[PATCH v6 06/20] block/export: wait for vhost-user-blk requests when draining

[Stefan Hajnoczi <stefanha@xxxxxxxxxx>]

Each vhost-user-blk request runs in a coroutine. When the BlockBackend
enters a drained section we need to enter a quiescent state. Currently
any in-flight requests race with bdrv_drained_begin() because it is
unaware of vhost-user-blk requests.

When blk_co_preadv/pwritev()/etc returns it wakes the
bdrv_drained_begin() thread but vhost-user-blk request processing has
not yet finished. The request coroutine continues executing while the
main loop thread thinks it is in a drained section.

One example where this is unsafe is for blk_set_aio_context() where
bdrv_drained_begin() is called before .aio_context_detached() and
.aio_context_attach(). If request coroutines are still running after
bdrv_drained_begin(), then the AioContext could change underneath them
and they race with new requests processed in the new AioContext. This
could lead to virtqueue corruption, for example.

(This example is theoretical, I came across this while reading the
code and have not tried to reproduce it.)

It's easy to make bdrv_drained_begin() wait for in-flight requests: add
a .drained_poll() callback that checks the VuServer's in-flight counter.
VuServer just needs an API that returns true when there are requests in
flight. The in-flight counter needs to be atomic.

Signed-off-by: Stefan Hajnoczi <stefanha@xxxxxxxxxx>
Reviewed-by: Kevin Wolf <kwolf@xxxxxxxxxx>
---
v5:
- Use atomic accesses for in_flight counter in vhost-user-server.c [Kevin]
---
 include/qemu/vhost-user-server.h     |  4 +++-
 block/export/vhost-user-blk-server.c | 13 +++++++++++++
 util/vhost-user-server.c             | 18 ++++++++++++------
 3 files changed, 28 insertions(+), 7 deletions(-)

diff --git a/include/qemu/vhost-user-server.h b/include/qemu/vhost-user-server.h
index bc0ac9ddb6..b1c1cda886 100644
--- a/include/qemu/vhost-user-server.h
+++ b/include/qemu/vhost-user-server.h
@@ -40,8 +40,9 @@ typedef struct {
     int max_queues;
     const VuDevIface *vu_iface;
 
+    unsigned int in_flight; /* atomic */
+
     /* Protected by ctx lock */
-    unsigned int in_flight;
     bool wait_idle;
     VuDev vu_dev;
     QIOChannel *ioc; /* The I/O channel with the client */
@@ -62,6 +63,7 @@ void vhost_user_server_stop(VuServer *server);
 
 void vhost_user_server_inc_in_flight(VuServer *server);
 void vhost_user_server_dec_in_flight(VuServer *server);
+bool vhost_user_server_has_in_flight(VuServer *server);
 
 void vhost_user_server_attach_aio_context(VuServer *server, AioContext *ctx);
 void vhost_user_server_detach_aio_context(VuServer *server);
diff --git a/block/export/vhost-user-blk-server.c 
b/block/export/vhost-user-blk-server.c
index 841acb36e3..f51a36a14f 100644
--- a/block/export/vhost-user-blk-server.c
+++ b/block/export/vhost-user-blk-server.c
@@ -272,7 +272,20 @@ static void vu_blk_exp_resize(void *opaque)
     vu_config_change_msg(&vexp->vu_server.vu_dev);
 }
 
+/*
+ * Ensures that bdrv_drained_begin() waits until in-flight requests complete.
+ *
+ * Called with vexp->export.ctx acquired.
+ */
+static bool vu_blk_drained_poll(void *opaque)
+{
+    VuBlkExport *vexp = opaque;
+
+    return vhost_user_server_has_in_flight(&vexp->vu_server);
+}
+
 static const BlockDevOps vu_blk_dev_ops = {
+    .drained_poll  = vu_blk_drained_poll,
     .resize_cb = vu_blk_exp_resize,
 };
 
diff --git a/util/vhost-user-server.c b/util/vhost-user-server.c
index 1622f8cfb3..68c3bf162f 100644
--- a/util/vhost-user-server.c
+++ b/util/vhost-user-server.c
@@ -78,17 +78,23 @@ static void panic_cb(VuDev *vu_dev, const char *buf)
 void vhost_user_server_inc_in_flight(VuServer *server)
 {
     assert(!server->wait_idle);
-    server->in_flight++;
+    qatomic_inc(&server->in_flight);
 }
 
 void vhost_user_server_dec_in_flight(VuServer *server)
 {
-    server->in_flight--;
-    if (server->wait_idle && !server->in_flight) {
-        aio_co_wake(server->co_trip);
+    if (qatomic_fetch_dec(&server->in_flight) == 1) {
+        if (server->wait_idle) {
+            aio_co_wake(server->co_trip);
+        }
     }
 }
 
+bool vhost_user_server_has_in_flight(VuServer *server)
+{
+    return qatomic_load_acquire(&server->in_flight) > 0;
+}
+
 static bool coroutine_fn
 vu_message_read(VuDev *vu_dev, int conn_fd, VhostUserMsg *vmsg)
 {
@@ -192,13 +198,13 @@ static coroutine_fn void vu_client_trip(void *opaque)
         /* Keep running */
     }
 
-    if (server->in_flight) {
+    if (vhost_user_server_has_in_flight(server)) {
         /* Wait for requests to complete before we can unmap the memory */
         server->wait_idle = true;
         qemu_coroutine_yield();
         server->wait_idle = false;
     }
-    assert(server->in_flight == 0);
+    assert(!vhost_user_server_has_in_flight(server));
 
     vu_deinit(vu_dev);
 
-- 
2.40.1