Virtualization - Jun 2023 - [PATCH 1/1] vhost: Fix crash during early vhost_transport_send_pkt calls

If userspace does VHOST_VSOCK_SET_GUEST_CID before VHOST_SET_OWNER we
can race where:
1. thread0 calls vhost_transport_send_pkt -> vhost_work_queue
2. thread1 does VHOST_SET_OWNER which calls vhost_worker_create.
3. vhost_worker_create will set the dev->worker pointer before setting
the worker->vtsk pointer.
4. thread0's vhost_work_queue will see the dev->worker pointer is
set and try to call vhost_task_wake using not yet set worker->vtsk
pointer.
5. We then crash since vtsk is NULL.

Before commit 6e890c5d5021 ("vhost: use vhost_tasks for worker
threads"), we only had the worker pointer so we could just check it to
see if VHOST_SET_OWNER has been done. After that commit we have the
vhost_worker and vhost_task pointers, so we can now hit the bug above.

This patch embeds the vhost_worker in the vhost_dev, so we can just
check the worker.vtsk pointer to check if VHOST_SET_OWNER has been done
like before.

Fixes: 6e890c5d5021 ("vhost: use vhost_tasks for worker threads")
Signed-off-by: Mike Christie <michael.christie at oracle.com>
---
 drivers/vhost/vhost.c | 50 +++++++++++++++----------------------------
 drivers/vhost/vhost.h |  2 +-
 2 files changed, 18 insertions(+), 34 deletions(-)

diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
index 074273020849..0ad9fea7c170 100644
--- a/drivers/vhost/vhost.c
+++ b/drivers/vhost/vhost.c
@@ -235,7 +235,7 @@ void vhost_dev_flush(struct vhost_dev *dev)
 {
 	struct vhost_flush_struct flush;
 
-	if (dev->worker) {
+	if (dev->worker.vtsk) {
 		init_completion(&flush.wait_event);
 		vhost_work_init(&flush.work, vhost_flush_work);
 
@@ -247,7 +247,7 @@ EXPORT_SYMBOL_GPL(vhost_dev_flush);
 
 void vhost_work_queue(struct vhost_dev *dev, struct vhost_work *work)
 {
-	if (!dev->worker)
+	if (!dev->worker.vtsk)
 		return;
 
 	if (!test_and_set_bit(VHOST_WORK_QUEUED, &work->flags)) {
@@ -255,8 +255,8 @@ void vhost_work_queue(struct vhost_dev *dev, struct
vhost_work *work)
 		 * sure it was not in the list.
 		 * test_and_set_bit() implies a memory barrier.
 		 */
-		llist_add(&work->node, &dev->worker->work_list);
-		vhost_task_wake(dev->worker->vtsk);
+		llist_add(&work->node, &dev->worker.work_list);
+		vhost_task_wake(dev->worker.vtsk);
 	}
 }
 EXPORT_SYMBOL_GPL(vhost_work_queue);
@@ -264,7 +264,7 @@ EXPORT_SYMBOL_GPL(vhost_work_queue);
 /* A lockless hint for busy polling code to exit the loop */
 bool vhost_has_work(struct vhost_dev *dev)
 {
-	return dev->worker &&
!llist_empty(&dev->worker->work_list);
+	return !llist_empty(&dev->worker.work_list);
 }
 EXPORT_SYMBOL_GPL(vhost_has_work);
 
@@ -456,7 +456,7 @@ void vhost_dev_init(struct vhost_dev *dev,
 	dev->umem = NULL;
 	dev->iotlb = NULL;
 	dev->mm = NULL;
-	dev->worker = NULL;
+	memset(&dev->worker, 0, sizeof(dev->worker));
 	dev->iov_limit = iov_limit;
 	dev->weight = weight;
 	dev->byte_weight = byte_weight;
@@ -530,47 +530,31 @@ static void vhost_detach_mm(struct vhost_dev *dev)
 
 static void vhost_worker_free(struct vhost_dev *dev)
 {
-	struct vhost_worker *worker = dev->worker;
-
-	if (!worker)
+	if (!dev->worker.vtsk)
 		return;
 
-	dev->worker = NULL;
-	WARN_ON(!llist_empty(&worker->work_list));
-	vhost_task_stop(worker->vtsk);
-	kfree(worker);
+	WARN_ON(!llist_empty(&dev->worker.work_list));
+	vhost_task_stop(dev->worker.vtsk);
+	dev->worker.kcov_handle = 0;
+	dev->worker.vtsk = NULL;
 }
 
 static int vhost_worker_create(struct vhost_dev *dev)
 {
-	struct vhost_worker *worker;
 	struct vhost_task *vtsk;
 	char name[TASK_COMM_LEN];
-	int ret;
-
-	worker = kzalloc(sizeof(*worker), GFP_KERNEL_ACCOUNT);
-	if (!worker)
-		return -ENOMEM;
 
-	dev->worker = worker;
-	worker->kcov_handle = kcov_common_handle();
-	init_llist_head(&worker->work_list);
+	init_llist_head(&dev->worker.work_list);
 	snprintf(name, sizeof(name), "vhost-%d", current->pid);
 
-	vtsk = vhost_task_create(vhost_worker, worker, name);
-	if (!vtsk) {
-		ret = -ENOMEM;
-		goto free_worker;
-	}
+	vtsk = vhost_task_create(vhost_worker, &dev->worker, name);
+	if (!vtsk)
+		return -ENOMEM;
 
-	worker->vtsk = vtsk;
+	dev->worker.kcov_handle = kcov_common_handle();
+	dev->worker.vtsk = vtsk;
 	vhost_task_start(vtsk);
 	return 0;
-
-free_worker:
-	kfree(worker);
-	dev->worker = NULL;
-	return ret;
 }
 
 /* Caller should have device mutex */
diff --git a/drivers/vhost/vhost.h b/drivers/vhost/vhost.h
index 0308638cdeee..305ec8593d46 100644
--- a/drivers/vhost/vhost.h
+++ b/drivers/vhost/vhost.h
@@ -154,7 +154,7 @@ struct vhost_dev {
 	struct vhost_virtqueue **vqs;
 	int nvqs;
 	struct eventfd_ctx *log_ctx;
-	struct vhost_worker *worker;
+	struct vhost_worker worker;
 	struct vhost_iotlb *umem;
 	struct vhost_iotlb *iotlb;
 	spinlock_t iotlb_lock;
-- 
2.25.1

On Mon, Jun 05, 2023 at 01:57:30PM -0500, Mike Christie
wrote:
>If userspace does VHOST_VSOCK_SET_GUEST_CID before VHOST_SET_OWNER we
>can race where:
>1. thread0 calls vhost_transport_send_pkt -> vhost_work_queue
>2. thread1 does VHOST_SET_OWNER which calls vhost_worker_create.
>3. vhost_worker_create will set the dev->worker pointer before setting
>the worker->vtsk pointer.
>4. thread0's vhost_work_queue will see the dev->worker pointer is
>set and try to call vhost_task_wake using not yet set worker->vtsk
>pointer.
>5. We then crash since vtsk is NULL.
>
>Before commit 6e890c5d5021 ("vhost: use vhost_tasks for worker
>threads"), we only had the worker pointer so we could just check it to
>see if VHOST_SET_OWNER has been done. After that commit we have the
>vhost_worker and vhost_task pointers, so we can now hit the bug above.
>
>This patch embeds the vhost_worker in the vhost_dev, so we can just
>check the worker.vtsk pointer to check if VHOST_SET_OWNER has been done
>like before.
>
>Fixes: 6e890c5d5021 ("vhost: use vhost_tasks for worker threads")
We should add:

Reported-by: syzbot+d0d442c22fa8db45ff0e at syzkaller.appspotmail.com
>Signed-off-by: Mike Christie <michael.christie at oracle.com>
>---
> drivers/vhost/vhost.c | 50 +++++++++++++++----------------------------
> drivers/vhost/vhost.h |  2 +-
> 2 files changed, 18 insertions(+), 34 deletions(-)
>
>diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
>index 074273020849..0ad9fea7c170 100644
>--- a/drivers/vhost/vhost.c
>+++ b/drivers/vhost/vhost.c
>@@ -235,7 +235,7 @@ void vhost_dev_flush(struct vhost_dev *dev)
> {
> 	struct vhost_flush_struct flush;
>
>-	if (dev->worker) {
>+	if (dev->worker.vtsk) {
> 		init_completion(&flush.wait_event);
> 		vhost_work_init(&flush.work, vhost_flush_work);
>
>@@ -247,7 +247,7 @@ EXPORT_SYMBOL_GPL(vhost_dev_flush);
>
> void vhost_work_queue(struct vhost_dev *dev, struct vhost_work *work)
> {
>-	if (!dev->worker)
>+	if (!dev->worker.vtsk)
> 		return;
>
> 	if (!test_and_set_bit(VHOST_WORK_QUEUED, &work->flags)) {
>@@ -255,8 +255,8 @@ void vhost_work_queue(struct vhost_dev *dev, struct
vhost_work *work)
> 		 * sure it was not in the list.
> 		 * test_and_set_bit() implies a memory barrier.
> 		 */
>-		llist_add(&work->node, &dev->worker->work_list);
>-		vhost_task_wake(dev->worker->vtsk);
>+		llist_add(&work->node, &dev->worker.work_list);
>+		vhost_task_wake(dev->worker.vtsk);
> 	}
> }
> EXPORT_SYMBOL_GPL(vhost_work_queue);
>@@ -264,7 +264,7 @@ EXPORT_SYMBOL_GPL(vhost_work_queue);
> /* A lockless hint for busy polling code to exit the loop */
> bool vhost_has_work(struct vhost_dev *dev)
> {
>-	return dev->worker &&
!llist_empty(&dev->worker->work_list);
>+	return !llist_empty(&dev->worker.work_list);
> }
> EXPORT_SYMBOL_GPL(vhost_has_work);
>
>@@ -456,7 +456,7 @@ void vhost_dev_init(struct vhost_dev *dev,
> 	dev->umem = NULL;
> 	dev->iotlb = NULL;
> 	dev->mm = NULL;
>-	dev->worker = NULL;
>+	memset(&dev->worker, 0, sizeof(dev->worker));
> 	dev->iov_limit = iov_limit;
> 	dev->weight = weight;
> 	dev->byte_weight = byte_weight;
>@@ -530,47 +530,31 @@ static void vhost_detach_mm(struct vhost_dev *dev)
>
> static void vhost_worker_free(struct vhost_dev *dev)
> {
>-	struct vhost_worker *worker = dev->worker;
>-
>-	if (!worker)
>+	if (!dev->worker.vtsk)
> 		return;
>
>-	dev->worker = NULL;
>-	WARN_ON(!llist_empty(&worker->work_list));
>-	vhost_task_stop(worker->vtsk);
>-	kfree(worker);
>+	WARN_ON(!llist_empty(&dev->worker.work_list));
>+	vhost_task_stop(dev->worker.vtsk);
>+	dev->worker.kcov_handle = 0;
>+	dev->worker.vtsk = NULL;
> }
>
> static int vhost_worker_create(struct vhost_dev *dev)
> {
>-	struct vhost_worker *worker;
> 	struct vhost_task *vtsk;
> 	char name[TASK_COMM_LEN];
>-	int ret;
>-
>-	worker = kzalloc(sizeof(*worker), GFP_KERNEL_ACCOUNT);
>-	if (!worker)
>-		return -ENOMEM;
>
>-	dev->worker = worker;
>-	worker->kcov_handle = kcov_common_handle();
>-	init_llist_head(&worker->work_list);
>+	init_llist_head(&dev->worker.work_list);
> 	snprintf(name, sizeof(name), "vhost-%d", current->pid);
>
>-	vtsk = vhost_task_create(vhost_worker, worker, name);
>-	if (!vtsk) {
>-		ret = -ENOMEM;
>-		goto free_worker;
>-	}
>+	vtsk = vhost_task_create(vhost_worker, &dev->worker, name);
>+	if (!vtsk)
>+		return -ENOMEM;
>
>-	worker->vtsk = vtsk;
>+	dev->worker.kcov_handle = kcov_common_handle();
>+	dev->worker.vtsk = vtsk;
vhost_work_queue() is called by vhost_transport_send_pkt() without
holding vhost_dev.mutex (like vhost_poll_queue() in several places).

If vhost_work_queue() finds dev->worker.vtsk not NULL, how can we
be sure that for example `work_list` has been initialized?

Maybe I'm overthinking since we didn't have this problem before or the
race is really short that it never happened.

Thanks,
Stefano