From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:48679) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1ey7fh-0003Sw-OI for qemu-devel@nongnu.org; Mon, 19 Mar 2018 23:10:46 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1ey7fg-0004rj-Mh for qemu-devel@nongnu.org; Mon, 19 Mar 2018 23:10:45 -0400 Date: Mon, 19 Mar 2018 23:10:28 -0400 From: Jeff Cody Message-ID: <20180320031028.GD2962@localhost.localdomain> References: <20180228180507.3964-1-mreitz@redhat.com> <20180228180507.3964-2-mreitz@redhat.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20180228180507.3964-2-mreitz@redhat.com> Subject: Re: [Qemu-devel] [Qemu-block] [PATCH v3 01/16] block: BDS deletion during bdrv_drain_recurse List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Max Reitz Cc: qemu-block@nongnu.org, Kevin Wolf , Fam Zheng , qemu-devel@nongnu.org, Stefan Hajnoczi On Wed, Feb 28, 2018 at 07:04:52PM +0100, Max Reitz wrote: > Draining a BDS child may lead to other children of the same parent being > detached and/or deleted. We should prepare for the former case (by > copying the children list before iterating through it) and prevent the > latter (by bdrv_ref()'ing all nodes if we are in the main loop). > > Signed-off-by: Max Reitz > --- > block/io.c | 40 ++++++++++++++++++++++++++++++---------- > 1 file changed, 30 insertions(+), 10 deletions(-) > > diff --git a/block/io.c b/block/io.c > index 4d3d1f640a..ead12c4136 100644 > --- a/block/io.c > +++ b/block/io.c > @@ -189,31 +189,51 @@ static void bdrv_drain_invoke(BlockDriverState *bs, bool begin, bool recursive) > > static bool bdrv_drain_recurse(BlockDriverState *bs) > { > - BdrvChild *child, *tmp; > + BdrvChild *child; > bool waited; > + struct BDSToDrain { > + BlockDriverState *bs; > + QLIST_ENTRY(BDSToDrain) next; > + }; > + QLIST_HEAD(, BDSToDrain) bs_list = QLIST_HEAD_INITIALIZER(bs_list); > + bool in_main_loop = > + qemu_get_current_aio_context() == qemu_get_aio_context(); > > /* Wait for drained requests to finish */ > waited = BDRV_POLL_WHILE(bs, atomic_read(&bs->in_flight) > 0); > > - QLIST_FOREACH_SAFE(child, &bs->children, next, tmp) { > - BlockDriverState *bs = child->bs; > - bool in_main_loop = > - qemu_get_current_aio_context() == qemu_get_aio_context(); > - assert(bs->refcnt > 0); > + /* Draining children may result in other children being removed from this > + * parent and maybe even deleted, so copy the children list first */ > + QLIST_FOREACH(child, &bs->children, next) { > + struct BDSToDrain *bs2d = g_new0(struct BDSToDrain, 1); > + > + bs2d->bs = child->bs; > if (in_main_loop) { > /* In case the recursive bdrv_drain_recurse processes a > * block_job_defer_to_main_loop BH and modifies the graph, > - * let's hold a reference to bs until we are done. > + * let's hold a reference to the BDS until we are done. > * > * IOThread doesn't have such a BH, and it is not safe to call > * bdrv_unref without BQL, so skip doing it there. > */ > - bdrv_ref(bs); > + bdrv_ref(bs2d->bs); > } > - waited |= bdrv_drain_recurse(bs); > + > + QLIST_INSERT_HEAD(&bs_list, bs2d, next); > + } > + > + while (!QLIST_EMPTY(&bs_list)) { > + struct BDSToDrain *bs2d = QLIST_FIRST(&bs_list); > + QLIST_REMOVE(bs2d, next); > + > + assert(bs2d->bs->refcnt > 0); > + > + waited |= bdrv_drain_recurse(bs2d->bs); > if (in_main_loop) { > - bdrv_unref(bs); > + bdrv_unref(bs2d->bs); > } > + > + g_free(bs2d); > } > > return waited; > -- > 2.14.3 > > Reviewed-by: Jeff Cody