Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1167639AbdDXKMn (ORCPT ); Mon, 24 Apr 2017 06:12:43 -0400 Received: from foss.arm.com ([217.140.101.70]:54422 "EHLO foss.arm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1167275AbdDXKMP (ORCPT ); Mon, 24 Apr 2017 06:12:15 -0400 From: Suzuki K Poulose To: pbonzini@redhat.com Cc: christoffer.dall@linaro.org, linux-kernel@vger.kernel.org, linux-arm-kernel@lists.infradead.org, kvmarm@lists.cs.columbia.edu, kvm@vger.kernel.org, marc.zyngier@arm.com, mark.rutland@arm.com, andreyknvl@google.com, rkrcmar@redhat.com, Suzuki K Poulose Subject: [PATCH 1/2] kvm: Fix mmu_notifier release race Date: Mon, 24 Apr 2017 11:10:23 +0100 Message-Id: <1493028624-29837-2-git-send-email-suzuki.poulose@arm.com> X-Mailer: git-send-email 2.7.4 In-Reply-To: <1493028624-29837-1-git-send-email-suzuki.poulose@arm.com> References: <1493028624-29837-1-git-send-email-suzuki.poulose@arm.com> MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 6025 Lines: 165 The KVM uses mmu_notifier (wherever available) to keep track of the changes to the mm of the guest. The guest shadow page tables are released when the VM exits via mmu_notifier->ops.release(). There is a rare chance that the mmu_notifier->release could be called more than once via two different paths, which could end up in use-after-free of kvm instance (such as [0]). e.g: thread A thread B ------- -------------- get_signal-> kvm_destroy_vm()-> do_exit-> mmu_notifier_unregister-> exit_mm-> kvm_arch_flush_shadow_all()-> exit_mmap-> spin_lock(&kvm->mmu_lock) mmu_notifier_release-> .... kvm_arch_flush_shadow_all()-> ..... ... spin_lock(&kvm->mmu_lock) ..... spin_unlock(&kvm->mmu_lock) kvm_arch_free_kvm() *** use after free of kvm *** This patch attempts to solve the problem by holding a reference to the KVM for the mmu_notifier, which is dropped only from notifier->ops.release(). This will ensure that the KVM struct is available till we reach the kvm_mmu_notifier_release, and the kvm_destroy_vm is called only from/after it. So, we can unregister the notifier with no_release option and hence avoiding the race above. However, we need to make sure that the KVM is freed only after the mmu_notifier has finished processing the notifier due to the following possible path of execution : mmu_notifier_release -> kvm_mmu_notifier_release -> kvm_put_kvm -> kvm_destroy_vm -> kvm_arch_free_kvm [0] http://lkml.kernel.org/r/CAAeHK+x8udHKq9xa1zkTO6ax5E8Dk32HYWfaT05FMchL2cr48g@mail.gmail.com Fixes: commit 85db06e514422 ("KVM: mmu_notifiers release method") Reported-by: andreyknvl@google.com Cc: Mark Rutland Cc: Paolo Bonzini Cc: Radim Krčmář Cc: Marc Zyngier Cc: Christoffer Dall Cc: andreyknvl@google.com Cc: Marc Zyngier Tested-by: Mark Rutland Signed-off-by: Suzuki K Poulose --- include/linux/kvm_host.h | 1 + virt/kvm/kvm_main.c | 59 ++++++++++++++++++++++++++++++++++++++++++------ 2 files changed, 53 insertions(+), 7 deletions(-) diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h index d025074..561e968 100644 --- a/include/linux/kvm_host.h +++ b/include/linux/kvm_host.h @@ -424,6 +424,7 @@ struct kvm { struct mmu_notifier mmu_notifier; unsigned long mmu_notifier_seq; long mmu_notifier_count; + struct rcu_head mmu_notifier_rcu; #endif long tlbs_dirty; struct list_head devices; diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c index 88257b3..2c3fdd4 100644 --- a/virt/kvm/kvm_main.c +++ b/virt/kvm/kvm_main.c @@ -471,6 +471,7 @@ static void kvm_mmu_notifier_release(struct mmu_notifier *mn, idx = srcu_read_lock(&kvm->srcu); kvm_arch_flush_shadow_all(kvm); srcu_read_unlock(&kvm->srcu, idx); + kvm_put_kvm(kvm); } static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { @@ -486,8 +487,46 @@ static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { static int kvm_init_mmu_notifier(struct kvm *kvm) { + int rc; kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; - return mmu_notifier_register(&kvm->mmu_notifier, current->mm); + rc = mmu_notifier_register(&kvm->mmu_notifier, current->mm); + /* + * We hold a reference to KVM here to make sure that the KVM + * doesn't get free'd before ops->release() completes. + */ + if (!rc) + kvm_get_kvm(kvm); + return rc; +} + +static void kvm_free_vm_rcu(struct rcu_head *rcu) +{ + struct kvm *kvm = container_of(rcu, struct kvm, mmu_notifier_rcu); + kvm_arch_free_vm(kvm); +} + +static void kvm_flush_shadow_mmu(struct kvm *kvm) +{ + /* + * We hold a reference to kvm instance for mmu_notifier and is + * only released when ops->release() is called via exit_mmap path. + * So, when we reach here ops->release() has been called already, which + * flushes the shadow page tables. Hence there is no need to call the + * release() again when we unregister the notifier. However, we need + * to delay freeing up the kvm until the release() completes, since + * we could reach here via : + * kvm_mmu_notifier_release() -> kvm_put_kvm() -> kvm_destroy_vm() + */ + mmu_notifier_unregister_no_release(&kvm->mmu_notifier, kvm->mm); +} + +static void kvm_free_vm(struct kvm *kvm) +{ + /* + * Wait until the mmu_notifier has finished the release(). + * See comments above in kvm_flush_shadow_mmu. + */ + mmu_notifier_call_srcu(&kvm->mmu_notifier_rcu, kvm_free_vm_rcu); } #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ @@ -497,6 +536,16 @@ static int kvm_init_mmu_notifier(struct kvm *kvm) return 0; } +static void kvm_flush_shadow_mmu(struct kvm *kvm) +{ + kvm_arch_flush_shadow_all(kvm); +} + +static void kvm_free_vm(struct kvm *kvm) +{ + kvm_arch_free_vm(kvm); +} + #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ static struct kvm_memslots *kvm_alloc_memslots(void) @@ -733,18 +782,14 @@ static void kvm_destroy_vm(struct kvm *kvm) kvm->buses[i] = NULL; } kvm_coalesced_mmio_free(kvm); -#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) - mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); -#else - kvm_arch_flush_shadow_all(kvm); -#endif + kvm_flush_shadow_mmu(kvm); kvm_arch_destroy_vm(kvm); kvm_destroy_devices(kvm); for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) kvm_free_memslots(kvm, kvm->memslots[i]); cleanup_srcu_struct(&kvm->irq_srcu); cleanup_srcu_struct(&kvm->srcu); - kvm_arch_free_vm(kvm); + kvm_free_vm(kvm); preempt_notifier_dec(); hardware_disable_all(); mmdrop(mm); -- 2.7.4