Skip to contentSkip to conversation

io: follow coroutine AioContext in qio_channel_yield()

PATCH Superseded Stefan Hajnoczi 4 participants 13 messages Aug 29, 2023 – Aug 31, 2023 Block layer

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

v2 cover io: follow coroutine AioContext in qio_channel_yield()

v2:
- Add Patch 1 & 2 to remove unused NBD aio_context arguments and dead code [Fabiano]
- Remove attach/detach prototypes from "io/channel.h" [Daniel]
- Add utility function to set fd handlers [Daniel]

The ongoing QEMU multi-queue block layer effort makes it possible for multiple threads to process I/O in parallel. The nbd block driver is not compatible with the multi-queue block layer yet because QIOChannel cannot be used easily from coroutines running in multiple threads. This series changes the QIOChannel API to make that possible.

Stefan Hajnoczi (4):
nbd: drop unused nbd_receive_negotiate() aio_context argument
nbd: drop unused nbd_start_negotiate() aio_context argument
io: check there are no qio_channel_yield() coroutines during

    ->finalize()

io: follow coroutine AioContext in qio_channel_yield()

 include/block/nbd.h              |   3 +-
 include/io/channel-util.h        |  23 ++++++
 include/io/channel.h             |  69 ++++++++---------
 include/qemu/vhost-user-server.h |   1 +
 block/nbd.c                      |  11 +--
 io/channel-command.c             |  10 ++-
 io/channel-file.c                |   9 ++-
 io/channel-null.c                |   3 +-
 io/channel-socket.c              |   9 ++-
 io/channel-tls.c                 |   6 +-
 io/channel-util.c                |  24 ++++++
 io/channel.c                     | 124 ++++++++++++++++++++++---------
 migration/channel-block.c        |   3 +-
 nbd/client-connection.c          |   3 +-
 nbd/client.c                     |  14 +---
 nbd/server.c                     |  14 +---
 qemu-nbd.c                       |   4 +-
 scsi/qemu-pr-helper.c            |   4 +-
 util/vhost-user-server.c         |  27 +++++--
 19 files changed, 228 insertions(+), 133 deletions(-)
Signature · 1 line
2.41.0
#1 lore raw

PATCH v2 1/4 nbd: drop unused nbd_receive_negotiate() aio_context argument

aio_context is always NULL, so drop it.

---
 include/block/nbd.h     | 3 +--
 nbd/client-connection.c | 3 +--
 nbd/client.c            | 5 ++---
 qemu-nbd.c              | 4 ++--
 4 files changed, 6 insertions(+), 9 deletions(-)
include/block/nbd.h +1 -2
index 4428bcffbb..f672b76173 100644
@@ -324,8 +324,7 @@ typedef struct NBDExportInfo {
     char **contexts;
 } NBDExportInfo;
 
-int nbd_receive_negotiate(AioContext *aio_context, QIOChannel *ioc,
-                          QCryptoTLSCreds *tlscreds,
+int nbd_receive_negotiate(QIOChannel *ioc, QCryptoTLSCreds *tlscreds,
                           const char *hostname, QIOChannel **outioc,
                           NBDExportInfo *info, Error **errp);
 void nbd_free_export_list(NBDExportInfo *info, int count);
nbd/client-connection.c +1 -2
index 3d14296c04..aafb3d0fb4 100644
@@ -146,8 +146,7 @@ static int nbd_connect(QIOChannelSocket *sioc, SocketAddress *addr,
         return 0;
     }
 
-    ret = nbd_receive_negotiate(NULL, QIO_CHANNEL(sioc), tlscreds,
-                                tlshostname,
+    ret = nbd_receive_negotiate(QIO_CHANNEL(sioc), tlscreds, tlshostname,
                                 outioc, info, errp);
     if (ret < 0) {
         /*
nbd/client.c +2 -3
index 479208d5d9..16ec10c8a9 100644
@@ -1014,8 +1014,7 @@ static int nbd_negotiate_finish_oldstyle(QIOChannel *ioc, NBDExportInfo *info,
  * Returns: negative errno: failure talking to server
  *          0: server is connected
  */
-int nbd_receive_negotiate(AioContext *aio_context, QIOChannel *ioc,
-                          QCryptoTLSCreds *tlscreds,
+int nbd_receive_negotiate(QIOChannel *ioc, QCryptoTLSCreds *tlscreds,
                           const char *hostname, QIOChannel **outioc,
                           NBDExportInfo *info, Error **errp)
 {
@@ -1027,7 +1026,7 @@ int nbd_receive_negotiate(AioContext *aio_context, QIOChannel *ioc,
     assert(info->name && strlen(info->name) <= NBD_MAX_STRING_SIZE);
     trace_nbd_receive_negotiate_name(info->name);
 
-    result = nbd_start_negotiate(aio_context, ioc, tlscreds, hostname, outioc,
+    result = nbd_start_negotiate(NULL, ioc, tlscreds, hostname, outioc,
                                  info->structured_reply, &zeroes, errp);
     if (result < 0) {
         return result;
qemu-nbd.c +2 -2
index aaccaa3318..b47459f781 100644
@@ -295,8 +295,8 @@ static void *nbd_client_thread(void *arg)
         goto out;
     }
 
-    if (nbd_receive_negotiate(NULL, QIO_CHANNEL(sioc),
-                              NULL, NULL, NULL, &info, &local_error) < 0) {
+    if (nbd_receive_negotiate(QIO_CHANNEL(sioc), NULL, NULL, NULL,
+                              &info, &local_error) < 0) {
         if (local_error) {
             error_report_err(local_error);
         }
Signature · 1 line
2.41.0
Reviewed-by on v2 1/4
#2 lore raw

Re: [PATCH v2 1/4] nbd: drop unused nbd_receive_negotiate() aio_context argument

Quoted 11 lines · Stefan Hajnoczi

aio_context is always NULL, so drop it.

---
 include/block/nbd.h     | 3 +--
 nbd/client-connection.c | 3 +--
 nbd/client.c            | 5 ++---
 qemu-nbd.c              | 4 ++--
 4 files changed, 6 insertions(+), 9 deletions(-)
Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.
Virtualization:  qemu.org | libguestfs.org
Reviewed-by on v2 1/4
#3 lore raw

Re: [PATCH v2 1/4] nbd: drop unused nbd_receive_negotiate() aio_context argument

In reply to Stefan Hajnoczi
Quoted 10 lines · Stefan Hajnoczi

aio_context is always NULL, so drop it.

---
  include/block/nbd.h     | 3 +--
  nbd/client-connection.c | 3 +--
  nbd/client.c            | 5 ++---
  qemu-nbd.c              | 4 ++--
  4 files changed, 6 insertions(+), 9 deletions(-)
#4 lore raw

PATCH v2 2/4 nbd: drop unused nbd_start_negotiate() aio_context argument

In reply to Stefan Hajnoczi

aio_context is always NULL, so drop it.

---
 nbd/client.c | 11 +++--------
 1 file changed, 3 insertions(+), 8 deletions(-)
nbd/client.c +3 -8
index 16ec10c8a9..bd7e200136 100644
@@ -877,8 +877,7 @@ static int nbd_list_meta_contexts(QIOChannel *ioc,
  * Returns: negative errno: failure talking to server
  *          non-negative: enum NBDMode describing server abilities
  */
-static int nbd_start_negotiate(AioContext *aio_context, QIOChannel *ioc,
-                               QCryptoTLSCreds *tlscreds,
+static int nbd_start_negotiate(QIOChannel *ioc, QCryptoTLSCreds *tlscreds,
                                const char *hostname, QIOChannel **outioc,
                                bool structured_reply, bool *zeroes,
                                Error **errp)
@@ -946,10 +945,6 @@ static int nbd_start_negotiate(AioContext *aio_context, QIOChannel *ioc,
                     return -EINVAL;
                 }
                 ioc = *outioc;
-                if (aio_context) {
-                    qio_channel_set_blocking(ioc, false, NULL);
-                    qio_channel_attach_aio_context(ioc, aio_context);
-                }
             } else {
                 error_setg(errp, "Server does not support STARTTLS");
                 return -EINVAL;
@@ -1026,7 +1021,7 @@ int nbd_receive_negotiate(QIOChannel *ioc, QCryptoTLSCreds *tlscreds,
     assert(info->name && strlen(info->name) <= NBD_MAX_STRING_SIZE);
     trace_nbd_receive_negotiate_name(info->name);
 
-    result = nbd_start_negotiate(NULL, ioc, tlscreds, hostname, outioc,
+    result = nbd_start_negotiate(ioc, tlscreds, hostname, outioc,
                                  info->structured_reply, &zeroes, errp);
     if (result < 0) {
         return result;
@@ -1149,7 +1144,7 @@ int nbd_receive_export_list(QIOChannel *ioc, QCryptoTLSCreds *tlscreds,
     QIOChannel *sioc = NULL;
 
     *info = NULL;
-    result = nbd_start_negotiate(NULL, ioc, tlscreds, hostname, &sioc, true,
+    result = nbd_start_negotiate(ioc, tlscreds, hostname, &sioc, true,
                                  NULL, errp);
     if (tlscreds && sioc) {
         ioc = sioc;
Signature · 1 line
2.41.0
Reviewed-by on v2 2/4
#5 lore raw

Re: [PATCH v2 2/4] nbd: drop unused nbd_start_negotiate() aio_context argument

Quoted 7 lines · Stefan Hajnoczi

aio_context is always NULL, so drop it.

---
 nbd/client.c | 11 +++--------
 1 file changed, 3 insertions(+), 8 deletions(-)
Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.
Virtualization:  qemu.org | libguestfs.org
#7 lore raw

PATCH v2 3/4 io: check there are no qio_channel_yield() coroutines during ->finalize()

In reply to Stefan Hajnoczi

Callers must clean up their coroutines before calling object_unref(OBJECT(ioc)) to prevent an fd handler leak. Add an assertion to check this.

This patch is preparation for the fd handler changes that follow.

---
 io/channel.c | 4 ++++
 1 file changed, 4 insertions(+)
io/channel.c +4 -0
index 72f0066af5..c415f3fc88 100644
@@ -653,6 +653,10 @@ static void qio_channel_finalize(Object *obj)
 {
     QIOChannel *ioc = QIO_CHANNEL(obj);
 
+    /* Must not have coroutines in qio_channel_yield() */
+    assert(!ioc->read_coroutine);
+    assert(!ioc->write_coroutine);
+
     g_free(ioc->name);
 
 #ifdef _WIN32
Signature · 1 line
2.41.0
#8 lore raw

PATCH v2 4/4 io: follow coroutine AioContext in qio_channel_yield()

In reply to Stefan Hajnoczi

The ongoing QEMU multi-queue block layer effort makes it possible for multiple threads to process I/O in parallel. The nbd block driver is not compatible with the multi-queue block layer yet because QIOChannel cannot be used easily from coroutines running in multiple threads. This series changes the QIOChannel API to make that possible.

In the current API, calling qio_channel_attach_aio_context() sets the AioContext where qio_channel_yield() installs an fd handler prior to yielding:

  qio_channel_attach_aio_context(ioc, my_ctx);
  ...
  qio_channel_yield(ioc); // my_ctx is used here
  ...
  qio_channel_detach_aio_context(ioc);

This API design has limitations: reading and writing must be done in the same AioContext and moving between AioContexts involves a cumbersome sequence of API calls that is not suitable for doing on a per-request basis.

There is no fundamental reason why a QIOChannel needs to run within the same AioContext every time qio_channel_yield() is called. QIOChannel only uses the AioContext while inside qio_channel_yield(). The rest of the time, QIOChannel is independent of any AioContext.

In the new API, qio_channel_yield() queries the AioContext from the current coroutine using qemu_coroutine_get_aio_context(). There is no need to explicitly attach/detach AioContexts anymore and qio_channel_attach_aio_context() and qio_channel_detach_aio_context() are gone. One coroutine can read from the QIOChannel while another coroutine writes from a different AioContext.

This API change allows the nbd block driver to use QIOChannel from any thread. It's important to keep in mind that the block driver already synchronizes QIOChannel access and ensures that two coroutines never read simultaneously or write simultaneously.

This patch updates all users of qio_channel_attach_aio_context() to the new API. Most conversions are simple, but vhost-user-server requires a new qemu_coroutine_yield() call to quiesce the vu_client_trip() coroutine when not attached to any AioContext.

While the API is has become simpler, there is one wart: QIOChannel has a special case for the iohandler AioContext (used for handlers that must not run in nested event loops). I didn't find an elegant way preserve that behavior, so I added a new API called qio_channel_set_follow_coroutine_ctx(ioc, true|false) for opting in to the new AioContext model. By default QIOChannel uses the iohandler AioHandler. Code that formerly called qio_channel_attach_aio_context() now calls qio_channel_set_follow_coroutine_ctx(ioc, true) once after the QIOChannel is created.

---
 include/io/channel-util.h        |  23 ++++++
 include/io/channel.h             |  69 ++++++++----------
 include/qemu/vhost-user-server.h |   1 +
 block/nbd.c                      |  11 +--
 io/channel-command.c             |  10 ++-
 io/channel-file.c                |   9 ++-
 io/channel-null.c                |   3 +-
 io/channel-socket.c              |   9 ++-
 io/channel-tls.c                 |   6 +-
 io/channel-util.c                |  24 +++++++
 io/channel.c                     | 120 ++++++++++++++++++++++---------
 migration/channel-block.c        |   3 +-
 nbd/server.c                     |  14 +---
 scsi/qemu-pr-helper.c            |   4 +-
 util/vhost-user-server.c         |  27 +++++--
 15 files changed, 216 insertions(+), 117 deletions(-)
include/io/channel-util.h +23 -0
index a5d720d9a0..74ed299fad 100644
@@ -49,4 +49,27 @@
 QIOChannel *qio_channel_new_fd(int fd,
                                Error **errp);
 
+/**
+ * qio_channel_yield:
+ * @read_fd: the file descriptor for the read handler
+ * @read_ctx: the AioContext for the read handler
+ * @io_read: the read handler
+ * @write_fd: the file descriptor for the write handler
+ * @write_ctx: the AioContext for the write handler
+ * @io_write: the write handler
+ * @opaque: the opaque argument to the read and write handler
+ *
+ * Set the read and write handlers when @read_ctx and @write_ctx are non-NULL,
+ * respectively. To leave a handler in its current state, pass a NULL
+ * AioContext. To clear a handler, pass a non-NULL AioContext and a NULL
+ * handler.
+ */
+void qio_channel_util_set_aio_fd_handler(int read_fd,
+                                         AioContext *read_ctx,
+                                         IOHandler *io_read,
+                                         int write_fd,
+                                         AioContext *write_ctx,
+                                         IOHandler *io_write,
+                                         void *opaque);
+
 #endif /* QIO_CHANNEL_UTIL_H */
include/io/channel.h +30 -39

112 lines not on this page · raw message

include/qemu/vhost-user-server.h +1 -0
index b1c1cda886..64ad701015 100644
@@ -43,6 +43,7 @@ typedef struct {
     unsigned int in_flight; /* atomic */
 
     /* Protected by ctx lock */
+    bool in_qio_channel_yield;
     bool wait_idle;
     VuDev vu_dev;
     QIOChannel *ioc; /* The I/O channel with the client */
block/nbd.c +1 -10
index 5322e66166..cc48580df7 100644
@@ -352,7 +352,7 @@ int coroutine_fn nbd_co_do_establish_connection(BlockDriverState *bs,
     }
 
     qio_channel_set_blocking(s->ioc, false, NULL);
-    qio_channel_attach_aio_context(s->ioc, bdrv_get_aio_context(bs));
+    qio_channel_set_follow_coroutine_ctx(s->ioc, true);
 
     /* successfully connected */
     WITH_QEMU_LOCK_GUARD(&s->requests_lock) {
@@ -397,7 +397,6 @@ static void coroutine_fn GRAPH_RDLOCK nbd_reconnect_attempt(BDRVNBDState *s)
 
     /* Finalize previous connection if any */
     if (s->ioc) {
-        qio_channel_detach_aio_context(s->ioc);
         yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name),
                                  nbd_yank, s->bs);
         object_unref(OBJECT(s->ioc));
@@ -2089,10 +2088,6 @@ static void nbd_attach_aio_context(BlockDriverState *bs,
      * the reconnect_delay_timer cannot be active here.
      */
     assert(!s->reconnect_delay_timer);
-
-    if (s->ioc) {
-        qio_channel_attach_aio_context(s->ioc, new_context);
-    }
 }
 
 static void nbd_detach_aio_context(BlockDriverState *bs)
@@ -2101,10 +2096,6 @@ static void nbd_detach_aio_context(BlockDriverState *bs)
 
     assert(!s->open_timer);
     assert(!s->reconnect_delay_timer);
-
-    if (s->ioc) {
-        qio_channel_detach_aio_context(s->ioc);
-    }
 }
 
 static BlockDriver bdrv_nbd = {
io/channel-command.c +7 -3
index 7ed726c802..6d5f64e146 100644
@@ -20,6 +20,7 @@
 
 #include "qemu/osdep.h"
 #include "io/channel-command.h"
+#include "io/channel-util.h"
 #include "io/channel-watch.h"
 #include "qapi/error.h"
 #include "qemu/module.h"
@@ -331,14 +332,17 @@ static int qio_channel_command_close(QIOChannel *ioc,
 
 
 static void qio_channel_command_set_aio_fd_handler(QIOChannel *ioc,
-                                                   AioContext *ctx,
+                                                   AioContext *read_ctx,
                                                    IOHandler *io_read,
+                                                   AioContext *write_ctx,
                                                    IOHandler *io_write,
                                                    void *opaque)
 {
     QIOChannelCommand *cioc = QIO_CHANNEL_COMMAND(ioc);
-    aio_set_fd_handler(ctx, cioc->readfd, io_read, NULL, NULL, NULL, opaque);
-    aio_set_fd_handler(ctx, cioc->writefd, NULL, io_write, NULL, NULL, opaque);
+
+    qio_channel_util_set_aio_fd_handler(cioc->readfd, read_ctx, io_read,
+                                        cioc->writefd, write_ctx, io_write,
+                                        opaque);
 }
 
 
io/channel-file.c +7 -2
index 8b5821f452..4a12c61886 100644
@@ -20,6 +20,7 @@
 
 #include "qemu/osdep.h"
 #include "io/channel-file.h"
+#include "io/channel-util.h"
 #include "io/channel-watch.h"
 #include "qapi/error.h"
 #include "qemu/module.h"
@@ -192,13 +193,17 @@ static int qio_channel_file_close(QIOChannel *ioc,
 
 
 static void qio_channel_file_set_aio_fd_handler(QIOChannel *ioc,
-                                                AioContext *ctx,
+                                                AioContext *read_ctx,
                                                 IOHandler *io_read,
+                                                AioContext *write_ctx,
                                                 IOHandler *io_write,
                                                 void *opaque)
 {
     QIOChannelFile *fioc = QIO_CHANNEL_FILE(ioc);
-    aio_set_fd_handler(ctx, fioc->fd, io_read, io_write, NULL, NULL, opaque);
+
+    qio_channel_util_set_aio_fd_handler(fioc->fd, read_ctx, io_read,
+                                        fioc->fd, write_ctx, io_write,
+                                        opaque);
 }
 
 static GSource *qio_channel_file_create_watch(QIOChannel *ioc,
io/channel-null.c +2 -1
index 4fafdb770d..ef99586348 100644
@@ -128,8 +128,9 @@ qio_channel_null_close(QIOChannel *ioc,
 
 static void
 qio_channel_null_set_aio_fd_handler(QIOChannel *ioc G_GNUC_UNUSED,
-                                    AioContext *ctx G_GNUC_UNUSED,
+                                    AioContext *read_ctx G_GNUC_UNUSED,
                                     IOHandler *io_read G_GNUC_UNUSED,
+                                    AioContext *write_ctx G_GNUC_UNUSED,
                                     IOHandler *io_write G_GNUC_UNUSED,
                                     void *opaque G_GNUC_UNUSED)
 {
io/channel-socket.c +7 -2
index d99945ebec..02ffb51e99 100644
@@ -22,6 +22,7 @@
 #include "qapi/qapi-visit-sockets.h"
 #include "qemu/module.h"
 #include "io/channel-socket.h"
+#include "io/channel-util.h"
 #include "io/channel-watch.h"
 #include "trace.h"
 #include "qapi/clone-visitor.h"
@@ -893,13 +894,17 @@ qio_channel_socket_shutdown(QIOChannel *ioc,
 }
 
 static void qio_channel_socket_set_aio_fd_handler(QIOChannel *ioc,
-                                                  AioContext *ctx,
+                                                  AioContext *read_ctx,
                                                   IOHandler *io_read,
+                                                  AioContext *write_ctx,
                                                   IOHandler *io_write,
                                                   void *opaque)
 {
     QIOChannelSocket *sioc = QIO_CHANNEL_SOCKET(ioc);
-    aio_set_fd_handler(ctx, sioc->fd, io_read, io_write, NULL, NULL, opaque);
+
+    qio_channel_util_set_aio_fd_handler(sioc->fd, read_ctx, io_read,
+                                        sioc->fd, write_ctx, io_write,
+                                        opaque);
 }
 
 static GSource *qio_channel_socket_create_watch(QIOChannel *ioc,
io/channel-tls.c +4 -2
index 847d5297c3..58fe1aceee 100644
@@ -388,14 +388,16 @@ static int qio_channel_tls_close(QIOChannel *ioc,
 }
 
 static void qio_channel_tls_set_aio_fd_handler(QIOChannel *ioc,
-                                               AioContext *ctx,
+                                               AioContext *read_ctx,
                                                IOHandler *io_read,
+                                               AioContext *write_ctx,
                                                IOHandler *io_write,
                                                void *opaque)
 {
     QIOChannelTLS *tioc = QIO_CHANNEL_TLS(ioc);
 
-    qio_channel_set_aio_fd_handler(tioc->master, ctx, io_read, io_write, opaque);
+    qio_channel_set_aio_fd_handler(tioc->master, read_ctx, io_read,
+            write_ctx, io_write, opaque);
 }
 
 typedef struct QIOChannelTLSSource QIOChannelTLSSource;
io/channel-util.c +24 -0
index 848a7a43d6..4b340d46d7 100644
@@ -36,3 +36,27 @@ QIOChannel *qio_channel_new_fd(int fd,
     }
     return ioc;
 }
+
+
+void qio_channel_util_set_aio_fd_handler(int read_fd,
+                                         AioContext *read_ctx,
+                                         IOHandler *io_read,
+                                         int write_fd,
+                                         AioContext *write_ctx,
+                                         IOHandler *io_write,
+                                         void *opaque)
+{
+    if (read_fd == write_fd && read_ctx == write_ctx) {
+        aio_set_fd_handler(read_ctx, read_fd, io_read, io_write,
+                NULL, NULL, opaque);
+    } else {
+        if (read_ctx) {
+            aio_set_fd_handler(read_ctx, read_fd, io_read, NULL,
+                    NULL, NULL, opaque);
+        }
+        if (write_ctx) {
+            aio_set_fd_handler(write_ctx, write_fd, NULL, io_write,
+                    NULL, NULL, opaque);
+        }
+    }
+}
io/channel.c +86 -34

173 lines not on this page · raw message

migration/channel-block.c +2 -1
index b7374363c3..fff8d87094 100644
@@ -158,8 +158,9 @@ qio_channel_block_close(QIOChannel *ioc,
 
 static void
 qio_channel_block_set_aio_fd_handler(QIOChannel *ioc,
-                                     AioContext *ctx,
+                                     AioContext *read_ctx,
                                      IOHandler *io_read,
+                                     AioContext *write_ctx,
                                      IOHandler *io_write,
                                      void *opaque)
 {
nbd/server.c +1 -13
index 8486b64b15..b5f93a20c9 100644
@@ -1333,6 +1333,7 @@ static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
      */
 
     qio_channel_set_blocking(client->ioc, false, NULL);
+    qio_channel_set_follow_coroutine_ctx(client->ioc, true);
 
     trace_nbd_negotiate_begin();
     memcpy(buf, "NBDMAGIC", 8);
@@ -1352,11 +1353,6 @@ static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
         return ret;
     }
 
-    /* Attach the channel to the same AioContext as the export */
-    if (client->exp && client->exp->common.ctx) {
-        qio_channel_attach_aio_context(client->ioc, client->exp->common.ctx);
-    }
-
     assert(!client->optlen);
     trace_nbd_negotiate_success();
 
@@ -1465,7 +1461,6 @@ void nbd_client_put(NBDClient *client)
          */
         assert(client->closing);
 
-        qio_channel_detach_aio_context(client->ioc);
         object_unref(OBJECT(client->sioc));
         object_unref(OBJECT(client->ioc));
         if (client->tlscreds) {
@@ -1544,8 +1539,6 @@ static void blk_aio_attached(AioContext *ctx, void *opaque)
     exp->common.ctx = ctx;
 
     QTAILQ_FOREACH(client, &exp->clients, next) {
-        qio_channel_attach_aio_context(client->ioc, ctx);
-
         assert(client->nb_requests == 0);
         assert(client->recv_coroutine == NULL);
         assert(client->send_coroutine == NULL);
@@ -1555,14 +1548,9 @@ static void blk_aio_attached(AioContext *ctx, void *opaque)
 static void blk_aio_detach(void *opaque)
 {
     NBDExport *exp = opaque;
-    NBDClient *client;
 
     trace_nbd_blk_aio_detach(exp->name, exp->common.ctx);
 
-    QTAILQ_FOREACH(client, &exp->clients, next) {
-        qio_channel_detach_aio_context(client->ioc);
-    }
-
     exp->common.ctx = NULL;
 }
 
scsi/qemu-pr-helper.c +1 -3
index ae44a816e1..c6c6347e9b 100644
@@ -735,8 +735,7 @@ static void coroutine_fn prh_co_entry(void *opaque)
 
     qio_channel_set_blocking(QIO_CHANNEL(client->ioc),
                              false, NULL);
-    qio_channel_attach_aio_context(QIO_CHANNEL(client->ioc),
-                                   qemu_get_aio_context());
+    qio_channel_set_follow_coroutine_ctx(QIO_CHANNEL(client->ioc), true);
 
     /* A very simple negotiation for future extensibility.  No features
      * are defined so write 0.
@@ -796,7 +795,6 @@ static void coroutine_fn prh_co_entry(void *opaque)
     }
 
 out:
-    qio_channel_detach_aio_context(QIO_CHANNEL(client->ioc));
     object_unref(OBJECT(client->ioc));
     g_free(client);
 }
util/vhost-user-server.c +20 -7
index cd17fb5326..b4b6bf30a2 100644
@@ -127,7 +127,14 @@ vu_message_read(VuDev *vu_dev, int conn_fd, VhostUserMsg *vmsg)
         if (rc < 0) {
             if (rc == QIO_CHANNEL_ERR_BLOCK) {
                 assert(local_err == NULL);
-                qio_channel_yield(ioc, G_IO_IN);
+                if (server->ctx) {
+                    server->in_qio_channel_yield = true;
+                    qio_channel_yield(ioc, G_IO_IN);
+                    server->in_qio_channel_yield = false;
+                } else {
+                    /* Wait until attached to an AioContext again */
+                    qemu_coroutine_yield();
+                }
                 continue;
             } else {
                 error_report_err(local_err);
@@ -278,7 +285,7 @@ set_watch(VuDev *vu_dev, int fd, int vu_evt,
         vu_fd_watch->fd = fd;
         vu_fd_watch->cb = cb;
         qemu_socket_set_nonblock(fd);
-        aio_set_fd_handler(server->ioc->ctx, fd, kick_handler,
+        aio_set_fd_handler(server->ctx, fd, kick_handler,
                            NULL, NULL, NULL, vu_fd_watch);
         vu_fd_watch->vu_dev = vu_dev;
         vu_fd_watch->pvt = pvt;
@@ -299,7 +306,7 @@ static void remove_watch(VuDev *vu_dev, int fd)
     if (!vu_fd_watch) {
         return;
     }
-    aio_set_fd_handler(server->ioc->ctx, fd, NULL, NULL, NULL, NULL, NULL);
+    aio_set_fd_handler(server->ctx, fd, NULL, NULL, NULL, NULL, NULL);
 
     QTAILQ_REMOVE(&server->vu_fd_watches, vu_fd_watch, next);
     g_free(vu_fd_watch);
@@ -344,6 +351,8 @@ static void vu_accept(QIONetListener *listener, QIOChannelSocket *sioc,
     /* TODO vu_message_write() spins if non-blocking! */
     qio_channel_set_blocking(server->ioc, false, NULL);
 
+    qio_channel_set_follow_coroutine_ctx(server->ioc, true);
+
     server->co_trip = qemu_coroutine_create(vu_client_trip, server);
 
     aio_context_acquire(server->ctx);
@@ -399,13 +408,12 @@ void vhost_user_server_attach_aio_context(VuServer *server, AioContext *ctx)
         return;
     }
 
-    qio_channel_attach_aio_context(server->ioc, ctx);
-
     QTAILQ_FOREACH(vu_fd_watch, &server->vu_fd_watches, next) {
         aio_set_fd_handler(ctx, vu_fd_watch->fd, kick_handler, NULL,
                            NULL, NULL, vu_fd_watch);
     }
 
+    assert(!server->in_qio_channel_yield);
     aio_co_schedule(ctx, server->co_trip);
 }
 
@@ -419,11 +427,16 @@ void vhost_user_server_detach_aio_context(VuServer *server)
             aio_set_fd_handler(server->ctx, vu_fd_watch->fd,
                                NULL, NULL, NULL, NULL, vu_fd_watch);
         }
-
-        qio_channel_detach_aio_context(server->ioc);
     }
 
     server->ctx = NULL;
+
+    if (server->ioc) {
+        if (server->in_qio_channel_yield) {
+            /* Stop receiving the next vhost-user message */
+            qio_channel_wake_read(server->ioc);
+        }
+    }
 }
 
 bool vhost_user_server_start(VuServer *server,
Signature · 1 line
2.41.0
Reviewed-by on v2 4/4
#9 lore raw
Quoted 55 lines · Stefan Hajnoczi

The ongoing QEMU multi-queue block layer effort makes it possible for multiple threads to process I/O in parallel. The nbd block driver is not compatible with the multi-queue block layer yet because QIOChannel cannot be used easily from coroutines running in multiple threads. This series changes the QIOChannel API to make that possible.

In the current API, calling qio_channel_attach_aio_context() sets the AioContext where qio_channel_yield() installs an fd handler prior to yielding:

  qio_channel_attach_aio_context(ioc, my_ctx);
  ...
  qio_channel_yield(ioc); // my_ctx is used here
  ...
  qio_channel_detach_aio_context(ioc);

This API design has limitations: reading and writing must be done in the same AioContext and moving between AioContexts involves a cumbersome sequence of API calls that is not suitable for doing on a per-request basis.

There is no fundamental reason why a QIOChannel needs to run within the same AioContext every time qio_channel_yield() is called. QIOChannel only uses the AioContext while inside qio_channel_yield(). The rest of the time, QIOChannel is independent of any AioContext.

In the new API, qio_channel_yield() queries the AioContext from the current coroutine using qemu_coroutine_get_aio_context(). There is no need to explicitly attach/detach AioContexts anymore and qio_channel_attach_aio_context() and qio_channel_detach_aio_context() are gone. One coroutine can read from the QIOChannel while another coroutine writes from a different AioContext.

This API change allows the nbd block driver to use QIOChannel from any thread. It's important to keep in mind that the block driver already synchronizes QIOChannel access and ensures that two coroutines never read simultaneously or write simultaneously.

This patch updates all users of qio_channel_attach_aio_context() to the new API. Most conversions are simple, but vhost-user-server requires a new qemu_coroutine_yield() call to quiesce the vu_client_trip() coroutine when not attached to any AioContext.

While the API is has become simpler, there is one wart: QIOChannel has a special case for the iohandler AioContext (used for handlers that must not run in nested event loops). I didn't find an elegant way preserve that behavior, so I added a new API called qio_channel_set_follow_coroutine_ctx(ioc, true|false) for opting in to the new AioContext model. By default QIOChannel uses the iohandler AioHandler. Code that formerly called qio_channel_attach_aio_context() now calls qio_channel_set_follow_coroutine_ctx(ioc, true) once after the QIOChannel is created.

---
+++ b/include/io/channel-util.h

include/io/channel-util.h line 53

Quoted 5 lines
include/io/channel-util.h +2 -0
@@ -49,4 +49,27 @@
 QIOChannel *qio_channel_new_fd(int fd,
                                Error **errp);
 
+/**
+ * qio_channel_yield:

Wrong function name...

Quoted 4 lines
include/io/channel-util.h +4 -0
+ * @read_fd: the file descriptor for the read handler
+ * @read_ctx: the AioContext for the read handler
+ * @io_read: the read handler
+ * @write_fd: the file descriptor for the write handler

include/io/channel-util.h line 67

Quoted 10 lines
include/io/channel-util.h +10 -0
+ * @write_ctx: the AioContext for the write handler
+ * @io_write: the write handler
+ * @opaque: the opaque argument to the read and write handler
+ *
+ * Set the read and write handlers when @read_ctx and @write_ctx are non-NULL,
+ * respectively. To leave a handler in its current state, pass a NULL
+ * AioContext. To clear a handler, pass a non-NULL AioContext and a NULL
+ * handler.
+ */
+void qio_channel_util_set_aio_fd_handler(int read_fd,

...should be this.

Quoted 27 lines
include/io/channel-util.h +7 -0
+                                         AioContext *read_ctx,
+                                         IOHandler *io_read,
+                                         int write_fd,
+                                         AioContext *write_ctx,
+                                         IOHandler *io_write,
+                                         void *opaque);
+
 #endif /* QIO_CHANNEL_UTIL_H */
include/io/channel.h +10 -3
index 229bf36910..5f9dbaab65 100644
--- a/include/io/channel.h
 
+++ b/io/channel.c
 
 
-static void qio_channel_set_aio_fd_handlers(QIOChannel *ioc)
+static void coroutine_fn
+qio_channel_set_fd_handlers(QIOChannel *ioc, GIOCondition condition)
 {
-    IOHandler *rd_handler = NULL, *wr_handler = NULL;
+    AioContext *ctx = ioc->follow_coroutine_ctx ?
+        qemu_coroutine_get_aio_context(qemu_coroutine_self()) :
+        iohandler_get_aio_context();
+    AioContext *read_ctx = NULL;
+    IOHandler *io_read = NULL;
+    AioContext *write_ctx = NULL;
+    IOHandler *io_write = NULL;

include/io/channel.h

Quoted 10 lines
include/io/channel.h +10 -0
+
+    if (condition == G_IO_IN) {
+        ioc->read_coroutine = qemu_coroutine_self();
+        ioc->read_ctx = ctx;
+        read_ctx = ctx;
+        io_read = qio_channel_restart_read;
+
+        /*
+         * Thread safety: if the other coroutine is set and its AioContext
+         * match ours, then there is mutual exclusion between read and write

matches

Quoted 16 lines
include/io/channel.h +16 -0
+         * because they share a single thread and it's safe to set both read
+         * and write fd handlers here. If the AioContext does not match ours,
+         * then both threads may run in parallel but there is no shared state
+         * to worry about.
+         */
+        if (ioc->write_coroutine && ioc->write_ctx == ctx) {
+            write_ctx = ctx;
+            io_write = qio_channel_restart_write;
+        }
+    } else if (condition == G_IO_OUT) {
+        ioc->write_coroutine = qemu_coroutine_self();
+        ioc->write_ctx = ctx;
+        write_ctx = ctx;
+        io_write = qio_channel_restart_write;
+        if (ioc->read_coroutine && ioc->read_ctx == ctx) {
+            read_ctx = ctx;

include/io/channel.h

Quoted 10 lines
include/io/channel.h +10 -0
+            io_read = qio_channel_restart_read;
+        }
+    } else {
+        abort();
+    }
+
+    qio_channel_set_aio_fd_handler(ioc, read_ctx, io_read,
+            write_ctx, io_write, ioc);
+}
+

With those minor fixes,

Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.
Virtualization:  qemu.org | libguestfs.org
#10 lore raw
Quoted 141 lines · Eric Blake
Quoted 55 lines · Stefan Hajnoczi

The ongoing QEMU multi-queue block layer effort makes it possible for multiple threads to process I/O in parallel. The nbd block driver is not compatible with the multi-queue block layer yet because QIOChannel cannot be used easily from coroutines running in multiple threads. This series changes the QIOChannel API to make that possible.

In the current API, calling qio_channel_attach_aio_context() sets the AioContext where qio_channel_yield() installs an fd handler prior to yielding:

  qio_channel_attach_aio_context(ioc, my_ctx);
  ...
  qio_channel_yield(ioc); // my_ctx is used here
  ...
  qio_channel_detach_aio_context(ioc);

This API design has limitations: reading and writing must be done in the same AioContext and moving between AioContexts involves a cumbersome sequence of API calls that is not suitable for doing on a per-request basis.

There is no fundamental reason why a QIOChannel needs to run within the same AioContext every time qio_channel_yield() is called. QIOChannel only uses the AioContext while inside qio_channel_yield(). The rest of the time, QIOChannel is independent of any AioContext.

In the new API, qio_channel_yield() queries the AioContext from the current coroutine using qemu_coroutine_get_aio_context(). There is no need to explicitly attach/detach AioContexts anymore and qio_channel_attach_aio_context() and qio_channel_detach_aio_context() are gone. One coroutine can read from the QIOChannel while another coroutine writes from a different AioContext.

This API change allows the nbd block driver to use QIOChannel from any thread. It's important to keep in mind that the block driver already synchronizes QIOChannel access and ensures that two coroutines never read simultaneously or write simultaneously.

This patch updates all users of qio_channel_attach_aio_context() to the new API. Most conversions are simple, but vhost-user-server requires a new qemu_coroutine_yield() call to quiesce the vu_client_trip() coroutine when not attached to any AioContext.

While the API is has become simpler, there is one wart: QIOChannel has a special case for the iohandler AioContext (used for handlers that must not run in nested event loops). I didn't find an elegant way preserve that behavior, so I added a new API called qio_channel_set_follow_coroutine_ctx(ioc, true|false) for opting in to the new AioContext model. By default QIOChannel uses the iohandler AioHandler. Code that formerly called qio_channel_attach_aio_context() now calls qio_channel_set_follow_coroutine_ctx(ioc, true) once after the QIOChannel is created.

---
+++ b/include/io/channel-util.h

include/io/channel-util.h line 53

Quoted 5 lines
include/io/channel-util.h +2 -0
@@ -49,4 +49,27 @@
 QIOChannel *qio_channel_new_fd(int fd,
                                Error **errp);
 
+/**
+ * qio_channel_yield:

Wrong function name...

Quoted 4 lines
include/io/channel-util.h +4 -0
+ * @read_fd: the file descriptor for the read handler
+ * @read_ctx: the AioContext for the read handler
+ * @io_read: the read handler
+ * @write_fd: the file descriptor for the write handler

include/io/channel-util.h line 67

Quoted 10 lines
include/io/channel-util.h +10 -0
+ * @write_ctx: the AioContext for the write handler
+ * @io_write: the write handler
+ * @opaque: the opaque argument to the read and write handler
+ *
+ * Set the read and write handlers when @read_ctx and @write_ctx are non-NULL,
+ * respectively. To leave a handler in its current state, pass a NULL
+ * AioContext. To clear a handler, pass a non-NULL AioContext and a NULL
+ * handler.
+ */
+void qio_channel_util_set_aio_fd_handler(int read_fd,

...should be this.

Quoted 27 lines
include/io/channel-util.h +7 -0
+                                         AioContext *read_ctx,
+                                         IOHandler *io_read,
+                                         int write_fd,
+                                         AioContext *write_ctx,
+                                         IOHandler *io_write,
+                                         void *opaque);
+
 #endif /* QIO_CHANNEL_UTIL_H */
include/io/channel.h +10 -3
index 229bf36910..5f9dbaab65 100644
--- a/include/io/channel.h
 
+++ b/io/channel.c
 
 
-static void qio_channel_set_aio_fd_handlers(QIOChannel *ioc)
+static void coroutine_fn
+qio_channel_set_fd_handlers(QIOChannel *ioc, GIOCondition condition)
 {
-    IOHandler *rd_handler = NULL, *wr_handler = NULL;
+    AioContext *ctx = ioc->follow_coroutine_ctx ?
+        qemu_coroutine_get_aio_context(qemu_coroutine_self()) :
+        iohandler_get_aio_context();
+    AioContext *read_ctx = NULL;
+    IOHandler *io_read = NULL;
+    AioContext *write_ctx = NULL;
+    IOHandler *io_write = NULL;

include/io/channel.h

Quoted 10 lines
include/io/channel.h +10 -0
+
+    if (condition == G_IO_IN) {
+        ioc->read_coroutine = qemu_coroutine_self();
+        ioc->read_ctx = ctx;
+        read_ctx = ctx;
+        io_read = qio_channel_restart_read;
+
+        /*
+         * Thread safety: if the other coroutine is set and its AioContext
+         * match ours, then there is mutual exclusion between read and write

matches

Quoted 20 lines
include/io/channel.h +20 -0
+         * because they share a single thread and it's safe to set both read
+         * and write fd handlers here. If the AioContext does not match ours,
+         * then both threads may run in parallel but there is no shared state
+         * to worry about.
+         */
+        if (ioc->write_coroutine && ioc->write_ctx == ctx) {
+            write_ctx = ctx;
+            io_write = qio_channel_restart_write;
+        }
+    } else if (condition == G_IO_OUT) {
+        ioc->write_coroutine = qemu_coroutine_self();
+        ioc->write_ctx = ctx;
+        write_ctx = ctx;
+        io_write = qio_channel_restart_write;
+        if (ioc->read_coroutine && ioc->read_ctx == ctx) {
+            read_ctx = ctx;
+            io_read = qio_channel_restart_read;
+        }
+    } else {
+        abort();
Quoted 8 lines

include/io/channel.h

Quoted 6 lines
include/io/channel.h +6 -0
+    }
+
+    qio_channel_set_aio_fd_handler(ioc, read_ctx, io_read,
+            write_ctx, io_write, ioc);
+}
+

With those minor fixes,

Thanks, will fix.

Quoted 8 lines
Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.
Virtualization:  qemu.org | libguestfs.org
#11 lore raw
In reply to Eric Blake
Quoted 150 lines · Eric Blake
Quoted 55 lines · Stefan Hajnoczi

The ongoing QEMU multi-queue block layer effort makes it possible for multiple threads to process I/O in parallel. The nbd block driver is not compatible with the multi-queue block layer yet because QIOChannel cannot be used easily from coroutines running in multiple threads. This series changes the QIOChannel API to make that possible.

In the current API, calling qio_channel_attach_aio_context() sets the AioContext where qio_channel_yield() installs an fd handler prior to yielding:

  qio_channel_attach_aio_context(ioc, my_ctx);
  ...
  qio_channel_yield(ioc); // my_ctx is used here
  ...
  qio_channel_detach_aio_context(ioc);

This API design has limitations: reading and writing must be done in the same AioContext and moving between AioContexts involves a cumbersome sequence of API calls that is not suitable for doing on a per-request basis.

There is no fundamental reason why a QIOChannel needs to run within the same AioContext every time qio_channel_yield() is called. QIOChannel only uses the AioContext while inside qio_channel_yield(). The rest of the time, QIOChannel is independent of any AioContext.

In the new API, qio_channel_yield() queries the AioContext from the current coroutine using qemu_coroutine_get_aio_context(). There is no need to explicitly attach/detach AioContexts anymore and qio_channel_attach_aio_context() and qio_channel_detach_aio_context() are gone. One coroutine can read from the QIOChannel while another coroutine writes from a different AioContext.

This API change allows the nbd block driver to use QIOChannel from any thread. It's important to keep in mind that the block driver already synchronizes QIOChannel access and ensures that two coroutines never read simultaneously or write simultaneously.

This patch updates all users of qio_channel_attach_aio_context() to the new API. Most conversions are simple, but vhost-user-server requires a new qemu_coroutine_yield() call to quiesce the vu_client_trip() coroutine when not attached to any AioContext.

While the API is has become simpler, there is one wart: QIOChannel has a special case for the iohandler AioContext (used for handlers that must not run in nested event loops). I didn't find an elegant way preserve that behavior, so I added a new API called qio_channel_set_follow_coroutine_ctx(ioc, true|false) for opting in to the new AioContext model. By default QIOChannel uses the iohandler AioHandler. Code that formerly called qio_channel_attach_aio_context() now calls qio_channel_set_follow_coroutine_ctx(ioc, true) once after the QIOChannel is created.

---
+++ b/include/io/channel-util.h

include/io/channel-util.h line 53

Quoted 5 lines
include/io/channel-util.h +2 -0
@@ -49,4 +49,27 @@
 QIOChannel *qio_channel_new_fd(int fd,
                                Error **errp);
 
+/**
+ * qio_channel_yield:

Wrong function name...

Quoted 4 lines
include/io/channel-util.h +4 -0
+ * @read_fd: the file descriptor for the read handler
+ * @read_ctx: the AioContext for the read handler
+ * @io_read: the read handler
+ * @write_fd: the file descriptor for the write handler

include/io/channel-util.h line 67

Quoted 10 lines
include/io/channel-util.h +10 -0
+ * @write_ctx: the AioContext for the write handler
+ * @io_write: the write handler
+ * @opaque: the opaque argument to the read and write handler
+ *
+ * Set the read and write handlers when @read_ctx and @write_ctx are non-NULL,
+ * respectively. To leave a handler in its current state, pass a NULL
+ * AioContext. To clear a handler, pass a non-NULL AioContext and a NULL
+ * handler.
+ */
+void qio_channel_util_set_aio_fd_handler(int read_fd,

...should be this.

Quoted 27 lines
include/io/channel-util.h +7 -0
+                                         AioContext *read_ctx,
+                                         IOHandler *io_read,
+                                         int write_fd,
+                                         AioContext *write_ctx,
+                                         IOHandler *io_write,
+                                         void *opaque);
+
 #endif /* QIO_CHANNEL_UTIL_H */
include/io/channel.h +10 -3
index 229bf36910..5f9dbaab65 100644
--- a/include/io/channel.h
 
+++ b/io/channel.c
 
 
-static void qio_channel_set_aio_fd_handlers(QIOChannel *ioc)
+static void coroutine_fn
+qio_channel_set_fd_handlers(QIOChannel *ioc, GIOCondition condition)
 {
-    IOHandler *rd_handler = NULL, *wr_handler = NULL;
+    AioContext *ctx = ioc->follow_coroutine_ctx ?
+        qemu_coroutine_get_aio_context(qemu_coroutine_self()) :
+        iohandler_get_aio_context();
+    AioContext *read_ctx = NULL;
+    IOHandler *io_read = NULL;
+    AioContext *write_ctx = NULL;
+    IOHandler *io_write = NULL;

include/io/channel.h

Quoted 10 lines
include/io/channel.h +10 -0
+
+    if (condition == G_IO_IN) {
+        ioc->read_coroutine = qemu_coroutine_self();
+        ioc->read_ctx = ctx;
+        read_ctx = ctx;
+        io_read = qio_channel_restart_read;
+
+        /*
+         * Thread safety: if the other coroutine is set and its AioContext
+         * match ours, then there is mutual exclusion between read and write

matches

Quoted 16 lines
include/io/channel.h +16 -0
+         * because they share a single thread and it's safe to set both read
+         * and write fd handlers here. If the AioContext does not match ours,
+         * then both threads may run in parallel but there is no shared state
+         * to worry about.
+         */
+        if (ioc->write_coroutine && ioc->write_ctx == ctx) {
+            write_ctx = ctx;
+            io_write = qio_channel_restart_write;
+        }
+    } else if (condition == G_IO_OUT) {
+        ioc->write_coroutine = qemu_coroutine_self();
+        ioc->write_ctx = ctx;
+        write_ctx = ctx;
+        io_write = qio_channel_restart_write;
+        if (ioc->read_coroutine && ioc->read_ctx == ctx) {
+            read_ctx = ctx;

include/io/channel.h

Quoted 10 lines
include/io/channel.h +10 -0
+            io_read = qio_channel_restart_read;
+        }
+    } else {
+        abort();
+    }
+
+    qio_channel_set_aio_fd_handler(ioc, read_ctx, io_read,
+            write_ctx, io_write, ioc);
+}
+

With those minor fixes,

Quoted 1 line

Eric, feel free to merge the two io subsystem changes at the same time as the NBD changes.

With regards,
Daniel

Signature · 3 lines
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|
#12 lore raw
Quoted 9 lines · Daniel P. Berrangé
Quoted 4 lines

With those minor fixes,

Eric, feel free to merge the two io subsystem changes at the same time as the NBD changes.

Ok; I'm queuing this series through my NBD branch, and will send a pull request within the next few days.

Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.
Virtualization:  qemu.org | libguestfs.org