> On 23.12.2015 11:48, Paolo Bonzini wrote: > > bdrv_close is used when ejecting a medium. > > Is it still? Other than it maybe being indirectly called through > bdrv_delete(), it shouldn't be.
Yes, through blk_remove_bs -> bdrv_unref -> bdrv_delete. > > Use a drained section to ensure > > that all I/O goes to either the old medium or the bitbucket. > > Since the BDS will be deleted after bdrv_close() has been called, where > else would the I/O go now? The ioeventfd could be processed during bs->drv->bdrv_close. For example I/O could be submitted while qcow2_close's qcow2_cache_flush yields, and the result would probably be corrupted metadata. Paolo > Max > > > Signed-off-by: Paolo Bonzini <[email protected]> > > --- > > block.c | 4 +++- > > 1 file changed, 3 insertions(+), 1 deletion(-) > > > > diff --git a/block.c b/block.c > > index 411edbf..01655de 100644 > > --- a/block.c > > +++ b/block.c > > @@ -2154,9 +2154,10 @@ void bdrv_close(BlockDriverState *bs) > > bdrv_io_limits_disable(bs); > > } > > > > - bdrv_drain(bs); /* complete I/O */ > > + bdrv_drained_begin(bs); /* complete I/O */ > > bdrv_flush(bs); > > bdrv_drain(bs); /* in case flush left pending I/O */ > > + > > notifier_list_notify(&bs->close_notifiers, bs); > > > > if (bs->blk) { > > @@ -2206,6 +2207,7 @@ void bdrv_close(BlockDriverState *bs) > > g_free(ban); > > } > > QLIST_INIT(&bs->aio_notifiers); > > + bdrv_drained_end(bs); > > } > > > > void bdrv_close_all(void) > > > > >
