block/file-posix.c | 7 +++++++ block/linux-aio.c | 22 +++++++++++++++++++++- 2 files changed, 28 insertions(+), 1 deletion(-)
From: Prasad Pandit <pjp@fedoraproject.org>
Libaio defines IO_CMD_FDSYNC command to sync all outstanding
asynchronous I/O operations, by flushing out file data to the
disk storage.
Enable linux-aio to submit such aio request. This helps to
reduce latency induced via pthread_create calls by
thread-pool (aio=threads).
Signed-off-by: Prasad Pandit <pjp@fedoraproject.org>
---
block/file-posix.c | 7 +++++++
block/linux-aio.c | 22 +++++++++++++++++++++-
2 files changed, 28 insertions(+), 1 deletion(-)
v3: check if host kernel supports aio_fsync call
-> https://lists.nongnu.org/archive/html/qemu-devel/2024-03/msg03210.html
diff --git a/block/file-posix.c b/block/file-posix.c
index 35684f7e21..a30494d1a8 100644
--- a/block/file-posix.c
+++ b/block/file-posix.c
@@ -2453,6 +2453,8 @@ static inline bool raw_check_linux_io_uring(BDRVRawState *s)
#endif
#ifdef CONFIG_LINUX_AIO
+extern bool laio_has_fdsync(int);
+
static inline bool raw_check_linux_aio(BDRVRawState *s)
{
Error *local_err = NULL;
@@ -2599,6 +2601,11 @@ static int coroutine_fn raw_co_flush_to_disk(BlockDriverState *bs)
if (raw_check_linux_io_uring(s)) {
return luring_co_submit(bs, s->fd, 0, NULL, QEMU_AIO_FLUSH);
}
+#endif
+#ifdef CONFIG_LINUX_AIO
+ if (raw_check_linux_aio(s) && laio_has_fdsync(s->fd)) {
+ return laio_co_submit(s->fd, 0, NULL, QEMU_AIO_FLUSH, 0);
+ }
#endif
return raw_thread_pool_submit(handle_aiocb_flush, &acb);
}
diff --git a/block/linux-aio.c b/block/linux-aio.c
index ec05d946f3..ed20a503e9 100644
--- a/block/linux-aio.c
+++ b/block/linux-aio.c
@@ -67,6 +67,7 @@ struct LinuxAioState {
int event_max;
};
+bool laio_has_fdsync(int);
static void ioq_submit(LinuxAioState *s);
static inline ssize_t io_event_ret(struct io_event *ev)
@@ -384,6 +385,9 @@ static int laio_do_submit(int fd, struct qemu_laiocb *laiocb, off_t offset,
case QEMU_AIO_READ:
io_prep_preadv(iocbs, fd, qiov->iov, qiov->niov, offset);
break;
+ case QEMU_AIO_FLUSH:
+ io_prep_fdsync(iocbs, fd);
+ break;
/* Currently Linux kernel does not support other operations */
default:
fprintf(stderr, "%s: invalid AIO request type 0x%x.\n",
@@ -412,7 +416,7 @@ int coroutine_fn laio_co_submit(int fd, uint64_t offset, QEMUIOVector *qiov,
AioContext *ctx = qemu_get_current_aio_context();
struct qemu_laiocb laiocb = {
.co = qemu_coroutine_self(),
- .nbytes = qiov->size,
+ .nbytes = qiov ? qiov->size : 0,
.ctx = aio_get_linux_aio(ctx),
.ret = -EINPROGRESS,
.is_read = (type == QEMU_AIO_READ),
@@ -486,3 +490,19 @@ void laio_cleanup(LinuxAioState *s)
}
g_free(s);
}
+
+bool laio_has_fdsync(int fd)
+{
+ struct iocb cb;
+ struct iocb *cbs[] = {&cb, NULL};
+
+ io_context_t ctx = 0;
+ io_setup(1, &ctx);
+
+ /* check if host kernel supports IO_CMD_FDSYNC */
+ io_prep_fdsync(&cb, fd);
+ int ret = io_submit(ctx, 1, cbs);
+
+ io_destroy(ctx);
+ return (ret == -EINVAL) ? false : true;
+}
--
2.44.0
On Wed, Mar 13, 2024 at 02:19:35PM +0530, Prasad Pandit wrote:
> From: Prasad Pandit <pjp@fedoraproject.org>
>
> Libaio defines IO_CMD_FDSYNC command to sync all outstanding
> asynchronous I/O operations, by flushing out file data to the
> disk storage.
>
> Enable linux-aio to submit such aio request. This helps to
> reduce latency induced via pthread_create calls by
> thread-pool (aio=threads).
>
> Signed-off-by: Prasad Pandit <pjp@fedoraproject.org>
> ---
> block/file-posix.c | 7 +++++++
> block/linux-aio.c | 22 +++++++++++++++++++++-
> 2 files changed, 28 insertions(+), 1 deletion(-)
>
> v3: check if host kernel supports aio_fsync call
> -> https://lists.nongnu.org/archive/html/qemu-devel/2024-03/msg03210.html
>
> diff --git a/block/file-posix.c b/block/file-posix.c
> index 35684f7e21..a30494d1a8 100644
> --- a/block/file-posix.c
> +++ b/block/file-posix.c
> @@ -2453,6 +2453,8 @@ static inline bool raw_check_linux_io_uring(BDRVRawState *s)
> #endif
>
> #ifdef CONFIG_LINUX_AIO
> +extern bool laio_has_fdsync(int);
Please declare this in include/block/raw-aio.h alongside the other laio APIs.
> +
> static inline bool raw_check_linux_aio(BDRVRawState *s)
> {
> Error *local_err = NULL;
> @@ -2599,6 +2601,11 @@ static int coroutine_fn raw_co_flush_to_disk(BlockDriverState *bs)
> if (raw_check_linux_io_uring(s)) {
> return luring_co_submit(bs, s->fd, 0, NULL, QEMU_AIO_FLUSH);
> }
> +#endif
> +#ifdef CONFIG_LINUX_AIO
> + if (raw_check_linux_aio(s) && laio_has_fdsync(s->fd)) {
FDSYNC support should be probed at open() time and the result should be
stored in a new bool field like s->laio_supports_fdsync. That way the
cost of laio_has_fdsync() on every flush request is avoided.
> + return laio_co_submit(s->fd, 0, NULL, QEMU_AIO_FLUSH, 0);
> + }
> #endif
> return raw_thread_pool_submit(handle_aiocb_flush, &acb);
> }
> diff --git a/block/linux-aio.c b/block/linux-aio.c
> index ec05d946f3..ed20a503e9 100644
> --- a/block/linux-aio.c
> +++ b/block/linux-aio.c
> @@ -67,6 +67,7 @@ struct LinuxAioState {
> int event_max;
> };
>
> +bool laio_has_fdsync(int);
> static void ioq_submit(LinuxAioState *s);
>
> static inline ssize_t io_event_ret(struct io_event *ev)
> @@ -384,6 +385,9 @@ static int laio_do_submit(int fd, struct qemu_laiocb *laiocb, off_t offset,
> case QEMU_AIO_READ:
> io_prep_preadv(iocbs, fd, qiov->iov, qiov->niov, offset);
> break;
> + case QEMU_AIO_FLUSH:
> + io_prep_fdsync(iocbs, fd);
> + break;
> /* Currently Linux kernel does not support other operations */
> default:
> fprintf(stderr, "%s: invalid AIO request type 0x%x.\n",
> @@ -412,7 +416,7 @@ int coroutine_fn laio_co_submit(int fd, uint64_t offset, QEMUIOVector *qiov,
> AioContext *ctx = qemu_get_current_aio_context();
> struct qemu_laiocb laiocb = {
> .co = qemu_coroutine_self(),
> - .nbytes = qiov->size,
> + .nbytes = qiov ? qiov->size : 0,
> .ctx = aio_get_linux_aio(ctx),
> .ret = -EINPROGRESS,
> .is_read = (type == QEMU_AIO_READ),
> @@ -486,3 +490,19 @@ void laio_cleanup(LinuxAioState *s)
> }
> g_free(s);
> }
> +
> +bool laio_has_fdsync(int fd)
> +{
> + struct iocb cb;
> + struct iocb *cbs[] = {&cb, NULL};
> +
> + io_context_t ctx = 0;
> + io_setup(1, &ctx);
> +
> + /* check if host kernel supports IO_CMD_FDSYNC */
> + io_prep_fdsync(&cb, fd);
> + int ret = io_submit(ctx, 1, cbs);
> +
> + io_destroy(ctx);
> + return (ret == -EINVAL) ? false : true;
> +}
> --
> 2.44.0
>
On Wed, 13 Mar 2024 at 20:48, Stefan Hajnoczi <stefanha@redhat.com> wrote: > > +extern bool laio_has_fdsync(int); > Please declare this in include/block/raw-aio.h alongside the other laio APIs. > > FDSYNC support should be probed at open() time and the result should be > stored in a new bool field like s->laio_supports_fdsync. That way the > cost of laio_has_fdsync() on every flush request is avoided. * Okay. Here 's' is a BDRVRawState object and file open seems to happen in the raw_open_common() function? I'll move the laio_has_fdsync() call there and see how it works. Thank you. --- - Prasad
On Wed, Mar 13, 2024 at 10:49:31PM +0530, Prasad Pandit wrote: > On Wed, 13 Mar 2024 at 20:48, Stefan Hajnoczi <stefanha@redhat.com> wrote: > > > +extern bool laio_has_fdsync(int); > > Please declare this in include/block/raw-aio.h alongside the other laio APIs. > > > > FDSYNC support should be probed at open() time and the result should be > > stored in a new bool field like s->laio_supports_fdsync. That way the > > cost of laio_has_fdsync() on every flush request is avoided. > > * Okay. Here 's' is a BDRVRawState object and file open seems to > happen in the raw_open_common() function? I'll move the > laio_has_fdsync() call there and see how it works. Yes. Thanks! Stefan
© 2016 - 2026 Red Hat, Inc.