Message ID | 1408622216-9578-5-git-send-email-famz@redhat.com |
---|---|
State | New |
Headers | show |
On Thu, Aug 21, 2014 at 07:56:51PM +0800, Fam Zheng wrote: > @@ -110,6 +109,22 @@ static void qemu_laio_completion_cb(EventNotifier *e) > } > } > > +static void laio_cancel_async(BlockDriverAIOCB *blockacb) > +{ > + struct qemu_laiocb *laiocb = (struct qemu_laiocb *)blockacb; > + struct io_event event; > + int ret; > + > + ret = io_cancel(laiocb->ctx->ctx, &laiocb->iocb, &event); > + laiocb->ret = -ECANCELED; > + if (!ret) { > + /* iocb is not cancelled, cb will be called by the event loop later */ > + return; > + } No callback will be invoked if io_cancel(2) every cancels the request immediately. The current kernel implementation always returns -EINPROGRESS or some of other error value. But some day it might return 0 and this would leak the request! > + > + laiocb->common.cb(laiocb->common.opaque, laiocb->ret); > +} It would be cleaner to reuse laio_cancel_async() from laio_cancel() to avoid code duplication. For example, there is a useful comment in laio_cancel() explaining that io_cancel(2) doesn't cancel I/O in practice on 2.6.31 era kernels.
On Thu, 08/21 17:31, Stefan Hajnoczi wrote: > On Thu, Aug 21, 2014 at 07:56:51PM +0800, Fam Zheng wrote: > > @@ -110,6 +109,22 @@ static void qemu_laio_completion_cb(EventNotifier *e) > > } > > } > > > > +static void laio_cancel_async(BlockDriverAIOCB *blockacb) > > +{ > > + struct qemu_laiocb *laiocb = (struct qemu_laiocb *)blockacb; > > + struct io_event event; > > + int ret; > > + > > + ret = io_cancel(laiocb->ctx->ctx, &laiocb->iocb, &event); > > + laiocb->ret = -ECANCELED; > > + if (!ret) { > > + /* iocb is not cancelled, cb will be called by the event loop later */ > > + return; > > + } > > No callback will be invoked if io_cancel(2) every cancels the request > immediately. > > The current kernel implementation always returns -EINPROGRESS or some of > other error value. But some day it might return 0 and this would leak > the request! > > > + > > + laiocb->common.cb(laiocb->common.opaque, laiocb->ret); > > +} > > It would be cleaner to reuse laio_cancel_async() from laio_cancel() to > avoid code duplication. For example, there is a useful comment in > laio_cancel() explaining that io_cancel(2) doesn't cancel I/O in > practice on 2.6.31 era kernels. I'll take a closer look at it. Fam
diff --git a/block/linux-aio.c b/block/linux-aio.c index 7ac7e8c..adf3b2e 100644 --- a/block/linux-aio.c +++ b/block/linux-aio.c @@ -79,9 +79,8 @@ static void qemu_laio_process_completion(struct qemu_laio_state *s, ret = -EINVAL; } } - - laiocb->common.cb(laiocb->common.opaque, ret); } + laiocb->common.cb(laiocb->common.opaque, ret); qemu_aio_release(laiocb); } @@ -110,6 +109,22 @@ static void qemu_laio_completion_cb(EventNotifier *e) } } +static void laio_cancel_async(BlockDriverAIOCB *blockacb) +{ + struct qemu_laiocb *laiocb = (struct qemu_laiocb *)blockacb; + struct io_event event; + int ret; + + ret = io_cancel(laiocb->ctx->ctx, &laiocb->iocb, &event); + laiocb->ret = -ECANCELED; + if (!ret) { + /* iocb is not cancelled, cb will be called by the event loop later */ + return; + } + + laiocb->common.cb(laiocb->common.opaque, laiocb->ret); +} + static void laio_cancel(BlockDriverAIOCB *blockacb) { struct qemu_laiocb *laiocb = (struct qemu_laiocb *)blockacb; @@ -145,6 +160,7 @@ static void laio_cancel(BlockDriverAIOCB *blockacb) static const AIOCBInfo laio_aiocb_info = { .aiocb_size = sizeof(struct qemu_laiocb), .cancel = laio_cancel, + .cancel_async = laio_cancel_async, }; static void ioq_init(LaioQueue *io_q)
Just call io_cancel (2), if it fails, it means the request is not canceled, so the event loop will eventually call qemu_laio_process_completion. In qemu_laio_process_completion, change to call the cb unconditionally. It is required by .cancel_async, and also acceptable by .cancel. Signed-off-by: Fam Zheng <famz@redhat.com> --- block/linux-aio.c | 20 ++++++++++++++++++-- 1 file changed, 18 insertions(+), 2 deletions(-)