From: Laurent Vivier <lviv...@redhat.com>
This commit adds support for the vhost-user interface to the passt
network backend, enabling high-performance, accelerated networking for
guests using passt.
The passt backend can now operate in a vhost-user mode, where it
communicates with the guest's virtio-net device over a socket pair
using the vhost-user protocol. This offloads the datapath from the
main QEMU loop, significantly improving network performance.
When the vhost-user=on option is used with -netdev passt, the new
vhost initialization path is taken instead of the standard
stream-based connection.
Signed-off-by: Laurent Vivier <lviv...@redhat.com>
Signed-off-by: Jason Wang <jasow...@redhat.com>
---
docs/system/devices/net.rst | 12 +-
net/passt.c | 346 ++++++++++++++++++++++++++++++++++++
qapi/net.json | 3 +
qemu-options.hx | 10 +-
4 files changed, 369 insertions(+), 2 deletions(-)
diff --git a/docs/system/devices/net.rst b/docs/system/devices/net.rst
index c586ee0f40..4d787c3aeb 100644
--- a/docs/system/devices/net.rst
+++ b/docs/system/devices/net.rst
@@ -104,7 +104,7 @@ To use the passt backend interface
There is no need to start the daemon as QEMU will do it for you.
-passt is started in the socket-based mode.
+By default, passt will be started in the socket-based mode.
.. parsed-literal::
|qemu_system| [...OPTIONS...] -nic passt
@@ -128,6 +128,16 @@ passt is started in the socket-based mode.
virtio0: index=0,type=nic,model=virtio-net-pci,macaddr=9a:2b:2c:2d:2e:2f
\ netdev0: index=0,type=passt,stream,connected to pid 25428
+To use the vhost-based interface, add the ``vhost-user=on`` parameter and
+select the virtio-net device:
+
+.. parsed-literal::
+ |qemu_system| [...OPTIONS...] -nic passt,model=virtio,vhost-user=on
+
+ (qemu) info network
+ virtio-net-pci.0:
index=0,type=nic,model=virtio-net-pci,macaddr=52:54:00:12:34:56
+ \ #net006: index=0,type=passt,vhost-user,connected to pid 25731
+
To use socket based passt interface:
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
diff --git a/net/passt.c b/net/passt.c
index 0a4a1ba6aa..6f616ba3c2 100644
--- a/net/passt.c
+++ b/net/passt.c
@@ -7,18 +7,75 @@
*/
#include "qemu/osdep.h"
#include <glib/gstdio.h>
+#include "qemu/error-report.h"
#include <gio/gio.h>
#include "net/net.h"
#include "clients.h"
#include "qapi/error.h"
#include "io/net-listener.h"
+#include "chardev/char-fe.h"
+#include "net/vhost_net.h"
+#include "hw/virtio/vhost.h"
+#include "hw/virtio/vhost-user.h"
+#include "standard-headers/linux/virtio_net.h"
#include "stream_data.h"
+#ifdef CONFIG_VHOST_USER
+static const int user_feature_bits[] = {
+ VIRTIO_F_NOTIFY_ON_EMPTY,
+ VIRTIO_F_NOTIFICATION_DATA,
+ VIRTIO_RING_F_INDIRECT_DESC,
+ VIRTIO_RING_F_EVENT_IDX,
+
+ VIRTIO_F_ANY_LAYOUT,
+ VIRTIO_F_VERSION_1,
+ VIRTIO_NET_F_CSUM,
+ VIRTIO_NET_F_GUEST_CSUM,
+ VIRTIO_NET_F_GSO,
+ VIRTIO_NET_F_GUEST_TSO4,
+ VIRTIO_NET_F_GUEST_TSO6,
+ VIRTIO_NET_F_GUEST_ECN,
+ VIRTIO_NET_F_GUEST_UFO,
+ VIRTIO_NET_F_HOST_TSO4,
+ VIRTIO_NET_F_HOST_TSO6,
+ VIRTIO_NET_F_HOST_ECN,
+ VIRTIO_NET_F_HOST_UFO,
+ VIRTIO_NET_F_MRG_RXBUF,
+ VIRTIO_NET_F_MTU,
+ VIRTIO_F_IOMMU_PLATFORM,
+ VIRTIO_F_RING_PACKED,
+ VIRTIO_F_RING_RESET,
+ VIRTIO_F_IN_ORDER,
+ VIRTIO_NET_F_RSS,
+ VIRTIO_NET_F_RSC_EXT,
+ VIRTIO_NET_F_HASH_REPORT,
+ VIRTIO_NET_F_GUEST_USO4,
+ VIRTIO_NET_F_GUEST_USO6,
+ VIRTIO_NET_F_HOST_USO,
+
+ /* This bit implies RARP isn't sent by QEMU out of band */
+ VIRTIO_NET_F_GUEST_ANNOUNCE,
+
+ VIRTIO_NET_F_MQ,
+
+ VHOST_INVALID_FEATURE_BIT
+};
+#endif
+
typedef struct NetPasstState {
NetStreamData data;
GPtrArray *args;
gchar *pidfile;
pid_t pid;
+#ifdef CONFIG_VHOST_USER
+ /* vhost user */
+ VhostUserState *vhost_user;
+ VHostNetState *vhost_net;
+ CharBackend vhost_chr;
+ guint vhost_watch;
+ uint64_t acked_features;
+ bool started;
+#endif
} NetPasstState;
static int net_passt_stream_start(NetPasstState *s, Error **errp);
@@ -27,6 +84,24 @@ static void net_passt_cleanup(NetClientState *nc)
{
NetPasstState *s = DO_UPCAST(NetPasstState, data.nc, nc);
+#ifdef CONFIG_VHOST_USER
+ if (s->vhost_net) {
+ vhost_net_cleanup(s->vhost_net);
+ g_free(s->vhost_net);
+ s->vhost_net = NULL;
+ }
+ if (s->vhost_watch) {
+ g_source_remove(s->vhost_watch);
+ s->vhost_watch = 0;
+ }
+ qemu_chr_fe_deinit(&s->vhost_chr, true);
+ if (s->vhost_user) {
+ vhost_user_cleanup(s->vhost_user);
+ g_free(s->vhost_user);
+ s->vhost_user = NULL;
+ }
+#endif
+
kill(s->pid, SIGTERM);
g_remove(s->pidfile);
g_free(s->pidfile);
@@ -60,11 +135,98 @@ static gboolean net_passt_send(QIOChannel *ioc,
GIOCondition condition,
return G_SOURCE_CONTINUE;
}
+#ifdef CONFIG_VHOST_USER
+static int passt_set_vnet_endianness(NetClientState *nc, bool enable)
+{
+ assert(nc->info->type == NET_CLIENT_DRIVER_PASST);
+
+ return 0;
+}
+
+static bool passt_has_vnet_hdr(NetClientState *nc)
+{
+ NetPasstState *s = DO_UPCAST(NetPasstState, data.nc, nc);
+
+ assert(nc->info->type == NET_CLIENT_DRIVER_PASST);
+
+ return s->vhost_user != NULL;
+}
+
+static bool passt_has_ufo(NetClientState *nc)
+{
+ NetPasstState *s = DO_UPCAST(NetPasstState, data.nc, nc);
+
+ assert(nc->info->type == NET_CLIENT_DRIVER_PASST);
+
+ return s->vhost_user != NULL;
+}
+
+static bool passt_check_peer_type(NetClientState *nc, ObjectClass *oc,
+ Error **errp)
+{
+ NetPasstState *s = DO_UPCAST(NetPasstState, data.nc, nc);
+ const char *driver = object_class_get_name(oc);
+
+ assert(nc->info->type == NET_CLIENT_DRIVER_PASST);
+
+ if (s->vhost_user == NULL) {
+ return true;
+ }
+
+ if (!g_str_has_prefix(driver, "virtio-net-")) {
+ error_setg(errp, "vhost-user requires frontend driver virtio-net-*");
+ return false;
+ }
+
+ return true;
+}
+
+static struct vhost_net *passt_get_vhost_net(NetClientState *nc)
+{
+ NetPasstState *s = DO_UPCAST(NetPasstState, data.nc, nc);
+
+ assert(nc->info->type == NET_CLIENT_DRIVER_PASST);
+
+ return s->vhost_net;
+}
+
+static uint64_t passt_get_acked_features(NetClientState *nc)
+{
+ NetPasstState *s = DO_UPCAST(NetPasstState, data.nc, nc);
+
+ assert(nc->info->type == NET_CLIENT_DRIVER_PASST);
+
+ return s->acked_features;
+}
+
+static void passt_save_acked_features(NetClientState *nc)
+{
+ NetPasstState *s = DO_UPCAST(NetPasstState, data.nc, nc);
+
+ assert(nc->info->type == NET_CLIENT_DRIVER_PASST);
+
+ if (s->vhost_net) {
+ uint64_t features = vhost_net_get_acked_features(s->vhost_net);
+ if (features) {
+ s->acked_features = features;
+ }
+ }
+}
+#endif
+
static NetClientInfo net_passt_info = {
.type = NET_CLIENT_DRIVER_PASST,
.size = sizeof(NetPasstState),
.receive = net_passt_receive,
.cleanup = net_passt_cleanup,
+#ifdef CONFIG_VHOST_USER
+ .has_vnet_hdr = passt_has_vnet_hdr,
+ .has_ufo = passt_has_ufo,
+ .set_vnet_be = passt_set_vnet_endianness,
+ .set_vnet_le = passt_set_vnet_endianness,
+ .check_peer_type = passt_check_peer_type,
+ .get_vhost_net = passt_get_vhost_net,
+#endif
};
static void net_passt_client_connected(QIOTask *task, gpointer opaque)
@@ -163,6 +325,177 @@ static int net_passt_stream_start(NetPasstState *s, Error
**errp)
return 0;
}
+#ifdef CONFIG_VHOST_USER
+static gboolean passt_vhost_user_watch(void *do_not_use, GIOCondition cond,
+ void *opaque)
+{
+ NetPasstState *s = opaque;
+
+ qemu_chr_fe_disconnect(&s->vhost_chr);
+
+ return G_SOURCE_CONTINUE;
+}
+
+static void passt_vhost_user_event(void *opaque, QEMUChrEvent event);
+
+static void chr_closed_bh(void *opaque)
+{
+ NetPasstState *s = opaque;
+
+ passt_save_acked_features(&s->data.nc);
+
+ net_client_set_link(&(NetClientState *){ &s->data.nc }, 1, false);
+
+ qemu_chr_fe_set_handlers(&s->vhost_chr, NULL, NULL, passt_vhost_user_event,
+ NULL, s, NULL, true);
+}
+
+static void passt_vhost_user_stop(NetPasstState *s)
+{
+ passt_save_acked_features(&s->data.nc);
+ vhost_net_cleanup(s->vhost_net);
+}
+
+static int passt_vhost_user_start(NetPasstState *s, VhostUserState *be)
+{
+ struct vhost_net *net = NULL;
+ VhostNetOptions options;
+
+ options.backend_type = VHOST_BACKEND_TYPE_USER;
+ options.net_backend = &s->data.nc;
+ options.opaque = be;
+ options.busyloop_timeout = 0;
+ options.nvqs = 2;
+ options.feature_bits = user_feature_bits;
+ options.max_tx_queue_size = VIRTQUEUE_MAX_SIZE;
+ options.get_acked_features = passt_get_acked_features;
+ options.save_acked_features = passt_save_acked_features;
+ options.is_vhost_user = true;
+
+ net = vhost_net_init(&options);
+ if (!net) {
+ error_report("failed to init passt vhost_net");
+ goto err;
+ }
+
+ if (s->vhost_net) {
+ vhost_net_cleanup(s->vhost_net);
+ g_free(s->vhost_net);
+ }
+ s->vhost_net = net;
+
+ return 0;
+err:
+ if (net) {
+ vhost_net_cleanup(net);
+ g_free(net);
+ }
+ passt_vhost_user_stop(s);
+ return -1;
+}
+
+static void passt_vhost_user_event(void *opaque, QEMUChrEvent event)
+{
+ NetPasstState *s = opaque;
+ Error *err = NULL;
+
+ switch (event) {
+ case CHR_EVENT_OPENED:
+ if (passt_vhost_user_start(s, s->vhost_user) < 0) {
+ qemu_chr_fe_disconnect(&s->vhost_chr);
+ return;
+ }
+ s->vhost_watch = qemu_chr_fe_add_watch(&s->vhost_chr, G_IO_HUP,
+ passt_vhost_user_watch, s);
+ net_client_set_link(&(NetClientState *){ &s->data.nc }, 1, true);
+ s->started = true;
+ break;
+ case CHR_EVENT_CLOSED:
+ if (s->vhost_watch) {
+ AioContext *ctx = qemu_get_current_aio_context();
+
+ g_source_remove(s->vhost_watch);
+ s->vhost_watch = 0;
+ qemu_chr_fe_set_handlers(&s->vhost_chr, NULL, NULL, NULL, NULL,
+ NULL, NULL, false);
+
+ aio_bh_schedule_oneshot(ctx, chr_closed_bh, s);
+ }
+ break;
+ case CHR_EVENT_BREAK:
+ case CHR_EVENT_MUX_IN:
+ case CHR_EVENT_MUX_OUT:
+ /* Ignore */
+ break;
+ }
+
+ if (err) {
+ error_report_err(err);
+ }
+}
+
+static int net_passt_vhost_user_init(NetPasstState *s, Error **errp)
+{
+ Chardev *chr;
+ int sv[2];
+
+ if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) {
+ error_setg_errno(errp, errno, "socketpair() failed");
+ return -1;
+ }
+
+ /* connect to passt */
+ qemu_set_info_str(&s->data.nc, "connecting to passt");
+
+ /* create chardev */
+
+ chr = CHARDEV(object_new(TYPE_CHARDEV_SOCKET));
+ if (!chr || qemu_chr_add_client(chr, sv[0]) == -1) {
+ object_unref(OBJECT(chr));
+ error_setg(errp, "Failed to make socket chardev");
+ goto err;
+ }
+
+ s->vhost_user = g_new0(struct VhostUserState, 1);
+ if (!qemu_chr_fe_init(&s->vhost_chr, chr, errp) ||
+ !vhost_user_init(s->vhost_user, &s->vhost_chr, errp)) {
+ goto err;
+ }
+
+ /* start passt */
+ if (net_passt_start_daemon(s, sv[1], errp) == -1) {
+ goto err;
+ }
+
+ do {
+ if (qemu_chr_fe_wait_connected(&s->vhost_chr, errp) < 0) {
+ goto err;
+ }
+
+ qemu_chr_fe_set_handlers(&s->vhost_chr, NULL, NULL,
+ passt_vhost_user_event, NULL, s, NULL,
+ true);
+ } while (!s->started);
+
+ qemu_set_info_str(&s->data.nc, "vhost-user,connected to pid %d", s->pid);
+
+ close(sv[1]);
+ return 0;
+err:
+ close(sv[0]);
+ close(sv[1]);
+
+ return -1;
+}
+#else
+static int net_passt_vhost_user_init(NetPasstState *s, Error **errp)
+{
+ error_setg(errp, "vhost-user support has not been built");
+
+ return -1;
+}
+#endif
+
static GPtrArray *net_passt_decode_args(const NetDevPasstOptions *passt,
gchar *pidfile, Error **errp)
{
@@ -174,6 +507,10 @@ static GPtrArray *net_passt_decode_args(const
NetDevPasstOptions *passt,
g_ptr_array_add(args, g_strdup("passt"));
}
+ if (passt->has_vhost_user && passt->vhost_user) {
+ g_ptr_array_add(args, g_strdup("--vhost-user"));
+ }
+
/* by default, be quiet */
if (!passt->has_quiet || passt->quiet) {
g_ptr_array_add(args, g_strdup("--quiet"));
@@ -398,6 +735,15 @@ int net_init_passt(const Netdev *netdev, const char *name,
s->args = args;
s->pidfile = pidfile;
+ if (netdev->u.passt.has_vhost_user && netdev->u.passt.vhost_user) {
+ if (net_passt_vhost_user_init(s, errp) == -1) {
+ qemu_del_net_client(nc);
+ return -1;
+ }
+
+ return 0;
+ }
+
if (net_passt_stream_start(s, errp) == -1) {
qemu_del_net_client(nc);
return -1;
diff --git a/qapi/net.json b/qapi/net.json
index 24999f6752..0f766041a3 100644
--- a/qapi/net.json
+++ b/qapi/net.json
@@ -121,6 +121,8 @@
#
# @quiet: don't print informational messages (default, passed as '--quiet')
#
+# @vhost-user: enable vhost-user
+#
# @mtu: assign MTU via DHCP/NDP
#
# @address: IPv4 or IPv6 address
@@ -189,6 +191,7 @@
'data': {
'*path': 'str',
'*quiet': 'bool',
+ '*vhost-user': 'bool',
'*mtu': 'int',
'*address': 'str',
'*netmask': 'str',
diff --git a/qemu-options.hx b/qemu-options.hx
index e8252cd5e8..a3c066c678 100644
--- a/qemu-options.hx
+++ b/qemu-options.hx
@@ -2797,7 +2797,7 @@ DEFHEADING(Network options:)
DEF("netdev", HAS_ARG, QEMU_OPTION_netdev,
#ifdef CONFIG_PASST
- "-netdev passt,id=str[,path=file][,quiet=on|off]\n"
+ "-netdev passt,id=str[,path=file][,quiet=on|off][,vhost-user=on|off]\n"
"[,mtu=mtu][,address=addr][,netmask=mask][,mac=addr][,gateway=addr]\n"
" [,interface=name][,outbound=address][,outbound-if4=name]\n"
" [,outbound-if6=name][,dns=addr][,search=list][,fqdn=name]\n"
@@ -2811,6 +2811,8 @@ DEF("netdev", HAS_ARG, QEMU_OPTION_netdev,
" if 'path' is not provided 'passt' will be started
according to PATH\n"
" by default, informational message of passt are not
displayed (quiet=on)\n"
" to display this message, use 'quiet=off'\n"
+ " by default, passt will be started in socket-based mode,
to enable vhost-mode,\n"
+ " use 'vhost-user=on'\n"
" for details on other options, refer to passt(1)\n"
" 'param' allows to pass any option defined by passt(1)\n"
#endif
@@ -3067,6 +3069,12 @@ SRST
By default, ``quiet=on`` to disable informational message from
passt. ``quiet=on`` is passed as ``--quiet`` to passt.
+ ``vhost-user=on|off``
+ By default, ``vhost-user=off`` and QEMU uses the stream network
+ backend to communicate with passt. If ``vhost-user=on``, passt is
+ started with ``--vhost-user`` and QEMU uses the vhost-user network
+ backend to communicate with passt.
+
``@mtu``
Assign MTU via DHCP/NDP
--
2.42.0
