[Qemu-devel] [PATCH 10/11] xen: add net backend driver.

[Gerd Hoffmann <kraxel@redhat.com>]

This patch adds a network interface backend driver to qemu.  It is a pure
userspace implemention using the gntdev interface.  It uses "qnet" as
backend name in xenstore so it doesn't interfere with the netback
backend (aka "vnif").

The network backend is hooked into the corrosponding qemu vlan, i.e.
vif 0 is hooked into vlan 0.  To make the packages actually arrive
somewhere you additionally have to link the vlan to the outside world
using the usual qemu command line options such as "-net tap,...".

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
---
 Makefile.target     |    2 +-
 hw/xen_backend.h    |    1 +
 hw/xen_machine_pv.c |    1 +
 hw/xen_nic.c        |  380 +++++++++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 383 insertions(+), 1 deletions(-)
 create mode 100644 hw/xen_nic.c

diff --git a/Makefile.target b/Makefile.target
index 41cc5e7..340de31 100644
--- a/Makefile.target
+++ b/Makefile.target
@@ -517,7 +517,7 @@ endif
 
 # xen backend driver support
 XEN_OBJS := xen_machine_pv.o xen_backend.o
-XEN_OBJS += xen_console.o xen_framebuffer.o xen_disk.o
+XEN_OBJS += xen_console.o xen_framebuffer.o xen_disk.o xen_nic.o
 ifeq ($(CONFIG_XEN), yes)
   OBJS += $(XEN_OBJS)
   LIBS += $(XEN_LIBS)
diff --git a/hw/xen_backend.h b/hw/xen_backend.h
index 9dbcb19..9ea4421 100644
--- a/hw/xen_backend.h
+++ b/hw/xen_backend.h
@@ -89,6 +89,7 @@ struct XenDevOps xen_console_ops;      /* xen_console.c     */
 struct XenDevOps xen_kbdmouse_ops;     /* xen_framebuffer.c */
 struct XenDevOps xen_framebuffer_ops;  /* xen_framebuffer.c */
 struct XenDevOps xen_blkdev_ops;       /* xen_disk.c        */
+struct XenDevOps xen_netdev_ops;       /* xen_nic.c         */
 
 void xen_set_display(int domid, DisplayState *ds);
 
diff --git a/hw/xen_machine_pv.c b/hw/xen_machine_pv.c
index 27b85e7..e6d2bd6 100644
--- a/hw/xen_machine_pv.c
+++ b/hw/xen_machine_pv.c
@@ -102,6 +102,7 @@ static int xen_init_pv(DisplayState *ds)
     xen_be_register("vkbd", &xen_kbdmouse_ops);
     xen_be_register("vfb", &xen_framebuffer_ops);
     xen_be_register("qdisk", &xen_blkdev_ops);
+    xen_be_register("qnic", &xen_netdev_ops);
 
     /* setup framebuffer */
     xen_set_display(xen_domid, ds);
diff --git a/hw/xen_nic.c b/hw/xen_nic.c
new file mode 100644
index 0000000..ce315c0
--- /dev/null
+++ b/hw/xen_nic.c
@@ -0,0 +1,380 @@
+/*
+ *  xen paravirt network card backend
+ *
+ *  (c) Gerd Hoffmann <kraxel@redhat.com>
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License as published by
+ *  the Free Software Foundation; under version 2 of the License.
+ *
+ *  This program is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *  GNU General Public License for more details.
+ *
+ *  You should have received a copy of the GNU General Public License
+ *  along with this program; if not, write to the Free Software
+ *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdarg.h>
+#include <string.h>
+#include <unistd.h>
+#include <signal.h>
+#include <inttypes.h>
+#include <fcntl.h>
+#include <errno.h>
+#include <pthread.h>
+#include <sys/socket.h>
+#include <sys/ioctl.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <sys/mman.h>
+#include <sys/wait.h>
+#include <linux/if.h>
+#include <linux/if_tun.h>
+
+#include <xs.h>
+#include <xenctrl.h>
+#include <xen/io/xenbus.h>
+#include <xen/io/netif.h>
+
+#include "hw.h"
+#include "net.h"
+#include "qemu-char.h"
+#include "xen_backend.h"
+
+/* ------------------------------------------------------------- */
+
+struct XenNetDev {
+    struct XenDevice      xendev;  /* must be first */
+    char                  *mac;
+    int                   tx_work;
+    int                   tx_ring_ref;
+    int                   rx_ring_ref;
+    struct netif_tx_sring *txs;
+    struct netif_rx_sring *rxs;
+    netif_tx_back_ring_t  tx_ring;
+    netif_rx_back_ring_t  rx_ring;
+    VLANClientState       *vs;
+};
+
+/* ------------------------------------------------------------- */
+
+static void net_tx_response(struct XenNetDev *netdev, netif_tx_request_t *txp, int8_t st)
+{
+    RING_IDX i = netdev->tx_ring.rsp_prod_pvt;
+    netif_tx_response_t *resp;
+    int notify;
+
+    resp = RING_GET_RESPONSE(&netdev->tx_ring, i);
+    resp->id     = txp->id;
+    resp->status = st;
+
+#if 0
+    if (txp->flags & NETTXF_extra_info)
+	RING_GET_RESPONSE(&netdev->tx_ring, ++i)->status = NETIF_RSP_NULL;
+#endif
+
+    netdev->tx_ring.rsp_prod_pvt = ++i;
+    RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&netdev->tx_ring, notify);
+    if (notify)
+	xen_be_send_notify(&netdev->xendev);
+
+    if (i == netdev->tx_ring.req_cons) {
+	int more_to_do;
+	RING_FINAL_CHECK_FOR_REQUESTS(&netdev->tx_ring, more_to_do);
+	if (more_to_do)
+	    netdev->tx_work++;
+    }
+}
+
+static void net_tx_error(struct XenNetDev *netdev, netif_tx_request_t *txp, RING_IDX end)
+{
+#if 0
+    /*
+     * Hmm, why netback fails everything in the ring?
+     * Should we do that even when not supporting SG and TSO?
+     */
+    RING_IDX cons = netdev->tx_ring.req_cons;
+
+    do {
+	make_tx_response(netif, txp, NETIF_RSP_ERROR);
+	if (cons >= end)
+	    break;
+	txp = RING_GET_REQUEST(&netdev->tx_ring, cons++);
+    } while (1);
+    netdev->tx_ring.req_cons = cons;
+    netif_schedule_work(netif);
+    netif_put(netif);
+#else
+    net_tx_response(netdev, txp, NETIF_RSP_ERROR);
+#endif
+}
+
+static void net_tx_packets(struct XenNetDev *netdev)
+{
+    netif_tx_request_t txreq;
+    RING_IDX rc, rp;
+    void *page;
+
+    for (;;) {
+	rc = netdev->tx_ring.req_cons;
+	rp = netdev->tx_ring.sring->req_prod;
+	xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
+
+	while ((rc != rp)) {
+	    if (RING_REQUEST_CONS_OVERFLOW(&netdev->tx_ring, rc))
+		break;
+	    memcpy(&txreq, RING_GET_REQUEST(&netdev->tx_ring, rc), sizeof(txreq));
+	    netdev->tx_ring.req_cons = ++rc;
+
+#if 1
+	    /* should not happen in theory, we don't announce the *
+	     * feature-{sg,gso,whatelse} flags in xenstore (yet?) */
+	    if (txreq.flags & NETTXF_extra_info) {
+		xen_be_printf(&netdev->xendev, 0, "FIXME: extra info flag\n");
+		net_tx_error(netdev, &txreq, rc);
+		continue;
+	    }
+	    if (txreq.flags & NETTXF_more_data) {
+		xen_be_printf(&netdev->xendev, 0, "FIXME: more data flag\n");
+		net_tx_error(netdev, &txreq, rc);
+		continue;
+	    }
+#endif
+
+	    if (txreq.size < 14) {
+		xen_be_printf(&netdev->xendev, 0, "bad packet size: %d\n", txreq.size);
+		net_tx_error(netdev, &txreq, rc);
+		continue;
+	    }
+
+	    if ((txreq.offset + txreq.size) > XC_PAGE_SIZE) {
+		xen_be_printf(&netdev->xendev, 0, "error: page crossing\n");
+		net_tx_error(netdev, &txreq, rc);
+		continue;
+	    }
+
+	    xen_be_printf(&netdev->xendev, 3, "tx packet ref %d, off %d, len %d, flags 0x%x%s%s%s%s\n",
+			  txreq.gref, txreq.offset, txreq.size, txreq.flags,
+			  (txreq.flags & NETTXF_csum_blank)     ? " csum_blank"     : "",
+			  (txreq.flags & NETTXF_data_validated) ? " data_validated" : "",
+			  (txreq.flags & NETTXF_more_data)      ? " more_data"      : "",
+			  (txreq.flags & NETTXF_extra_info)     ? " extra_info"     : "");
+
+	    page = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
+					   netdev->xendev.dom,
+					   txreq.gref, PROT_READ);
+	    if (NULL == page) {
+		xen_be_printf(&netdev->xendev, 0, "error: gref dereference failed\n");
+		net_tx_error(netdev, &txreq, rc);
+		continue;
+	    }
+	    if (txreq.flags & NETTXF_csum_blank)
+		net_checksum_calculate(page + txreq.offset, txreq.size);
+	    qemu_send_packet(netdev->vs, page + txreq.offset, txreq.size);
+	    xc_gnttab_munmap(netdev->xendev.gnttabdev, page, 1);
+	    net_tx_response(netdev, &txreq, NETIF_RSP_OKAY);
+	}
+	if (!netdev->tx_work)
+	    break;
+	netdev->tx_work = 0;
+    }
+}
+
+/* ------------------------------------------------------------- */
+
+static void net_rx_response(struct XenNetDev *netdev,
+			    netif_rx_request_t *req, int8_t st,
+			    uint16_t offset, uint16_t size,
+			    uint16_t flags)
+{
+    RING_IDX i = netdev->rx_ring.rsp_prod_pvt;
+    netif_rx_response_t *resp;
+    int notify;
+
+    resp = RING_GET_RESPONSE(&netdev->rx_ring, i);
+    resp->offset     = offset;
+    resp->flags      = flags;
+    resp->id         = req->id;
+    resp->status     = (int16_t)size;
+    if (st < 0)
+	resp->status = (int16_t)st;
+
+    xen_be_printf(&netdev->xendev, 3, "rx response: idx %d, status %d, flags 0x%x\n",
+		  i, resp->status, resp->flags);
+
+    netdev->rx_ring.rsp_prod_pvt = ++i;
+    RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&netdev->rx_ring, notify);
+    if (notify)
+	xen_be_send_notify(&netdev->xendev);
+}
+
+#define NET_IP_ALIGN 2
+
+static int net_rx_ok(void *opaque)
+{
+    struct XenNetDev *netdev = opaque;
+    RING_IDX rc, rp;
+
+    if (netdev->xendev.be_state != XenbusStateConnected)
+	return 0;
+
+    rc = netdev->rx_ring.req_cons;
+    rp = netdev->rx_ring.sring->req_prod;
+    xen_rmb();
+
+    if (rc == rp || RING_REQUEST_CONS_OVERFLOW(&netdev->rx_ring, rc)) {
+	xen_be_printf(&netdev->xendev, 2, "%s: no rx buffers (%d/%d)\n",
+		      __FUNCTION__, rc, rp);
+	return 0;
+    }
+    return 1;
+}
+
+static void net_rx_packet(void *opaque, const uint8_t *buf, int size)
+{
+    struct XenNetDev *netdev = opaque;
+    netif_rx_request_t rxreq;
+    RING_IDX rc, rp;
+    void *page;
+
+    if (netdev->xendev.be_state != XenbusStateConnected)
+	return;
+
+    rc = netdev->rx_ring.req_cons;
+    rp = netdev->rx_ring.sring->req_prod;
+    xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
+
+    if (rc == rp || RING_REQUEST_CONS_OVERFLOW(&netdev->rx_ring, rc)) {
+	xen_be_printf(&netdev->xendev, 2, "no buffer, drop packet\n");
+	return;
+    }
+    if (size > XC_PAGE_SIZE - NET_IP_ALIGN) {
+	xen_be_printf(&netdev->xendev, 0, "packet too big (%d > %ld)",
+		      size, XC_PAGE_SIZE - NET_IP_ALIGN);
+	return;
+    }
+
+    memcpy(&rxreq, RING_GET_REQUEST(&netdev->rx_ring, rc), sizeof(rxreq));
+    netdev->rx_ring.req_cons = ++rc;
+
+    page = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
+				   netdev->xendev.dom,
+				   rxreq.gref, PROT_WRITE);
+    if (NULL == page) {
+	xen_be_printf(&netdev->xendev, 0, "error: gref dereference failed\n");
+	net_rx_response(netdev, &rxreq, NETIF_RSP_ERROR, 0, 0, 0);
+	return;
+    }
+    memcpy(page + NET_IP_ALIGN, buf, size);
+    xc_gnttab_munmap(netdev->xendev.gnttabdev, page, 1);
+    net_rx_response(netdev, &rxreq, NETIF_RSP_OKAY, NET_IP_ALIGN, size, 0);
+}
+
+/* ------------------------------------------------------------- */
+
+static int net_init(struct XenDevice *xendev)
+{
+    struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
+    VLANState *vlan;
+
+    /* read xenstore entries */
+    if (NULL == netdev->mac)
+	netdev->mac = xenstore_read_be_str(&netdev->xendev, "mac");
+
+    /* do we have all we need? */
+    if (NULL == netdev->mac)
+	return -1;
+
+    vlan = qemu_find_vlan(netdev->xendev.dev);
+    netdev->vs = qemu_new_vlan_client(vlan, net_rx_packet, net_rx_ok, netdev);
+    snprintf(netdev->vs->info_str, sizeof(netdev->vs->info_str),
+             "nic: xenbus vif macaddr=%s", netdev->mac);
+
+    /* fill info */
+    xenstore_write_be_int(&netdev->xendev, "feature-rx-copy", 1);
+    xenstore_write_be_int(&netdev->xendev, "feature-rx-flip", 0);
+
+    return 0;
+}
+
+static int net_connect(struct XenDevice *xendev)
+{
+    struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
+    int rx_copy;
+
+    if (-1 == xenstore_read_fe_int(&netdev->xendev, "tx-ring-ref",
+				   &netdev->tx_ring_ref))
+	return -1;
+    if (-1 == xenstore_read_fe_int(&netdev->xendev, "rx-ring-ref",
+				   &netdev->rx_ring_ref))
+	return 1;
+    if (-1 == xenstore_read_fe_int(&netdev->xendev, "event-channel",
+				   &netdev->xendev.remote_port))
+	return -1;
+
+    if (-1 == xenstore_read_fe_int(&netdev->xendev, "request-rx-copy", &rx_copy))
+	rx_copy = 0;
+    if (0 == rx_copy) {
+	xen_be_printf(&netdev->xendev, 0, "frontend doesn't support rx-copy.\n");
+	return -1;
+    }
+
+    netdev->txs = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
+					  netdev->xendev.dom,
+					  netdev->tx_ring_ref,
+					  PROT_READ | PROT_WRITE);
+    netdev->rxs = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
+					  netdev->xendev.dom,
+					  netdev->rx_ring_ref,
+					  PROT_READ | PROT_WRITE);
+    if (!netdev->txs || !netdev->rxs)
+	return -1;
+    BACK_RING_INIT(&netdev->tx_ring, netdev->txs, XC_PAGE_SIZE);
+    BACK_RING_INIT(&netdev->rx_ring, netdev->rxs, XC_PAGE_SIZE);
+
+    xen_be_bind_evtchn(&netdev->xendev);
+
+    xen_be_printf(&netdev->xendev, 1, "ok: tx-ring-ref %d, rx-ring-ref %d, "
+		  "remote port %d, local port %d\n",
+		  netdev->tx_ring_ref, netdev->rx_ring_ref,
+		  netdev->xendev.remote_port, netdev->xendev.local_port);
+    return 0;
+}
+
+static void net_disconnect(struct XenDevice *xendev)
+{
+    struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
+
+    xen_be_unbind_evtchn(&netdev->xendev);
+
+    if (netdev->txs) {
+	xc_gnttab_munmap(netdev->xendev.gnttabdev, netdev->txs, 1);
+	netdev->txs = NULL;
+    }
+    if (netdev->rxs) {
+	xc_gnttab_munmap(netdev->xendev.gnttabdev, netdev->rxs, 1);
+	netdev->rxs = NULL;
+    }
+}
+
+static void net_event(struct XenDevice *xendev)
+{
+    struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
+    net_tx_packets(netdev);
+}
+
+/* ------------------------------------------------------------- */
+
+struct XenDevOps xen_netdev_ops = {
+    .size       = sizeof(struct XenNetDev),
+    .flags      = DEVOPS_FLAG_NEED_GNTDEV,
+    .init       = net_init,
+    .connect    = net_connect,
+    .event      = net_event,
+    .disconnect = net_disconnect,
+};
-- 
1.5.5.1