 @@ static int kvm_vm_ioctl_set_nr_mmu_pages
 		return -EINVAL;
 
 	down_write(&kvm->slots_lock);
+	spin_lock(&kvm->mmu_lock);
 
 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
 
+	spin_unlock(&kvm->mmu_lock);
 	up_write(&kvm->slots_lock);
 	return 0;
 }
@@ -1786,7 +1788,9 @@ int kvm_vm_ioctl_get_dirty_log(struct kv
 
 	/* If nothing is dirty, don't bother messing with page tables. */
 	if (is_dirty) {
+		spin_lock(&kvm->mmu_lock);
 		kvm_mmu_slot_remove_write_access(kvm, log->slot);
+		spin_unlock(&kvm->mmu_lock);
 		kvm_flush_remote_tlbs(kvm);
 		memslot = &kvm->memslots[log->slot];
 		n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
@@ -4530,12 +4534,14 @@ int kvm_arch_set_memory_region(struct kv
 		}
 	}
 
+	spin_lock(&kvm->mmu_lock);
 	if (!kvm->arch.n_requested_mmu_pages) {
 		unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
 	}
 
 	kvm_mmu_slot_remove_write_access(kvm, mem->slot);
+	spin_unlock(&kvm->mmu_lock);
 	kvm_flush_remote_tlbs(kvm);
 
 	return 0;

[patch 2/4] KVM: take mmu_lock when updating a deleted slot

[mtosatti]

[-- Attachment #1: slot-delete-update-mmu-lock --]
[-- Type: text/plain, Size: 716 bytes --]

kvm_handle_hva relies on mmu_lock protection to safely access 
the memslot structures.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

Index: kvm-pending/virt/kvm/kvm_main.c
===================================================================
--- kvm-pending.orig/virt/kvm/kvm_main.c
+++ kvm-pending/virt/kvm/kvm_main.c
@@ -1199,8 +1199,10 @@ int __kvm_set_memory_region(struct kvm *
 
 	kvm_free_physmem_slot(&old, npages ? &new : NULL);
 	/* Slot deletion case: we have to update the current slot */
+	spin_lock(&kvm->mmu_lock);
 	if (!npages)
 		*memslot = old;
+	spin_unlock(&kvm->mmu_lock);
 #ifdef CONFIG_DMAR
 	/* map the pages in iommu page table */
 	r = kvm_iommu_map_pages(kvm, base_gfn, npages);

[patch 3/4] KVM: introduce kvm_arch_can_free_memslot, disallow slot deletion if cached cr3

[mtosatti]

[-- Attachment #1: verify-cr3 --]
[-- Type: text/plain, Size: 4286 bytes --]

Disallow the deletion of memory slots (and aliases, for x86 case), if a
vcpu contains a cr3 that points to such slot/alias.

This complements commit 6c20e1442bb1c62914bb85b7f4a38973d2a423ba.

v2:
- set KVM_REQ_TRIPLE_FAULT
- use __KVM_HAVE_ARCH_CAN_FREE_MEMSLOT to avoid duplication of stub

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

Index: kvm-pending/arch/x86/kvm/x86.c
===================================================================
--- kvm-pending.orig/arch/x86/kvm/x86.c
+++ kvm-pending/arch/x86/kvm/x86.c
@@ -1636,6 +1636,29 @@ gfn_t unalias_gfn(struct kvm *kvm, gfn_t
 	return gfn;
 }
 
+static int kvm_root_gfn_in_range(struct kvm *kvm, gfn_t base_gfn,
+				 gfn_t end_gfn, bool unalias)
+{
+	struct kvm_vcpu *vcpu;
+	gfn_t root_gfn;
+	int i;
+
+	for (i = 0; i < KVM_MAX_VCPUS; ++i) {
+		vcpu = kvm->vcpus[i];
+		if (!vcpu)
+			continue;
+		root_gfn = vcpu->arch.cr3 >> PAGE_SHIFT;
+		if (unalias)
+			root_gfn = unalias_gfn(kvm, root_gfn);
+		if (root_gfn >= base_gfn && root_gfn <= end_gfn) {
+			set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
+			return 1;
+		}
+	}
+
+	return 0;
+}
+
 /*
  * Set a new alias region.  Aliases map a portion of physical memory into
  * another portion.  This is useful for memory windows, for example the PC
@@ -1666,6 +1689,19 @@ static int kvm_vm_ioctl_set_memory_alias
 	spin_lock(&kvm->mmu_lock);
 
 	p = &kvm->arch.aliases[alias->slot];
+
+	/* FIXME: either disallow shrinking alias slots or disable
+ 	 * size changes as done with memslots
+ 	 */
+	if (!alias->memory_size) {
+		r = -EBUSY;
+		if (kvm_root_gfn_in_range(kvm, p->base_gfn,
+					       p->base_gfn + p->npages - 1,
+					       false))
+			goto out_unlock;
+	}
+
+
 	p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
 	p->npages = alias->memory_size >> PAGE_SHIFT;
 	p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
@@ -1682,6 +1718,9 @@ static int kvm_vm_ioctl_set_memory_alias
 
 	return 0;
 
+out_unlock:
+	spin_unlock(&kvm->mmu_lock);
+	up_write(&kvm->slots_lock);
 out:
 	return r;
 }
@@ -4552,6 +4591,15 @@ void kvm_arch_flush_shadow(struct kvm *k
 	kvm_mmu_zap_all(kvm);
 }
 
+int kvm_arch_can_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
+{
+	int ret;
+
+	ret = kvm_root_gfn_in_range(kvm, slot->base_gfn,
+				    slot->base_gfn + slot->npages - 1, true);
+	return !ret;
+}
+
 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
 {
 	return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
Index: kvm-pending/include/linux/kvm_host.h
===================================================================
--- kvm-pending.orig/include/linux/kvm_host.h
+++ kvm-pending/include/linux/kvm_host.h
@@ -200,6 +200,7 @@ int kvm_arch_set_memory_region(struct kv
 				struct kvm_memory_slot old,
 				int user_alloc);
 void kvm_arch_flush_shadow(struct kvm *kvm);
+int kvm_arch_can_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot);
 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn);
 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn);
 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn);
Index: kvm-pending/virt/kvm/kvm_main.c
===================================================================
--- kvm-pending.orig/virt/kvm/kvm_main.c
+++ kvm-pending/virt/kvm/kvm_main.c
@@ -1061,6 +1061,13 @@ static int kvm_vm_release(struct inode *
 	return 0;
 }
 
+#ifndef __KVM_HAVE_ARCH_CAN_FREE_MEMSLOT
+int kvm_arch_can_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
+{
+	return 1;
+}
+#endif
+
 /*
  * Allocate some memory and give it an address in the guest physical address
  * space.
@@ -1179,8 +1186,13 @@ int __kvm_set_memory_region(struct kvm *
 	}
 #endif /* not defined CONFIG_S390 */
 
-	if (!npages)
+	if (!npages) {
 		kvm_arch_flush_shadow(kvm);
+		if (!kvm_arch_can_free_memslot(kvm, memslot)) {
+			r = -EBUSY;
+			goto out_free;
+		}
+	}
 
 	spin_lock(&kvm->mmu_lock);
 	if (mem->slot >= kvm->nmemslots)
Index: kvm-pending/arch/x86/include/asm/kvm.h
===================================================================
--- kvm-pending.orig/arch/x86/include/asm/kvm.h
+++ kvm-pending/arch/x86/include/asm/kvm.h
@@ -18,6 +18,8 @@
 #define __KVM_HAVE_GUEST_DEBUG
 #define __KVM_HAVE_MSIX
 
+#define __KVM_HAVE_ARCH_CAN_FREE_MEMSLOT
+
 /* Architectural interrupt line count. */
 #define KVM_NR_INTERRUPTS 256
 

[patch 4/4] KVM: x86: disallow changing a slots size

[mtosatti]

[-- Attachment #1: disallow-alias-size-change --]
[-- Type: text/plain, Size: 1063 bytes --]

Support to shrinking aliases complicates kernel code unnecessarily,
while userspace can do the same with two operations, delete an alias,
and create a new alias.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

Index: kvm-pending/arch/x86/kvm/x86.c
===================================================================
--- kvm-pending.orig/arch/x86/kvm/x86.c
+++ kvm-pending/arch/x86/kvm/x86.c
@@ -1669,6 +1669,7 @@ static int kvm_vm_ioctl_set_memory_alias
 {
 	int r, n;
 	struct kvm_mem_alias *p;
+	unsigned long npages;
 
 	r = -EINVAL;
 	/* General sanity checks */
@@ -1690,9 +1691,12 @@ static int kvm_vm_ioctl_set_memory_alias
 
 	p = &kvm->arch.aliases[alias->slot];
 
-	/* FIXME: either disallow shrinking alias slots or disable
- 	 * size changes as done with memslots
- 	 */
+	/* Disallow changing an alias slot's size. */
+	npages = alias->memory_size >> PAGE_SHIFT;
+	r = -EINVAL;
+	if (npages && p->npages && npages != p->npages)
+		goto out_unlock;
+
 	if (!alias->memory_size) {
 		r = -EBUSY;
 		if (kvm_root_gfn_in_range(kvm, p->base_gfn,

Re: [patch 3/4] KVM: introduce kvm_arch_can_free_memslot, disallow slot deletion if cached cr3

[Avi Kivity]

mtosatti@redhat.com wrote:
> Disallow the deletion of memory slots (and aliases, for x86 case), if a
> vcpu contains a cr3 that points to such slot/alias.
>
> This complements commit 6c20e1442bb1c62914bb85b7f4a38973d2a423ba.
>
> v2:
> - set KVM_REQ_TRIPLE_FAULT
> - use __KVM_HAVE_ARCH_CAN_FREE_MEMSLOT to avoid duplication of stub
>
> Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
>
> Index: kvm-pending/arch/x86/kvm/x86.c
> ===================================================================
> --- kvm-pending.orig/arch/x86/kvm/x86.c
> +++ kvm-pending/arch/x86/kvm/x86.c
> @@ -1636,6 +1636,29 @@ gfn_t unalias_gfn(struct kvm *kvm, gfn_t
>  	return gfn;
>  }
>  
> +static int kvm_root_gfn_in_range(struct kvm *kvm, gfn_t base_gfn,
> +				 gfn_t end_gfn, bool unalias)
> +{
> +	struct kvm_vcpu *vcpu;
> +	gfn_t root_gfn;
> +	int i;
> +
> +	for (i = 0; i < KVM_MAX_VCPUS; ++i) {
> +		vcpu = kvm->vcpus[i];
> +		if (!vcpu)
> +			continue;
> +		root_gfn = vcpu->arch.cr3 >> PAGE_SHIFT;
>   

The guest may have changed this by now.

> +		if (unalias)
> +			root_gfn = unalias_gfn(kvm, root_gfn);
> +		if (root_gfn >= base_gfn && root_gfn <= end_gfn) {
> +			set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
> +			return 1;
> +		}
> +	}
> +
> +	return 0;
> +}
> +
>   

The naming is bad, a function named as a predicate shouldn't have side 
effects.

Also, we should allow deleting the slot.  There's no reason to deny 
userspace something just because the guest is playing around

I think this should be enough:
- take mmu lock
- request an mmu reload from all vcpus
- drop the slot
- release mmu lock

The reload will inject a #GP if cr3 is now out of bounds, should be 
changed to triple fault, but everything is in place (set_cr3 already 
checks).

-- 
error compiling committee.c: too many arguments to function

[patch 0/3] locking fixes / cr3 validation v3

[mtosatti]

Addressing comments.

[patch 1/3] KVM: MMU: protect kvm_mmu_change_mmu_pages with mmu_lock

[mtosatti]

[-- Attachment #1: set-mem-lock --]
[-- Type: text/plain, Size: 2144 bytes --]

kvm_handle_hva, called by MMU notifiers, manipulates mmu data only with
the protection of mmu_lock.

Update kvm_mmu_change_mmu_pages callers to take mmu_lock, thus protecting
against kvm_handle_hva.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

Index: kvm-pending/arch/x86/kvm/mmu.c
===================================================================
--- kvm-pending.orig/arch/x86/kvm/mmu.c
+++ kvm-pending/arch/x86/kvm/mmu.c
@@ -2723,7 +2723,6 @@ void kvm_mmu_slot_remove_write_access(st
 {
 	struct kvm_mmu_page *sp;
 
-	spin_lock(&kvm->mmu_lock);
 	list_for_each_entry(sp, &kvm->arch.active_mmu_pages, link) {
 		int i;
 		u64 *pt;
@@ -2738,7 +2737,6 @@ void kvm_mmu_slot_remove_write_access(st
 				pt[i] &= ~PT_WRITABLE_MASK;
 	}
 	kvm_flush_remote_tlbs(kvm);
-	spin_unlock(&kvm->mmu_lock);
 }
 
 void kvm_mmu_zap_all(struct kvm *kvm)
Index: kvm-pending/arch/x86/kvm/x86.c
===================================================================
--- kvm-pending.orig/arch/x86/kvm/x86.c
+++ kvm-pending/arch/x86/kvm/x86.c
@@ -1607,10 +1607,12 @@ static int kvm_vm_ioctl_set_nr_mmu_pages
 		return -EINVAL;
 
 	down_write(&kvm->slots_lock);
+	spin_lock(&kvm->mmu_lock);
 
 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
 
+	spin_unlock(&kvm->mmu_lock);
 	up_write(&kvm->slots_lock);
 	return 0;
 }
@@ -1786,7 +1788,9 @@ int kvm_vm_ioctl_get_dirty_log(struct kv
 
 	/* If nothing is dirty, don't bother messing with page tables. */
 	if (is_dirty) {
+		spin_lock(&kvm->mmu_lock);
 		kvm_mmu_slot_remove_write_access(kvm, log->slot);
+		spin_unlock(&kvm->mmu_lock);
 		kvm_flush_remote_tlbs(kvm);
 		memslot = &kvm->memslots[log->slot];
 		n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
@@ -4530,12 +4534,14 @@ int kvm_arch_set_memory_region(struct kv
 		}
 	}
 
+	spin_lock(&kvm->mmu_lock);
 	if (!kvm->arch.n_requested_mmu_pages) {
 		unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
 	}
 
 	kvm_mmu_slot_remove_write_access(kvm, mem->slot);
+	spin_unlock(&kvm->mmu_lock);
 	kvm_flush_remote_tlbs(kvm);
 
 	return 0;

[patch 2/3] KVM: take mmu_lock when updating a deleted slot

[mtosatti]

[-- Attachment #1: slot-delete-update-mmu-lock --]
[-- Type: text/plain, Size: 716 bytes --]

kvm_handle_hva relies on mmu_lock protection to safely access 
the memslot structures.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

Index: kvm-pending/virt/kvm/kvm_main.c
===================================================================
--- kvm-pending.orig/virt/kvm/kvm_main.c
+++ kvm-pending/virt/kvm/kvm_main.c
@@ -1199,8 +1199,10 @@ int __kvm_set_memory_region(struct kvm *
 
 	kvm_free_physmem_slot(&old, npages ? &new : NULL);
 	/* Slot deletion case: we have to update the current slot */
+	spin_lock(&kvm->mmu_lock);
 	if (!npages)
 		*memslot = old;
+	spin_unlock(&kvm->mmu_lock);
 #ifdef CONFIG_DMAR
 	/* map the pages in iommu page table */
 	r = kvm_iommu_map_pages(kvm, base_gfn, npages);

[patch 3/3] KVM: x86: check for cr3 validity in mmu_alloc_roots

[mtosatti]

[-- Attachment #1: reload-cr3 --]
[-- Type: text/plain, Size: 2863 bytes --]

Verify the cr3 address stored in vcpu->arch.cr3 points to an existant
memslot. If not, inject a triple fault.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

Index: kvm/arch/x86/kvm/mmu.c
===================================================================
--- kvm.orig/arch/x86/kvm/mmu.c
+++ kvm/arch/x86/kvm/mmu.c
@@ -1912,7 +1912,19 @@ static void mmu_free_roots(struct kvm_vc
 	vcpu->arch.mmu.root_hpa = INVALID_PAGE;
 }
 
-static void mmu_alloc_roots(struct kvm_vcpu *vcpu)
+static int mmu_check_root(struct kvm_vcpu *vcpu, gfn_t root_gfn)
+{
+	int ret = 0;
+
+	if (!kvm_is_visible_gfn(vcpu->kvm, root_gfn)) {
+		set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
+		ret = 1;
+	}
+
+	return ret;
+}
+
+static int mmu_alloc_roots(struct kvm_vcpu *vcpu)
 {
 	int i;
 	gfn_t root_gfn;
@@ -1927,13 +1939,15 @@ static void mmu_alloc_roots(struct kvm_v
 		ASSERT(!VALID_PAGE(root));
 		if (tdp_enabled)
 			direct = 1;
+		if (mmu_check_root(vcpu, root_gfn))
+			return 1;
 		sp = kvm_mmu_get_page(vcpu, root_gfn, 0,
 				      PT64_ROOT_LEVEL, direct,
 				      ACC_ALL, NULL);
 		root = __pa(sp->spt);
 		++sp->root_count;
 		vcpu->arch.mmu.root_hpa = root;
-		return;
+		return 0;
 	}
 	direct = !is_paging(vcpu);
 	if (tdp_enabled)
@@ -1950,6 +1964,8 @@ static void mmu_alloc_roots(struct kvm_v
 			root_gfn = vcpu->arch.pdptrs[i] >> PAGE_SHIFT;
 		} else if (vcpu->arch.mmu.root_level == 0)
 			root_gfn = 0;
+		if (mmu_check_root(vcpu, root_gfn))
+			return 1;
 		sp = kvm_mmu_get_page(vcpu, root_gfn, i << 30,
 				      PT32_ROOT_LEVEL, direct,
 				      ACC_ALL, NULL);
@@ -1958,6 +1974,7 @@ static void mmu_alloc_roots(struct kvm_v
 		vcpu->arch.mmu.pae_root[i] = root | PT_PRESENT_MASK;
 	}
 	vcpu->arch.mmu.root_hpa = __pa(vcpu->arch.mmu.pae_root);
+	return 0;
 }
 
 static void mmu_sync_roots(struct kvm_vcpu *vcpu)
@@ -1976,7 +1993,7 @@ static void mmu_sync_roots(struct kvm_vc
 	for (i = 0; i < 4; ++i) {
 		hpa_t root = vcpu->arch.mmu.pae_root[i];
 
-		if (root) {
+		if (root && VALID_PAGE(root)) {
 			root &= PT64_BASE_ADDR_MASK;
 			sp = page_header(root);
 			mmu_sync_children(vcpu, sp);
@@ -2311,9 +2328,11 @@ int kvm_mmu_load(struct kvm_vcpu *vcpu)
 		goto out;
 	spin_lock(&vcpu->kvm->mmu_lock);
 	kvm_mmu_free_some_pages(vcpu);
-	mmu_alloc_roots(vcpu);
+	r = mmu_alloc_roots(vcpu);
 	mmu_sync_roots(vcpu);
 	spin_unlock(&vcpu->kvm->mmu_lock);
+	if (r)
+		goto out;
 	kvm_x86_ops->set_cr3(vcpu, vcpu->arch.mmu.root_hpa);
 	kvm_mmu_flush_tlb(vcpu);
 out:
Index: kvm/arch/x86/kvm/x86.c
===================================================================
--- kvm.orig/arch/x86/kvm/x86.c
+++ kvm/arch/x86/kvm/x86.c
@@ -4554,6 +4554,7 @@ int kvm_arch_set_memory_region(struct kv
 void kvm_arch_flush_shadow(struct kvm *kvm)
 {
 	kvm_mmu_zap_all(kvm);
+	kvm_reload_remote_mmus(kvm);
 }
 
 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)

Re: [patch 0/3] locking fixes / cr3 validation v3

[Avi Kivity]

mtosatti@redhat.com wrote:
> Addressing comments.
>
>
>   

Applied all.  But please fix you From: header.

-- 
Do not meddle in the internals of kernels, for they are subtle and quick to panic.