From 9a789d104a4a69031ad95d7fad6380ab21e82503 Mon Sep 17 00:00:00 2001

From: Kevin Wolf <kwolf@redhat.com>

Date: Fri, 18 Nov 2022 18:41:08 +0100

Subject: [PATCH 26/31] block: Drop out of coroutine in

 bdrv_do_drained_begin_quiesce()

RH-Author: Stefano Garzarella <sgarzare@redhat.com>

RH-MergeRequest: 135: block: Simplify drain to prevent QEMU from crashing during snapshot

RH-Bugzilla: 2155112

RH-Acked-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>

RH-Acked-by: Hanna Czenczek <hreitz@redhat.com>

RH-Acked-by: Kevin Wolf <kwolf@redhat.com>

RH-Commit: [14/16] c9266663b822f703e55b6a07de98ceb56e69e924 (sgarzarella/qemu-kvm-c-9-s)

The next patch adds a parent drain to bdrv_attach_child_common(), which

shouldn't be, but is currently called from coroutines in some cases (e.g.

.bdrv_co_create implementations generally open new nodes). Therefore,

the assertion that we're not in a coroutine doesn't hold true any more.

We could just remove the assertion because there is nothing in the

function that should be in conflict with running in a coroutine, but

just to be on the safe side, we can reverse the caller relationship

between bdrv_do_drained_begin() and bdrv_do_drained_begin_quiesce() so

that the latter also just drops out of coroutine context and we can

still be certain in the future that any drain code doesn't run in

coroutines.

As a nice side effect, the structure of bdrv_do_drained_begin() is now

symmetrical with bdrv_do_drained_end().

Signed-off-by: Kevin Wolf <kwolf@redhat.com>

Message-Id: <20221118174110.55183-14-kwolf@redhat.com>

Reviewed-by: Hanna Reitz <hreitz@redhat.com>

Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>

Signed-off-by: Kevin Wolf <kwolf@redhat.com>

(cherry picked from commit 05c272ff0cf1b16cc3606f746182dd99b774f553)

Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>

---

 block/io.c | 25 ++++++++++++-------------

 1 file changed, 12 insertions(+), 13 deletions(-)

diff --git a/block/io.c b/block/io.c

index 2e9503df6a..5e9150d92c 100644

--- a/block/io.c

+++ b/block/io.c

@@ -346,10 +346,15 @@ static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs,

     }

 }

-void bdrv_do_drained_begin_quiesce(BlockDriverState *bs, BdrvChild *parent)

+static void bdrv_do_drained_begin(BlockDriverState *bs, BdrvChild *parent,

+                                  bool poll)

 {

     IO_OR_GS_CODE();

-    assert(!qemu_in_coroutine());

+

+    if (qemu_in_coroutine()) {

+        bdrv_co_yield_to_drain(bs, true, parent, poll);

+        return;

+    }

     /* Stop things in parent-to-child order */

     if (qatomic_fetch_inc(&bs->quiesce_counter) == 0) {

@@ -359,17 +364,6 @@ void bdrv_do_drained_begin_quiesce(BlockDriverState *bs, BdrvChild *parent)

             bs->drv->bdrv_drain_begin(bs);

         }

     }

-}

-

-static void bdrv_do_drained_begin(BlockDriverState *bs, BdrvChild *parent,

-                                  bool poll)

-{

-    if (qemu_in_coroutine()) {

-        bdrv_co_yield_to_drain(bs, true, parent, poll);

-        return;

-    }

-

-    bdrv_do_drained_begin_quiesce(bs, parent);

     /*

      * Wait for drained requests to finish.

@@ -385,6 +379,11 @@ static void bdrv_do_drained_begin(BlockDriverState *bs, BdrvChild *parent,

     }

 }

+void bdrv_do_drained_begin_quiesce(BlockDriverState *bs, BdrvChild *parent)

+{

+    bdrv_do_drained_begin(bs, parent, false);

+}

+

 void bdrv_drained_begin(BlockDriverState *bs)

 {

     IO_OR_GS_CODE();

-- 

2.31.1