|
1 |
/*
|
|
2 |
* xen paravirt network card backend
|
|
3 |
*
|
|
4 |
* (c) Gerd Hoffmann <kraxel@redhat.com>
|
|
5 |
*
|
|
6 |
* This program is free software; you can redistribute it and/or modify
|
|
7 |
* it under the terms of the GNU General Public License as published by
|
|
8 |
* the Free Software Foundation; under version 2 of the License.
|
|
9 |
*
|
|
10 |
* This program is distributed in the hope that it will be useful,
|
|
11 |
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
12 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
13 |
* GNU General Public License for more details.
|
|
14 |
*
|
|
15 |
* You should have received a copy of the GNU General Public License along
|
|
16 |
* with this program; if not, write to the Free Software Foundation, Inc.,
|
|
17 |
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
|
|
18 |
*/
|
|
19 |
|
|
20 |
#include <stdio.h>
|
|
21 |
#include <stdlib.h>
|
|
22 |
#include <stdarg.h>
|
|
23 |
#include <string.h>
|
|
24 |
#include <unistd.h>
|
|
25 |
#include <signal.h>
|
|
26 |
#include <inttypes.h>
|
|
27 |
#include <fcntl.h>
|
|
28 |
#include <errno.h>
|
|
29 |
#include <pthread.h>
|
|
30 |
#include <sys/socket.h>
|
|
31 |
#include <sys/ioctl.h>
|
|
32 |
#include <sys/types.h>
|
|
33 |
#include <sys/stat.h>
|
|
34 |
#include <sys/mman.h>
|
|
35 |
#include <sys/wait.h>
|
|
36 |
#include <linux/if.h>
|
|
37 |
#include <linux/if_tun.h>
|
|
38 |
|
|
39 |
#include <xs.h>
|
|
40 |
#include <xenctrl.h>
|
|
41 |
#include <xen/io/xenbus.h>
|
|
42 |
#include <xen/io/netif.h>
|
|
43 |
|
|
44 |
#include "hw.h"
|
|
45 |
#include "net.h"
|
|
46 |
#include "qemu-char.h"
|
|
47 |
#include "xen_backend.h"
|
|
48 |
|
|
49 |
/* ------------------------------------------------------------- */
|
|
50 |
|
|
51 |
struct XenNetDev {
|
|
52 |
struct XenDevice xendev; /* must be first */
|
|
53 |
char *mac;
|
|
54 |
int tx_work;
|
|
55 |
int tx_ring_ref;
|
|
56 |
int rx_ring_ref;
|
|
57 |
struct netif_tx_sring *txs;
|
|
58 |
struct netif_rx_sring *rxs;
|
|
59 |
netif_tx_back_ring_t tx_ring;
|
|
60 |
netif_rx_back_ring_t rx_ring;
|
|
61 |
VLANClientState *vs;
|
|
62 |
};
|
|
63 |
|
|
64 |
/* ------------------------------------------------------------- */
|
|
65 |
|
|
66 |
static void net_tx_response(struct XenNetDev *netdev, netif_tx_request_t *txp, int8_t st)
|
|
67 |
{
|
|
68 |
RING_IDX i = netdev->tx_ring.rsp_prod_pvt;
|
|
69 |
netif_tx_response_t *resp;
|
|
70 |
int notify;
|
|
71 |
|
|
72 |
resp = RING_GET_RESPONSE(&netdev->tx_ring, i);
|
|
73 |
resp->id = txp->id;
|
|
74 |
resp->status = st;
|
|
75 |
|
|
76 |
#if 0
|
|
77 |
if (txp->flags & NETTXF_extra_info)
|
|
78 |
RING_GET_RESPONSE(&netdev->tx_ring, ++i)->status = NETIF_RSP_NULL;
|
|
79 |
#endif
|
|
80 |
|
|
81 |
netdev->tx_ring.rsp_prod_pvt = ++i;
|
|
82 |
RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&netdev->tx_ring, notify);
|
|
83 |
if (notify)
|
|
84 |
xen_be_send_notify(&netdev->xendev);
|
|
85 |
|
|
86 |
if (i == netdev->tx_ring.req_cons) {
|
|
87 |
int more_to_do;
|
|
88 |
RING_FINAL_CHECK_FOR_REQUESTS(&netdev->tx_ring, more_to_do);
|
|
89 |
if (more_to_do)
|
|
90 |
netdev->tx_work++;
|
|
91 |
}
|
|
92 |
}
|
|
93 |
|
|
94 |
static void net_tx_error(struct XenNetDev *netdev, netif_tx_request_t *txp, RING_IDX end)
|
|
95 |
{
|
|
96 |
#if 0
|
|
97 |
/*
|
|
98 |
* Hmm, why netback fails everything in the ring?
|
|
99 |
* Should we do that even when not supporting SG and TSO?
|
|
100 |
*/
|
|
101 |
RING_IDX cons = netdev->tx_ring.req_cons;
|
|
102 |
|
|
103 |
do {
|
|
104 |
make_tx_response(netif, txp, NETIF_RSP_ERROR);
|
|
105 |
if (cons >= end)
|
|
106 |
break;
|
|
107 |
txp = RING_GET_REQUEST(&netdev->tx_ring, cons++);
|
|
108 |
} while (1);
|
|
109 |
netdev->tx_ring.req_cons = cons;
|
|
110 |
netif_schedule_work(netif);
|
|
111 |
netif_put(netif);
|
|
112 |
#else
|
|
113 |
net_tx_response(netdev, txp, NETIF_RSP_ERROR);
|
|
114 |
#endif
|
|
115 |
}
|
|
116 |
|
|
117 |
static void net_tx_packets(struct XenNetDev *netdev)
|
|
118 |
{
|
|
119 |
netif_tx_request_t txreq;
|
|
120 |
RING_IDX rc, rp;
|
|
121 |
void *page;
|
|
122 |
void *tmpbuf = NULL;
|
|
123 |
|
|
124 |
for (;;) {
|
|
125 |
rc = netdev->tx_ring.req_cons;
|
|
126 |
rp = netdev->tx_ring.sring->req_prod;
|
|
127 |
xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
|
|
128 |
|
|
129 |
while ((rc != rp)) {
|
|
130 |
if (RING_REQUEST_CONS_OVERFLOW(&netdev->tx_ring, rc))
|
|
131 |
break;
|
|
132 |
memcpy(&txreq, RING_GET_REQUEST(&netdev->tx_ring, rc), sizeof(txreq));
|
|
133 |
netdev->tx_ring.req_cons = ++rc;
|
|
134 |
|
|
135 |
#if 1
|
|
136 |
/* should not happen in theory, we don't announce the *
|
|
137 |
* feature-{sg,gso,whatelse} flags in xenstore (yet?) */
|
|
138 |
if (txreq.flags & NETTXF_extra_info) {
|
|
139 |
xen_be_printf(&netdev->xendev, 0, "FIXME: extra info flag\n");
|
|
140 |
net_tx_error(netdev, &txreq, rc);
|
|
141 |
continue;
|
|
142 |
}
|
|
143 |
if (txreq.flags & NETTXF_more_data) {
|
|
144 |
xen_be_printf(&netdev->xendev, 0, "FIXME: more data flag\n");
|
|
145 |
net_tx_error(netdev, &txreq, rc);
|
|
146 |
continue;
|
|
147 |
}
|
|
148 |
#endif
|
|
149 |
|
|
150 |
if (txreq.size < 14) {
|
|
151 |
xen_be_printf(&netdev->xendev, 0, "bad packet size: %d\n", txreq.size);
|
|
152 |
net_tx_error(netdev, &txreq, rc);
|
|
153 |
continue;
|
|
154 |
}
|
|
155 |
|
|
156 |
if ((txreq.offset + txreq.size) > XC_PAGE_SIZE) {
|
|
157 |
xen_be_printf(&netdev->xendev, 0, "error: page crossing\n");
|
|
158 |
net_tx_error(netdev, &txreq, rc);
|
|
159 |
continue;
|
|
160 |
}
|
|
161 |
|
|
162 |
xen_be_printf(&netdev->xendev, 3, "tx packet ref %d, off %d, len %d, flags 0x%x%s%s%s%s\n",
|
|
163 |
txreq.gref, txreq.offset, txreq.size, txreq.flags,
|
|
164 |
(txreq.flags & NETTXF_csum_blank) ? " csum_blank" : "",
|
|
165 |
(txreq.flags & NETTXF_data_validated) ? " data_validated" : "",
|
|
166 |
(txreq.flags & NETTXF_more_data) ? " more_data" : "",
|
|
167 |
(txreq.flags & NETTXF_extra_info) ? " extra_info" : "");
|
|
168 |
|
|
169 |
page = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
|
|
170 |
netdev->xendev.dom,
|
|
171 |
txreq.gref, PROT_READ);
|
|
172 |
if (page == NULL) {
|
|
173 |
xen_be_printf(&netdev->xendev, 0, "error: tx gref dereference failed (%d)\n",
|
|
174 |
txreq.gref);
|
|
175 |
net_tx_error(netdev, &txreq, rc);
|
|
176 |
continue;
|
|
177 |
}
|
|
178 |
if (txreq.flags & NETTXF_csum_blank) {
|
|
179 |
/* have read-only mapping -> can't fill checksum in-place */
|
|
180 |
if (!tmpbuf)
|
|
181 |
tmpbuf = malloc(PAGE_SIZE);
|
|
182 |
memcpy(tmpbuf, page + txreq.offset, txreq.size);
|
|
183 |
net_checksum_calculate(tmpbuf, txreq.size);
|
|
184 |
qemu_send_packet(netdev->vs, tmpbuf, txreq.size);
|
|
185 |
} else {
|
|
186 |
qemu_send_packet(netdev->vs, page + txreq.offset, txreq.size);
|
|
187 |
}
|
|
188 |
xc_gnttab_munmap(netdev->xendev.gnttabdev, page, 1);
|
|
189 |
net_tx_response(netdev, &txreq, NETIF_RSP_OKAY);
|
|
190 |
}
|
|
191 |
if (!netdev->tx_work)
|
|
192 |
break;
|
|
193 |
netdev->tx_work = 0;
|
|
194 |
}
|
|
195 |
free(tmpbuf);
|
|
196 |
}
|
|
197 |
|
|
198 |
/* ------------------------------------------------------------- */
|
|
199 |
|
|
200 |
static void net_rx_response(struct XenNetDev *netdev,
|
|
201 |
netif_rx_request_t *req, int8_t st,
|
|
202 |
uint16_t offset, uint16_t size,
|
|
203 |
uint16_t flags)
|
|
204 |
{
|
|
205 |
RING_IDX i = netdev->rx_ring.rsp_prod_pvt;
|
|
206 |
netif_rx_response_t *resp;
|
|
207 |
int notify;
|
|
208 |
|
|
209 |
resp = RING_GET_RESPONSE(&netdev->rx_ring, i);
|
|
210 |
resp->offset = offset;
|
|
211 |
resp->flags = flags;
|
|
212 |
resp->id = req->id;
|
|
213 |
resp->status = (int16_t)size;
|
|
214 |
if (st < 0)
|
|
215 |
resp->status = (int16_t)st;
|
|
216 |
|
|
217 |
xen_be_printf(&netdev->xendev, 3, "rx response: idx %d, status %d, flags 0x%x\n",
|
|
218 |
i, resp->status, resp->flags);
|
|
219 |
|
|
220 |
netdev->rx_ring.rsp_prod_pvt = ++i;
|
|
221 |
RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&netdev->rx_ring, notify);
|
|
222 |
if (notify)
|
|
223 |
xen_be_send_notify(&netdev->xendev);
|
|
224 |
}
|
|
225 |
|
|
226 |
#define NET_IP_ALIGN 2
|
|
227 |
|
|
228 |
static int net_rx_ok(void *opaque)
|
|
229 |
{
|
|
230 |
struct XenNetDev *netdev = opaque;
|
|
231 |
RING_IDX rc, rp;
|
|
232 |
|
|
233 |
if (netdev->xendev.be_state != XenbusStateConnected)
|
|
234 |
return 0;
|
|
235 |
|
|
236 |
rc = netdev->rx_ring.req_cons;
|
|
237 |
rp = netdev->rx_ring.sring->req_prod;
|
|
238 |
xen_rmb();
|
|
239 |
|
|
240 |
if (rc == rp || RING_REQUEST_CONS_OVERFLOW(&netdev->rx_ring, rc)) {
|
|
241 |
xen_be_printf(&netdev->xendev, 2, "%s: no rx buffers (%d/%d)\n",
|
|
242 |
__FUNCTION__, rc, rp);
|
|
243 |
return 0;
|
|
244 |
}
|
|
245 |
return 1;
|
|
246 |
}
|
|
247 |
|
|
248 |
static void net_rx_packet(void *opaque, const uint8_t *buf, int size)
|
|
249 |
{
|
|
250 |
struct XenNetDev *netdev = opaque;
|
|
251 |
netif_rx_request_t rxreq;
|
|
252 |
RING_IDX rc, rp;
|
|
253 |
void *page;
|
|
254 |
|
|
255 |
if (netdev->xendev.be_state != XenbusStateConnected)
|
|
256 |
return;
|
|
257 |
|
|
258 |
rc = netdev->rx_ring.req_cons;
|
|
259 |
rp = netdev->rx_ring.sring->req_prod;
|
|
260 |
xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
|
|
261 |
|
|
262 |
if (rc == rp || RING_REQUEST_CONS_OVERFLOW(&netdev->rx_ring, rc)) {
|
|
263 |
xen_be_printf(&netdev->xendev, 2, "no buffer, drop packet\n");
|
|
264 |
return;
|
|
265 |
}
|
|
266 |
if (size > XC_PAGE_SIZE - NET_IP_ALIGN) {
|
|
267 |
xen_be_printf(&netdev->xendev, 0, "packet too big (%d > %ld)",
|
|
268 |
size, XC_PAGE_SIZE - NET_IP_ALIGN);
|
|
269 |
return;
|
|
270 |
}
|
|
271 |
|
|
272 |
memcpy(&rxreq, RING_GET_REQUEST(&netdev->rx_ring, rc), sizeof(rxreq));
|
|
273 |
netdev->rx_ring.req_cons = ++rc;
|
|
274 |
|
|
275 |
page = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
|
|
276 |
netdev->xendev.dom,
|
|
277 |
rxreq.gref, PROT_WRITE);
|
|
278 |
if (page == NULL) {
|
|
279 |
xen_be_printf(&netdev->xendev, 0, "error: rx gref dereference failed (%d)\n",
|
|
280 |
rxreq.gref);
|
|
281 |
net_rx_response(netdev, &rxreq, NETIF_RSP_ERROR, 0, 0, 0);
|
|
282 |
return;
|
|
283 |
}
|
|
284 |
memcpy(page + NET_IP_ALIGN, buf, size);
|
|
285 |
xc_gnttab_munmap(netdev->xendev.gnttabdev, page, 1);
|
|
286 |
net_rx_response(netdev, &rxreq, NETIF_RSP_OKAY, NET_IP_ALIGN, size, 0);
|
|
287 |
}
|
|
288 |
|
|
289 |
/* ------------------------------------------------------------- */
|
|
290 |
|
|
291 |
static int net_init(struct XenDevice *xendev)
|
|
292 |
{
|
|
293 |
struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
|
|
294 |
VLANState *vlan;
|
|
295 |
|
|
296 |
/* read xenstore entries */
|
|
297 |
if (netdev->mac == NULL)
|
|
298 |
netdev->mac = xenstore_read_be_str(&netdev->xendev, "mac");
|
|
299 |
|
|
300 |
/* do we have all we need? */
|
|
301 |
if (netdev->mac == NULL)
|
|
302 |
return -1;
|
|
303 |
|
|
304 |
vlan = qemu_find_vlan(netdev->xendev.dev);
|
|
305 |
netdev->vs = qemu_new_vlan_client(vlan, "xen", NULL,
|
|
306 |
net_rx_packet, net_rx_ok, NULL,
|
|
307 |
netdev);
|
|
308 |
snprintf(netdev->vs->info_str, sizeof(netdev->vs->info_str),
|
|
309 |
"nic: xenbus vif macaddr=%s", netdev->mac);
|
|
310 |
|
|
311 |
/* fill info */
|
|
312 |
xenstore_write_be_int(&netdev->xendev, "feature-rx-copy", 1);
|
|
313 |
xenstore_write_be_int(&netdev->xendev, "feature-rx-flip", 0);
|
|
314 |
|
|
315 |
return 0;
|
|
316 |
}
|
|
317 |
|
|
318 |
static int net_connect(struct XenDevice *xendev)
|
|
319 |
{
|
|
320 |
struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
|
|
321 |
int rx_copy;
|
|
322 |
|
|
323 |
if (xenstore_read_fe_int(&netdev->xendev, "tx-ring-ref",
|
|
324 |
&netdev->tx_ring_ref) == -1)
|
|
325 |
return -1;
|
|
326 |
if (xenstore_read_fe_int(&netdev->xendev, "rx-ring-ref",
|
|
327 |
&netdev->rx_ring_ref) == -1)
|
|
328 |
return 1;
|
|
329 |
if (xenstore_read_fe_int(&netdev->xendev, "event-channel",
|
|
330 |
&netdev->xendev.remote_port) == -1)
|
|
331 |
return -1;
|
|
332 |
|
|
333 |
if (xenstore_read_fe_int(&netdev->xendev, "request-rx-copy", &rx_copy) == -1)
|
|
334 |
rx_copy = 0;
|
|
335 |
if (rx_copy == 0) {
|
|
336 |
xen_be_printf(&netdev->xendev, 0, "frontend doesn't support rx-copy.\n");
|
|
337 |
return -1;
|
|
338 |
}
|
|
339 |
|
|
340 |
netdev->txs = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
|
|
341 |
netdev->xendev.dom,
|
|
342 |
netdev->tx_ring_ref,
|
|
343 |
PROT_READ | PROT_WRITE);
|
|
344 |
netdev->rxs = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
|
|
345 |
netdev->xendev.dom,
|
|
346 |
netdev->rx_ring_ref,
|
|
347 |
PROT_READ | PROT_WRITE);
|
|
348 |
if (!netdev->txs || !netdev->rxs)
|
|
349 |
return -1;
|
|
350 |
BACK_RING_INIT(&netdev->tx_ring, netdev->txs, XC_PAGE_SIZE);
|
|
351 |
BACK_RING_INIT(&netdev->rx_ring, netdev->rxs, XC_PAGE_SIZE);
|
|
352 |
|
|
353 |
xen_be_bind_evtchn(&netdev->xendev);
|
|
354 |
|
|
355 |
xen_be_printf(&netdev->xendev, 1, "ok: tx-ring-ref %d, rx-ring-ref %d, "
|
|
356 |
"remote port %d, local port %d\n",
|
|
357 |
netdev->tx_ring_ref, netdev->rx_ring_ref,
|
|
358 |
netdev->xendev.remote_port, netdev->xendev.local_port);
|
|
359 |
return 0;
|
|
360 |
}
|
|
361 |
|
|
362 |
static void net_disconnect(struct XenDevice *xendev)
|
|
363 |
{
|
|
364 |
struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
|
|
365 |
|
|
366 |
xen_be_unbind_evtchn(&netdev->xendev);
|
|
367 |
|
|
368 |
if (netdev->txs) {
|
|
369 |
xc_gnttab_munmap(netdev->xendev.gnttabdev, netdev->txs, 1);
|
|
370 |
netdev->txs = NULL;
|
|
371 |
}
|
|
372 |
if (netdev->rxs) {
|
|
373 |
xc_gnttab_munmap(netdev->xendev.gnttabdev, netdev->rxs, 1);
|
|
374 |
netdev->rxs = NULL;
|
|
375 |
}
|
|
376 |
if (netdev->vs) {
|
|
377 |
qemu_del_vlan_client(netdev->vs);
|
|
378 |
netdev->vs = NULL;
|
|
379 |
}
|
|
380 |
}
|
|
381 |
|
|
382 |
static void net_event(struct XenDevice *xendev)
|
|
383 |
{
|
|
384 |
struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
|
|
385 |
net_tx_packets(netdev);
|
|
386 |
}
|
|
387 |
|
|
388 |
static int net_free(struct XenDevice *xendev)
|
|
389 |
{
|
|
390 |
struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
|
|
391 |
|
|
392 |
qemu_free(netdev->mac);
|
|
393 |
return 0;
|
|
394 |
}
|
|
395 |
|
|
396 |
/* ------------------------------------------------------------- */
|
|
397 |
|
|
398 |
struct XenDevOps xen_netdev_ops = {
|
|
399 |
.size = sizeof(struct XenNetDev),
|
|
400 |
.flags = DEVOPS_FLAG_NEED_GNTDEV,
|
|
401 |
.init = net_init,
|
|
402 |
.connect = net_connect,
|
|
403 |
.event = net_event,
|
|
404 |
.disconnect = net_disconnect,
|
|
405 |
.free = net_free,
|
|
406 |
};
|