Skip to contentSkip to conversation

powerpc/kvm/book3s: Fixes possible 'use after release' of kvm

PATCH In review Leonardo Bras 4 participants 7 messages Nov 26, 2019 – Nov 28, 2019

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 powerpc/kvm/book3s: Fixes possible 'use after release' of kvm

Fixes a possible 'use after free' of kvm variable.
It does use mutex_unlock(&kvm->lock) after possible freeing a variable with kvm_put_kvm(kvm).

---
 arch/powerpc/kvm/book3s_64_vio.c | 3 +--
 virt/kvm/kvm_main.c              | 8 ++++----
 2 files changed, 5 insertions(+), 6 deletions(-)
arch/powerpc/kvm/book3s_64_vio.c +1 -2
index 5834db0a54c6..a402ead833b6 100644
@@ -316,14 +316,13 @@ long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
 
 	if (ret >= 0)
 		list_add_rcu(&stt->list, &kvm->arch.spapr_tce_tables);
-	else
-		kvm_put_kvm(kvm);
 
 	mutex_unlock(&kvm->lock);
 
 	if (ret >= 0)
 		return ret;
 
+	kvm_put_kvm(kvm);
 	kfree(stt);
  fail_acct:
 	account_locked_vm(current->mm, kvmppc_stt_pages(npages), false);
virt/kvm/kvm_main.c +4 -4
index 13efc291b1c7..f37089b60d09 100644
@@ -2744,10 +2744,8 @@ static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
 	/* Now it's all set up, let userspace reach it */
 	kvm_get_kvm(kvm);
 	r = create_vcpu_fd(vcpu);
-	if (r < 0) {
-		kvm_put_kvm(kvm);
+	if (r < 0)
 		goto unlock_vcpu_destroy;
-	}
 
 	kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
 
@@ -2771,6 +2769,8 @@ static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
 	mutex_lock(&kvm->lock);
 	kvm->created_vcpus--;
 	mutex_unlock(&kvm->lock);
+	if (r < 0)
+		kvm_put_kvm(kvm);
 	return r;
 }
 
@@ -3183,10 +3183,10 @@ static int kvm_ioctl_create_device(struct kvm *kvm,
 	kvm_get_kvm(kvm);
 	ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
 	if (ret < 0) {
-		kvm_put_kvm(kvm);
 		mutex_lock(&kvm->lock);
 		list_del(&dev->vm_node);
 		mutex_unlock(&kvm->lock);
+		kvm_put_kvm(kvm);
 		ops->destroy(dev);
 		return ret;
 	}
Signature · 1 line
2.23.0
#1 lore raw
Quoted 3 lines · Leonardo Bras

Fixes a possible 'use after free' of kvm variable.
It does use mutex_unlock(&kvm->lock) after possible freeing a variable with kvm_put_kvm(kvm).

Moving the calls to kvm_put_kvm() to the end of the functions doesn't actually fix a use-after-free. In these flows, the reference being released is a borrowed reference that KVM takes on behalf of the entity it is creating, e.g. device, vcpu, or spapr tce. The caller of these create helpers must also hold its own reference to @kvm on top of the borrowed reference, i.e. these kvm_put_kvm() calls will never free @kvm (assuming there are no refcounting bugs elsewhere in KVM).

If one these kvm_put_kvm() calls did unexpectedly free @kvm (due to a bug somewhere else), KVM would still hit a use-after-free scenario as the caller still thinks @kvm is valid. Currently, this would only happen on a subsequent ioctl() on the caller's file descriptor (which holds a pointer to @kvm), as the callers of these functions don't directly dereference @kvm after the functions return. But, not deferencing @kvm isn't deliberate or functionally required, it's just how the code happens to be written.

The intent of adding kvm_put_kvm_no_destroy() was primarily to document that under no circumstance should the to-be-put reference be the *last* reference to @kvm. Moving the call to kvm_put_kvm{_no_destroy}() doesn't change that

Quoted 66 lines
---
 arch/powerpc/kvm/book3s_64_vio.c | 3 +--
 virt/kvm/kvm_main.c              | 8 ++++----
 2 files changed, 5 insertions(+), 6 deletions(-)
arch/powerpc/kvm/book3s_64_vio.c +1 -2
index 5834db0a54c6..a402ead833b6 100644
@@ -316,14 +316,13 @@ long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
 
 	if (ret >= 0)
 		list_add_rcu(&stt->list, &kvm->arch.spapr_tce_tables);
-	else
-		kvm_put_kvm(kvm);
 
 	mutex_unlock(&kvm->lock);
 
 	if (ret >= 0)
 		return ret;
 
+	kvm_put_kvm(kvm);
 	kfree(stt);
  fail_acct:
 	account_locked_vm(current->mm, kvmppc_stt_pages(npages), false);
virt/kvm/kvm_main.c +4 -4
index 13efc291b1c7..f37089b60d09 100644
@@ -2744,10 +2744,8 @@ static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
 	/* Now it's all set up, let userspace reach it */
 	kvm_get_kvm(kvm);
 	r = create_vcpu_fd(vcpu);
-	if (r < 0) {
-		kvm_put_kvm(kvm);
+	if (r < 0)
 		goto unlock_vcpu_destroy;
-	}
 
 	kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
 
@@ -2771,6 +2769,8 @@ static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
 	mutex_lock(&kvm->lock);
 	kvm->created_vcpus--;
 	mutex_unlock(&kvm->lock);
+	if (r < 0)
+		kvm_put_kvm(kvm);
 	return r;
 }
 
@@ -3183,10 +3183,10 @@ static int kvm_ioctl_create_device(struct kvm *kvm,
 	kvm_get_kvm(kvm);
 	ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
 	if (ret < 0) {
-		kvm_put_kvm(kvm);
 		mutex_lock(&kvm->lock);
 		list_del(&dev->vm_node);
 		mutex_unlock(&kvm->lock);
+		kvm_put_kvm(kvm);
 		ops->destroy(dev);
 		return ret;
 	}
Signature · 1 line
2.23.0
maintainer
#2 lore raw
In reply to Leonardo Bras
Quoted 20 lines · Leonardo Bras

Fixes a possible 'use after free' of kvm variable.
It does use mutex_unlock(&kvm->lock) after possible freeing a variable with kvm_put_kvm(kvm).

---
 arch/powerpc/kvm/book3s_64_vio.c | 3 +--
 virt/kvm/kvm_main.c              | 8 ++++----
 2 files changed, 5 insertions(+), 6 deletions(-)
arch/powerpc/kvm/book3s_64_vio.c +0 -2
index 5834db0a54c6..a402ead833b6 100644
@@ -316,14 +316,13 @@ long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
 
 	if (ret >= 0)
 		list_add_rcu(&stt->list, &kvm->arch.spapr_tce_tables);
-	else
-		kvm_put_kvm(kvm);

arch/powerpc/kvm/book3s_64_vio.c long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm, · line 328

Quoted 10 lines
arch/powerpc/kvm/book3s_64_vio.c +1 -0
 
 	mutex_unlock(&kvm->lock);
 
 	if (ret >= 0)
 		return ret;
 
+	kvm_put_kvm(kvm);
 	kfree(stt);
  fail_acct:
 	account_locked_vm(current->mm, kvmppc_stt_pages(npages), false);

This part is a good change, as it makes the code clearer. The virt/kvm/kvm_main.c bits, however, are not necessary as explained by Sean.

Paolo

Quoted 38 lines
virt/kvm/kvm_main.c +4 -4
index 13efc291b1c7..f37089b60d09 100644
@@ -2744,10 +2744,8 @@ static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
 	/* Now it's all set up, let userspace reach it */
 	kvm_get_kvm(kvm);
 	r = create_vcpu_fd(vcpu);
-	if (r < 0) {
-		kvm_put_kvm(kvm);
+	if (r < 0)
 		goto unlock_vcpu_destroy;
-	}
 
 	kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
 
@@ -2771,6 +2769,8 @@ static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
 	mutex_lock(&kvm->lock);
 	kvm->created_vcpus--;
 	mutex_unlock(&kvm->lock);
+	if (r < 0)
+		kvm_put_kvm(kvm);
 	return r;
 }
 
@@ -3183,10 +3183,10 @@ static int kvm_ioctl_create_device(struct kvm *kvm,
 	kvm_get_kvm(kvm);
 	ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
 	if (ret < 0) {
-		kvm_put_kvm(kvm);
 		mutex_lock(&kvm->lock);
 		list_del(&dev->vm_node);
 		mutex_unlock(&kvm->lock);
+		kvm_put_kvm(kvm);
 		ops->destroy(dev);
 		return ret;
 	}
#3 lore raw
Quoted 11 lines · Paolo Bonzini
Quoted 11 lines
  
       if (ret >= 0)
               list_add_rcu(&stt->list, &kvm->arch.spapr_tce_tables);
arch/powerpc/kvm/book3s_64_vio.c +0 -2
-     else
-             kvm_put_kvm(kvm);
 
       mutex_unlock(&kvm->lock);
 
       if (ret >= 0)
               return ret;
Quoted 8 lines
Quoted 4 lines
+     kvm_put_kvm(kvm);
       kfree(stt);
   fail_acct:
       account_locked_vm(current->mm, kvmppc_stt_pages(npages), false);

This part is a good change, as it makes the code clearer. The virt/kvm/kvm_main.c bits, however, are not necessary as explained by Sean.

Thanks!
So, like this patch?
https://lkml.org/lkml/2019/11/7/763

Best regards,

Leonardo

#4 lore raw
In reply to Paolo Bonzini
Quoted 13 lines · Paolo Bonzini
Quoted 13 lines
arch/powerpc/kvm/book3s_64_vio.c +0 -2
index 5834db0a54c6..a402ead833b6 100644
@@ -316,14 +316,13 @@ long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
 
       if (ret >= 0)
               list_add_rcu(&stt->list, &kvm->arch.spapr_tce_tables);
-     else
-             kvm_put_kvm(kvm);
 
       mutex_unlock(&kvm->lock);
 
Quoted 7 lines

arch/powerpc/kvm/book3s_64_vio.c long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm, · line 328

Quoted 7 lines
arch/powerpc/kvm/book3s_64_vio.c +1 -0
       if (ret >= 0)
               return ret;
 
+     kvm_put_kvm(kvm);
       kfree(stt);
   fail_acct:
       account_locked_vm(current->mm, kvmppc_stt_pages(npages), false);

Paul, do you think this change is still valid as it 'makes the code clearer', as said by Paolo before? I would write a new commit message to match the change.

Best regards,
Leonardo

#5 lore raw
In reply to Leonardo Bras
Quoted 3 lines · Leonardo Bras

Fixes a possible 'use after free' of kvm variable.
It does use mutex_unlock(&kvm->lock) after possible freeing a variable with kvm_put_kvm(kvm).

Comments below...

Quoted 7 lines
arch/powerpc/kvm/book3s_64_vio.c +0 -0
index 5834db0a54c6..a402ead833b6 100644
@@ -316,14 +316,13 @@ long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
 
 	if (ret >= 0)

arch/powerpc/kvm/book3s_64_vio.c long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm, · line 325

Quoted 10 lines
arch/powerpc/kvm/book3s_64_vio.c +1 -2
 		list_add_rcu(&stt->list, &kvm->arch.spapr_tce_tables);
-	else
-		kvm_put_kvm(kvm);
 
 	mutex_unlock(&kvm->lock);
 
 	if (ret >= 0)
 		return ret;
 
+	kvm_put_kvm(kvm);

There isn't a potential use-after-free here. We are relying on the property that the release function (kvm_vm_release) cannot be called in parallel with this function. The reason is that this function (kvm_vm_ioctl_create_spapr_tce) is handling an ioctl on a kvm VM file descriptor. That means that a userspace process has the file descriptor still open. The code that implements the close() system call makes sure that no thread is still executing inside any system call that is using the same file descriptor before calling the file descriptor's release function (in this case, kvm_vm_release). That means that this kvm_put_kvm() call here cannot make the reference count go to zero.

Quoted 20 lines
arch/powerpc/kvm/book3s_64_vio.c +0 -0
 	kfree(stt);
  fail_acct:
 	account_locked_vm(current->mm, kvmppc_stt_pages(npages), false);
virt/kvm/kvm_main.c +1 -3
index 13efc291b1c7..f37089b60d09 100644
@@ -2744,10 +2744,8 @@ static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
 	/* Now it's all set up, let userspace reach it */
 	kvm_get_kvm(kvm);
 	r = create_vcpu_fd(vcpu);
-	if (r < 0) {
-		kvm_put_kvm(kvm);
+	if (r < 0)
 		goto unlock_vcpu_destroy;
-	}
 
 	kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
 

virt/kvm/kvm_main.c static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) · line 2775

Quoted 7 lines
virt/kvm/kvm_main.c +2 -0
@@ -2771,6 +2769,8 @@ static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
 	mutex_lock(&kvm->lock);
 	kvm->created_vcpus--;
 	mutex_unlock(&kvm->lock);
+	if (r < 0)
+		kvm_put_kvm(kvm);
 	return r;
 }

Once again we are inside an ioctl on the kvm VM file descriptor, so the reference count cannot go to zero.

virt/kvm/kvm_main.c static int kvm_ioctl_create_device(struct kvm *kvm, · line 3192

Quoted 12 lines
virt/kvm/kvm_main.c +1 -1
@@ -3183,10 +3183,10 @@ static int kvm_ioctl_create_device(struct kvm *kvm,
 	kvm_get_kvm(kvm);
 	ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
 	if (ret < 0) {
-		kvm_put_kvm(kvm);
 		mutex_lock(&kvm->lock);
 		list_del(&dev->vm_node);
 		mutex_unlock(&kvm->lock);
+		kvm_put_kvm(kvm);
 		ops->destroy(dev);
 		return ret;
 	}

Same again here.

Paul.

#6 lore raw
Quoted 11 lines · Paul Mackerras

There isn't a potential use-after-free here. We are relying on the property that the release function (kvm_vm_release) cannot be called in parallel with this function. The reason is that this function (kvm_vm_ioctl_create_spapr_tce) is handling an ioctl on a kvm VM file descriptor. That means that a userspace process has the file descriptor still open. The code that implements the close() system call makes sure that no thread is still executing inside any system call that is using the same file descriptor before calling the file descriptor's release function (in this case, kvm_vm_release). That means that this kvm_put_kvm() call here cannot make the reference count go to zero.

That was very informative. A lot of things are clear to me now. Thanks for explaining this Paul.

Best regards,
Leonardo