]> git.proxmox.com Git - mirror_qemu.git/blame - net/tap.c
nbd sockets vnc: Mark problematic address family tests TODO
[mirror_qemu.git] / net / tap.c
CommitLineData
5281d757
MM
1/*
2 * QEMU System Emulator
3 *
4 * Copyright (c) 2003-2008 Fabrice Bellard
5 * Copyright (c) 2009 Red Hat, Inc.
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
13 *
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
23 * THE SOFTWARE.
24 */
25
2744d920 26#include "qemu/osdep.h"
1422e32d 27#include "tap_int.h"
5281d757 28
5281d757 29
5281d757 30#include <sys/ioctl.h>
5281d757 31#include <sys/wait.h>
71f4effc 32#include <sys/socket.h>
5281d757
MM
33#include <net/if.h>
34
1422e32d 35#include "net/net.h"
a245fc18 36#include "clients.h"
83c9089e 37#include "monitor/monitor.h"
9c17d615 38#include "sysemu/sysemu.h"
da34e65c 39#include "qapi/error.h"
5281d757 40#include "qemu-common.h"
f348b6d1 41#include "qemu/cutils.h"
1de7afc9 42#include "qemu/error-report.h"
5281d757 43
1422e32d 44#include "net/tap.h"
5281d757 45
0d09e41a 46#include "net/vhost_net.h"
82b0d80e 47
5281d757 48typedef struct TAPState {
4e68f7a0 49 NetClientState nc;
5281d757
MM
50 int fd;
51 char down_script[1024];
52 char down_script_arg[128];
d32fcad3 53 uint8_t buf[NET_BUFSIZE];
ec45f083
JW
54 bool read_poll;
55 bool write_poll;
56 bool using_vnet_hdr;
57 bool has_ufo;
16dbaf90 58 bool enabled;
82b0d80e 59 VHostNetState *vhost_net;
ef4252b1 60 unsigned host_vnet_hdr_len;
9e32ff32 61 Notifier exit;
5281d757
MM
62} TAPState;
63
ac4fcf56
MA
64static void launch_script(const char *setup_script, const char *ifname,
65 int fd, Error **errp);
5281d757 66
5281d757
MM
67static void tap_send(void *opaque);
68static void tap_writable(void *opaque);
69
70static void tap_update_fd_handler(TAPState *s)
71{
82e1cc4b
FZ
72 qemu_set_fd_handler(s->fd,
73 s->read_poll && s->enabled ? tap_send : NULL,
74 s->write_poll && s->enabled ? tap_writable : NULL,
75 s);
5281d757
MM
76}
77
ec45f083 78static void tap_read_poll(TAPState *s, bool enable)
5281d757 79{
ec45f083 80 s->read_poll = enable;
5281d757
MM
81 tap_update_fd_handler(s);
82}
83
ec45f083 84static void tap_write_poll(TAPState *s, bool enable)
5281d757 85{
ec45f083 86 s->write_poll = enable;
5281d757
MM
87 tap_update_fd_handler(s);
88}
89
90static void tap_writable(void *opaque)
91{
92 TAPState *s = opaque;
93
ec45f083 94 tap_write_poll(s, false);
5281d757 95
3e35ba93 96 qemu_flush_queued_packets(&s->nc);
5281d757
MM
97}
98
99static ssize_t tap_write_packet(TAPState *s, const struct iovec *iov, int iovcnt)
100{
101 ssize_t len;
102
103 do {
104 len = writev(s->fd, iov, iovcnt);
105 } while (len == -1 && errno == EINTR);
106
107 if (len == -1 && errno == EAGAIN) {
ec45f083 108 tap_write_poll(s, true);
5281d757
MM
109 return 0;
110 }
111
112 return len;
113}
114
4e68f7a0 115static ssize_t tap_receive_iov(NetClientState *nc, const struct iovec *iov,
5281d757
MM
116 int iovcnt)
117{
3e35ba93 118 TAPState *s = DO_UPCAST(TAPState, nc, nc);
5281d757
MM
119 const struct iovec *iovp = iov;
120 struct iovec iov_copy[iovcnt + 1];
ef4252b1 121 struct virtio_net_hdr_mrg_rxbuf hdr = { };
5281d757 122
ef4252b1 123 if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
5281d757 124 iov_copy[0].iov_base = &hdr;
ef4252b1 125 iov_copy[0].iov_len = s->host_vnet_hdr_len;
5281d757
MM
126 memcpy(&iov_copy[1], iov, iovcnt * sizeof(*iov));
127 iovp = iov_copy;
128 iovcnt++;
129 }
130
131 return tap_write_packet(s, iovp, iovcnt);
132}
133
4e68f7a0 134static ssize_t tap_receive_raw(NetClientState *nc, const uint8_t *buf, size_t size)
5281d757 135{
3e35ba93 136 TAPState *s = DO_UPCAST(TAPState, nc, nc);
5281d757
MM
137 struct iovec iov[2];
138 int iovcnt = 0;
ef4252b1 139 struct virtio_net_hdr_mrg_rxbuf hdr = { };
5281d757 140
ef4252b1 141 if (s->host_vnet_hdr_len) {
5281d757 142 iov[iovcnt].iov_base = &hdr;
ef4252b1 143 iov[iovcnt].iov_len = s->host_vnet_hdr_len;
5281d757
MM
144 iovcnt++;
145 }
146
147 iov[iovcnt].iov_base = (char *)buf;
148 iov[iovcnt].iov_len = size;
149 iovcnt++;
150
151 return tap_write_packet(s, iov, iovcnt);
152}
153
4e68f7a0 154static ssize_t tap_receive(NetClientState *nc, const uint8_t *buf, size_t size)
5281d757 155{
3e35ba93 156 TAPState *s = DO_UPCAST(TAPState, nc, nc);
5281d757
MM
157 struct iovec iov[1];
158
ef4252b1 159 if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
3e35ba93 160 return tap_receive_raw(nc, buf, size);
5281d757
MM
161 }
162
163 iov[0].iov_base = (char *)buf;
164 iov[0].iov_len = size;
165
166 return tap_write_packet(s, iov, 1);
167}
168
966ea5ec
MM
169#ifndef __sun__
170ssize_t tap_read_packet(int tapfd, uint8_t *buf, int maxlen)
5281d757
MM
171{
172 return read(tapfd, buf, maxlen);
173}
174#endif
175
4e68f7a0 176static void tap_send_completed(NetClientState *nc, ssize_t len)
5281d757 177{
3e35ba93 178 TAPState *s = DO_UPCAST(TAPState, nc, nc);
ec45f083 179 tap_read_poll(s, true);
5281d757
MM
180}
181
182static void tap_send(void *opaque)
183{
184 TAPState *s = opaque;
185 int size;
756ae78b 186 int packets = 0;
5281d757 187
a90a7425 188 while (true) {
5819c918
MM
189 uint8_t *buf = s->buf;
190
191 size = tap_read_packet(s->fd, s->buf, sizeof(s->buf));
192 if (size <= 0) {
193 break;
194 }
195
ef4252b1
MT
196 if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
197 buf += s->host_vnet_hdr_len;
198 size -= s->host_vnet_hdr_len;
5819c918
MM
199 }
200
3e35ba93 201 size = qemu_send_packet_async(&s->nc, buf, size, tap_send_completed);
5819c918 202 if (size == 0) {
ec45f083 203 tap_read_poll(s, false);
68e5ec64
SH
204 break;
205 } else if (size < 0) {
206 break;
5819c918 207 }
756ae78b
WK
208
209 /*
210 * When the host keeps receiving more packets while tap_send() is
211 * running we can hog the QEMU global mutex. Limit the number of
212 * packets that are processed per tap_send() callback to prevent
213 * stalling the guest.
214 */
215 packets++;
216 if (packets >= 50) {
217 break;
218 }
68e5ec64 219 }
5281d757
MM
220}
221
3bac80d3 222static bool tap_has_ufo(NetClientState *nc)
5281d757 223{
3e35ba93 224 TAPState *s = DO_UPCAST(TAPState, nc, nc);
5281d757 225
f394b2e2 226 assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
5281d757
MM
227
228 return s->has_ufo;
229}
230
3bac80d3 231static bool tap_has_vnet_hdr(NetClientState *nc)
5281d757 232{
3e35ba93 233 TAPState *s = DO_UPCAST(TAPState, nc, nc);
5281d757 234
f394b2e2 235 assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
5281d757 236
ef4252b1 237 return !!s->host_vnet_hdr_len;
5281d757
MM
238}
239
3bac80d3 240static bool tap_has_vnet_hdr_len(NetClientState *nc, int len)
445d892f
MT
241{
242 TAPState *s = DO_UPCAST(TAPState, nc, nc);
243
f394b2e2 244 assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
445d892f 245
e96dfd11 246 return !!tap_probe_vnet_hdr_len(s->fd, len);
445d892f
MT
247}
248
3bac80d3 249static void tap_set_vnet_hdr_len(NetClientState *nc, int len)
445d892f
MT
250{
251 TAPState *s = DO_UPCAST(TAPState, nc, nc);
252
f394b2e2 253 assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
445d892f
MT
254 assert(len == sizeof(struct virtio_net_hdr_mrg_rxbuf) ||
255 len == sizeof(struct virtio_net_hdr));
256
257 tap_fd_set_vnet_hdr_len(s->fd, len);
258 s->host_vnet_hdr_len = len;
259}
260
3bac80d3 261static void tap_using_vnet_hdr(NetClientState *nc, bool using_vnet_hdr)
5281d757 262{
3e35ba93 263 TAPState *s = DO_UPCAST(TAPState, nc, nc);
5281d757 264
f394b2e2 265 assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
ef4252b1 266 assert(!!s->host_vnet_hdr_len == using_vnet_hdr);
5281d757
MM
267
268 s->using_vnet_hdr = using_vnet_hdr;
269}
270
c80cd6bb
GK
271static int tap_set_vnet_le(NetClientState *nc, bool is_le)
272{
273 TAPState *s = DO_UPCAST(TAPState, nc, nc);
274
275 return tap_fd_set_vnet_le(s->fd, is_le);
276}
277
278static int tap_set_vnet_be(NetClientState *nc, bool is_be)
279{
280 TAPState *s = DO_UPCAST(TAPState, nc, nc);
281
282 return tap_fd_set_vnet_be(s->fd, is_be);
283}
284
3bac80d3 285static void tap_set_offload(NetClientState *nc, int csum, int tso4,
5281d757
MM
286 int tso6, int ecn, int ufo)
287{
3e35ba93 288 TAPState *s = DO_UPCAST(TAPState, nc, nc);
27a6375d
MT
289 if (s->fd < 0) {
290 return;
291 }
5281d757 292
27a6375d 293 tap_fd_set_offload(s->fd, csum, tso4, tso6, ecn, ufo);
5281d757
MM
294}
295
9e32ff32
MAL
296static void tap_exit_notify(Notifier *notifier, void *data)
297{
298 TAPState *s = container_of(notifier, TAPState, exit);
299 Error *err = NULL;
300
301 if (s->down_script[0]) {
302 launch_script(s->down_script, s->down_script_arg, s->fd, &err);
303 if (err) {
304 error_report_err(err);
305 }
306 }
307}
308
4e68f7a0 309static void tap_cleanup(NetClientState *nc)
5281d757 310{
3e35ba93 311 TAPState *s = DO_UPCAST(TAPState, nc, nc);
5281d757 312
82b0d80e
MT
313 if (s->vhost_net) {
314 vhost_net_cleanup(s->vhost_net);
e6bcb1b6 315 g_free(s->vhost_net);
43849424 316 s->vhost_net = NULL;
82b0d80e
MT
317 }
318
3e35ba93 319 qemu_purge_queued_packets(nc);
5281d757 320
9e32ff32
MAL
321 tap_exit_notify(&s->exit, NULL);
322 qemu_remove_exit_notifier(&s->exit);
5281d757 323
ec45f083
JW
324 tap_read_poll(s, false);
325 tap_write_poll(s, false);
5281d757 326 close(s->fd);
27a6375d 327 s->fd = -1;
5281d757
MM
328}
329
4e68f7a0 330static void tap_poll(NetClientState *nc, bool enable)
ceb69615
MT
331{
332 TAPState *s = DO_UPCAST(TAPState, nc, nc);
333 tap_read_poll(s, enable);
334 tap_write_poll(s, enable);
335}
336
4e68f7a0 337int tap_get_fd(NetClientState *nc)
95d528a2
MT
338{
339 TAPState *s = DO_UPCAST(TAPState, nc, nc);
f394b2e2 340 assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
95d528a2
MT
341 return s->fd;
342}
343
5281d757
MM
344/* fd support */
345
3e35ba93 346static NetClientInfo net_tap_info = {
f394b2e2 347 .type = NET_CLIENT_DRIVER_TAP,
3e35ba93
MM
348 .size = sizeof(TAPState),
349 .receive = tap_receive,
350 .receive_raw = tap_receive_raw,
351 .receive_iov = tap_receive_iov,
ceb69615 352 .poll = tap_poll,
3e35ba93 353 .cleanup = tap_cleanup,
2e753bcc
VM
354 .has_ufo = tap_has_ufo,
355 .has_vnet_hdr = tap_has_vnet_hdr,
356 .has_vnet_hdr_len = tap_has_vnet_hdr_len,
357 .using_vnet_hdr = tap_using_vnet_hdr,
358 .set_offload = tap_set_offload,
359 .set_vnet_hdr_len = tap_set_vnet_hdr_len,
c80cd6bb
GK
360 .set_vnet_le = tap_set_vnet_le,
361 .set_vnet_be = tap_set_vnet_be,
3e35ba93
MM
362};
363
4e68f7a0 364static TAPState *net_tap_fd_init(NetClientState *peer,
5281d757
MM
365 const char *model,
366 const char *name,
367 int fd,
368 int vnet_hdr)
369{
4e68f7a0 370 NetClientState *nc;
5281d757 371 TAPState *s;
5281d757 372
ab5f3f84 373 nc = qemu_new_net_client(&net_tap_info, peer, model, name);
3e35ba93
MM
374
375 s = DO_UPCAST(TAPState, nc, nc);
376
5281d757 377 s->fd = fd;
ef4252b1 378 s->host_vnet_hdr_len = vnet_hdr ? sizeof(struct virtio_net_hdr) : 0;
ec45f083 379 s->using_vnet_hdr = false;
9c282718 380 s->has_ufo = tap_probe_has_ufo(s->fd);
16dbaf90 381 s->enabled = true;
3e35ba93 382 tap_set_offload(&s->nc, 0, 0, 0, 0, 0);
58ddcd50
MT
383 /*
384 * Make sure host header length is set correctly in tap:
385 * it might have been modified by another instance of qemu.
386 */
387 if (tap_probe_vnet_hdr_len(s->fd, s->host_vnet_hdr_len)) {
388 tap_fd_set_vnet_hdr_len(s->fd, s->host_vnet_hdr_len);
389 }
ec45f083 390 tap_read_poll(s, true);
82b0d80e 391 s->vhost_net = NULL;
9e32ff32
MAL
392
393 s->exit.notify = tap_exit_notify;
394 qemu_add_exit_notifier(&s->exit);
395
5281d757
MM
396 return s;
397}
398
ac4fcf56
MA
399static void launch_script(const char *setup_script, const char *ifname,
400 int fd, Error **errp)
5281d757 401{
5281d757
MM
402 int pid, status;
403 char *args[3];
404 char **parg;
405
5281d757
MM
406 /* try to launch network script */
407 pid = fork();
ac4fcf56
MA
408 if (pid < 0) {
409 error_setg_errno(errp, errno, "could not launch network script %s",
410 setup_script);
411 return;
412 }
5281d757
MM
413 if (pid == 0) {
414 int open_max = sysconf(_SC_OPEN_MAX), i;
415
13a12f86
PG
416 for (i = 3; i < open_max; i++) {
417 if (i != fd) {
5281d757
MM
418 close(i);
419 }
420 }
421 parg = args;
422 *parg++ = (char *)setup_script;
423 *parg++ = (char *)ifname;
9678d950 424 *parg = NULL;
5281d757
MM
425 execv(setup_script, args);
426 _exit(1);
ac4fcf56 427 } else {
5281d757
MM
428 while (waitpid(pid, &status, 0) != pid) {
429 /* loop */
430 }
5281d757
MM
431
432 if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
ac4fcf56 433 return;
5281d757 434 }
ac4fcf56
MA
435 error_setg(errp, "network script %s failed with status %d",
436 setup_script, status);
5281d757 437 }
5281d757
MM
438}
439
a7c36ee4
CB
440static int recv_fd(int c)
441{
442 int fd;
443 uint8_t msgbuf[CMSG_SPACE(sizeof(fd))];
444 struct msghdr msg = {
445 .msg_control = msgbuf,
446 .msg_controllen = sizeof(msgbuf),
447 };
448 struct cmsghdr *cmsg;
449 struct iovec iov;
450 uint8_t req[1];
451 ssize_t len;
452
453 cmsg = CMSG_FIRSTHDR(&msg);
454 cmsg->cmsg_level = SOL_SOCKET;
455 cmsg->cmsg_type = SCM_RIGHTS;
456 cmsg->cmsg_len = CMSG_LEN(sizeof(fd));
457 msg.msg_controllen = cmsg->cmsg_len;
458
459 iov.iov_base = req;
460 iov.iov_len = sizeof(req);
461
462 msg.msg_iov = &iov;
463 msg.msg_iovlen = 1;
464
465 len = recvmsg(c, &msg, 0);
466 if (len > 0) {
467 memcpy(&fd, CMSG_DATA(cmsg), sizeof(fd));
468 return fd;
469 }
470
471 return len;
472}
473
a8a21be9
MA
474static int net_bridge_run_helper(const char *helper, const char *bridge,
475 Error **errp)
a7c36ee4
CB
476{
477 sigset_t oldmask, mask;
478 int pid, status;
479 char *args[5];
480 char **parg;
481 int sv[2];
482
483 sigemptyset(&mask);
484 sigaddset(&mask, SIGCHLD);
485 sigprocmask(SIG_BLOCK, &mask, &oldmask);
486
487 if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) {
a8a21be9 488 error_setg_errno(errp, errno, "socketpair() failed");
a7c36ee4
CB
489 return -1;
490 }
491
492 /* try to launch bridge helper */
493 pid = fork();
a8a21be9
MA
494 if (pid < 0) {
495 error_setg_errno(errp, errno, "Can't fork bridge helper");
496 return -1;
497 }
a7c36ee4
CB
498 if (pid == 0) {
499 int open_max = sysconf(_SC_OPEN_MAX), i;
500 char fd_buf[6+10];
501 char br_buf[6+IFNAMSIZ] = {0};
502 char helper_cmd[PATH_MAX + sizeof(fd_buf) + sizeof(br_buf) + 15];
503
13a12f86
PG
504 for (i = 3; i < open_max; i++) {
505 if (i != sv[1]) {
a7c36ee4
CB
506 close(i);
507 }
508 }
509
510 snprintf(fd_buf, sizeof(fd_buf), "%s%d", "--fd=", sv[1]);
511
512 if (strrchr(helper, ' ') || strrchr(helper, '\t')) {
513 /* assume helper is a command */
514
515 if (strstr(helper, "--br=") == NULL) {
516 snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge);
517 }
518
519 snprintf(helper_cmd, sizeof(helper_cmd), "%s %s %s %s",
520 helper, "--use-vnet", fd_buf, br_buf);
521
522 parg = args;
523 *parg++ = (char *)"sh";
524 *parg++ = (char *)"-c";
525 *parg++ = helper_cmd;
526 *parg++ = NULL;
527
528 execv("/bin/sh", args);
529 } else {
530 /* assume helper is just the executable path name */
531
532 snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge);
533
534 parg = args;
535 *parg++ = (char *)helper;
536 *parg++ = (char *)"--use-vnet";
537 *parg++ = fd_buf;
538 *parg++ = br_buf;
539 *parg++ = NULL;
540
541 execv(helper, args);
542 }
543 _exit(1);
544
a8a21be9 545 } else {
a7c36ee4 546 int fd;
a8a21be9 547 int saved_errno;
a7c36ee4
CB
548
549 close(sv[1]);
550
551 do {
552 fd = recv_fd(sv[0]);
553 } while (fd == -1 && errno == EINTR);
a8a21be9 554 saved_errno = errno;
a7c36ee4
CB
555
556 close(sv[0]);
557
558 while (waitpid(pid, &status, 0) != pid) {
559 /* loop */
560 }
561 sigprocmask(SIG_SETMASK, &oldmask, NULL);
562 if (fd < 0) {
a8a21be9
MA
563 error_setg_errno(errp, saved_errno,
564 "failed to recv file descriptor");
a7c36ee4
CB
565 return -1;
566 }
a8a21be9
MA
567 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
568 error_setg(errp, "bridge helper failed");
569 return -1;
a7c36ee4 570 }
a8a21be9 571 return fd;
a7c36ee4 572 }
a7c36ee4
CB
573}
574
cebea510 575int net_init_bridge(const Netdev *netdev, const char *name,
a30ecde6 576 NetClientState *peer, Error **errp)
a7c36ee4 577{
f79b51b0
LE
578 const NetdevBridgeOptions *bridge;
579 const char *helper, *br;
a7c36ee4
CB
580 TAPState *s;
581 int fd, vnet_hdr;
582
f394b2e2
EB
583 assert(netdev->type == NET_CLIENT_DRIVER_BRIDGE);
584 bridge = &netdev->u.bridge;
f79b51b0
LE
585
586 helper = bridge->has_helper ? bridge->helper : DEFAULT_BRIDGE_HELPER;
587 br = bridge->has_br ? bridge->br : DEFAULT_BRIDGE_INTERFACE;
a7c36ee4 588
a8a21be9 589 fd = net_bridge_run_helper(helper, br, errp);
a7c36ee4
CB
590 if (fd == -1) {
591 return -1;
592 }
593
594 fcntl(fd, F_SETFL, O_NONBLOCK);
a7c36ee4 595 vnet_hdr = tap_probe_vnet_hdr(fd);
d33d93b2 596 s = net_tap_fd_init(peer, "bridge", name, fd, vnet_hdr);
a7c36ee4 597
f79b51b0
LE
598 snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s,br=%s", helper,
599 br);
a7c36ee4
CB
600
601 return 0;
602}
603
08c573a8
LE
604static int net_tap_init(const NetdevTapOptions *tap, int *vnet_hdr,
605 const char *setup_script, char *ifname,
468dd824 606 size_t ifname_sz, int mq_required, Error **errp)
5281d757 607{
ac4fcf56 608 Error *err = NULL;
5281d757 609 int fd, vnet_hdr_required;
5281d757 610
08c573a8
LE
611 if (tap->has_vnet_hdr) {
612 *vnet_hdr = tap->vnet_hdr;
5281d757
MM
613 vnet_hdr_required = *vnet_hdr;
614 } else {
08c573a8 615 *vnet_hdr = 1;
5281d757
MM
616 vnet_hdr_required = 0;
617 }
618
264986e2 619 TFR(fd = tap_open(ifname, ifname_sz, vnet_hdr, vnet_hdr_required,
468dd824 620 mq_required, errp));
5281d757
MM
621 if (fd < 0) {
622 return -1;
623 }
624
5281d757
MM
625 if (setup_script &&
626 setup_script[0] != '\0' &&
ac4fcf56
MA
627 strcmp(setup_script, "no") != 0) {
628 launch_script(setup_script, ifname, fd, &err);
629 if (err) {
468dd824 630 error_propagate(errp, err);
ac4fcf56
MA
631 close(fd);
632 return -1;
633 }
5281d757
MM
634 }
635
5281d757
MM
636 return fd;
637}
638
264986e2
JW
639#define MAX_TAP_QUEUES 1024
640
445f116c
MA
641static void net_init_tap_one(const NetdevTapOptions *tap, NetClientState *peer,
642 const char *model, const char *name,
643 const char *ifname, const char *script,
644 const char *downscript, const char *vhostfdname,
645 int vnet_hdr, int fd, Error **errp)
5193e5fb 646{
1677f4c6 647 Error *err = NULL;
da4a4eac 648 TAPState *s = net_tap_fd_init(peer, model, name, fd, vnet_hdr);
81647a65 649 int vhostfd;
5193e5fb 650
80b832c3
MA
651 tap_set_sndbuf(s->fd, tap, &err);
652 if (err) {
445f116c
MA
653 error_propagate(errp, err);
654 return;
5193e5fb
JW
655 }
656
264986e2 657 if (tap->has_fd || tap->has_fds) {
5193e5fb
JW
658 snprintf(s->nc.info_str, sizeof(s->nc.info_str), "fd=%d", fd);
659 } else if (tap->has_helper) {
660 snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s",
661 tap->helper);
662 } else {
5193e5fb
JW
663 snprintf(s->nc.info_str, sizeof(s->nc.info_str),
664 "ifname=%s,script=%s,downscript=%s", ifname, script,
665 downscript);
666
667 if (strcmp(downscript, "no") != 0) {
668 snprintf(s->down_script, sizeof(s->down_script), "%s", downscript);
669 snprintf(s->down_script_arg, sizeof(s->down_script_arg),
670 "%s", ifname);
671 }
672 }
673
674 if (tap->has_vhost ? tap->vhost :
675 vhostfdname || (tap->has_vhostforce && tap->vhostforce)) {
81647a65
NN
676 VhostNetOptions options;
677
1a1bfac9 678 options.backend_type = VHOST_BACKEND_TYPE_KERNEL;
81647a65 679 options.net_backend = &s->nc;
69e87b32
JW
680 if (tap->has_poll_us) {
681 options.busyloop_timeout = tap->poll_us;
682 } else {
683 options.busyloop_timeout = 0;
684 }
5193e5fb 685
3a2d44f6 686 if (vhostfdname) {
1677f4c6 687 vhostfd = monitor_fd_param(cur_mon, vhostfdname, &err);
5193e5fb 688 if (vhostfd == -1) {
445f116c
MA
689 error_propagate(errp, err);
690 return;
5193e5fb
JW
691 }
692 } else {
81647a65
NN
693 vhostfd = open("/dev/vhost-net", O_RDWR);
694 if (vhostfd < 0) {
445f116c
MA
695 error_setg_errno(errp, errno,
696 "tap: open vhost char device failed");
697 return;
81647a65 698 }
c471ad0e 699 fcntl(vhostfd, F_SETFL, O_NONBLOCK);
5193e5fb 700 }
81647a65 701 options.opaque = (void *)(uintptr_t)vhostfd;
5193e5fb 702
81647a65 703 s->vhost_net = vhost_net_init(&options);
5193e5fb 704 if (!s->vhost_net) {
445f116c
MA
705 error_setg(errp,
706 "vhost-net requested but could not be initialized");
707 return;
5193e5fb 708 }
3a2d44f6 709 } else if (vhostfdname) {
69e87b32 710 error_setg(errp, "vhostfd(s)= is not valid without vhost");
5193e5fb 711 }
5193e5fb
JW
712}
713
264986e2
JW
714static int get_fds(char *str, char *fds[], int max)
715{
716 char *ptr = str, *this;
717 size_t len = strlen(str);
718 int i = 0;
719
720 while (i < max && ptr < str + len) {
721 this = strchr(ptr, ':');
722
723 if (this == NULL) {
724 fds[i] = g_strdup(ptr);
725 } else {
726 fds[i] = g_strndup(ptr, this - ptr);
727 }
728
729 i++;
730 if (this == NULL) {
731 break;
732 } else {
733 ptr = this + 1;
734 }
735 }
736
737 return i;
738}
739
cebea510 740int net_init_tap(const Netdev *netdev, const char *name,
a30ecde6 741 NetClientState *peer, Error **errp)
5281d757 742{
08c573a8 743 const NetdevTapOptions *tap;
264986e2 744 int fd, vnet_hdr = 0, i = 0, queues;
08c573a8
LE
745 /* for the no-fd, no-helper case */
746 const char *script = NULL; /* suppress wrong "uninit'd use" gcc warning */
5193e5fb 747 const char *downscript = NULL;
1677f4c6 748 Error *err = NULL;
264986e2 749 const char *vhostfdname;
08c573a8
LE
750 char ifname[128];
751
f394b2e2
EB
752 assert(netdev->type == NET_CLIENT_DRIVER_TAP);
753 tap = &netdev->u.tap;
264986e2
JW
754 queues = tap->has_queues ? tap->queues : 1;
755 vhostfdname = tap->has_vhostfd ? tap->vhostfd : NULL;
5281d757 756
ce675a75
JW
757 /* QEMU vlans does not support multiqueue tap, in this case peer is set.
758 * For -netdev, peer is always NULL. */
759 if (peer && (tap->has_queues || tap->has_fds || tap->has_vhostfds)) {
a3088177 760 error_setg(errp, "Multiqueue tap cannot be used with QEMU vlans");
ce675a75
JW
761 return -1;
762 }
763
08c573a8
LE
764 if (tap->has_fd) {
765 if (tap->has_ifname || tap->has_script || tap->has_downscript ||
264986e2 766 tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
c87826a8 767 tap->has_fds || tap->has_vhostfds) {
a3088177
MA
768 error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
769 "helper=, queues=, fds=, and vhostfds= "
770 "are invalid with fd=");
5281d757
MM
771 return -1;
772 }
773
1677f4c6 774 fd = monitor_fd_param(cur_mon, tap->fd, &err);
5281d757 775 if (fd == -1) {
a3088177 776 error_propagate(errp, err);
5281d757
MM
777 return -1;
778 }
779
780 fcntl(fd, F_SETFL, O_NONBLOCK);
781
782 vnet_hdr = tap_probe_vnet_hdr(fd);
a7c36ee4 783
445f116c
MA
784 net_init_tap_one(tap, peer, "tap", name, NULL,
785 script, downscript,
786 vhostfdname, vnet_hdr, fd, &err);
787 if (err) {
a3088177 788 error_propagate(errp, err);
264986e2
JW
789 return -1;
790 }
791 } else if (tap->has_fds) {
fac7d7b1
PM
792 char **fds;
793 char **vhost_fds;
264986e2
JW
794 int nfds, nvhosts;
795
796 if (tap->has_ifname || tap->has_script || tap->has_downscript ||
797 tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
c87826a8 798 tap->has_vhostfd) {
a3088177
MA
799 error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
800 "helper=, queues=, and vhostfd= "
801 "are invalid with fds=");
264986e2
JW
802 return -1;
803 }
804
fac7d7b1
PM
805 fds = g_new0(char *, MAX_TAP_QUEUES);
806 vhost_fds = g_new0(char *, MAX_TAP_QUEUES);
807
264986e2
JW
808 nfds = get_fds(tap->fds, fds, MAX_TAP_QUEUES);
809 if (tap->has_vhostfds) {
810 nvhosts = get_fds(tap->vhostfds, vhost_fds, MAX_TAP_QUEUES);
811 if (nfds != nvhosts) {
a3088177
MA
812 error_setg(errp, "The number of fds passed does not match "
813 "the number of vhostfds passed");
091a6b2a 814 goto free_fail;
264986e2
JW
815 }
816 }
817
818 for (i = 0; i < nfds; i++) {
1677f4c6 819 fd = monitor_fd_param(cur_mon, fds[i], &err);
264986e2 820 if (fd == -1) {
a3088177 821 error_propagate(errp, err);
091a6b2a 822 goto free_fail;
264986e2
JW
823 }
824
825 fcntl(fd, F_SETFL, O_NONBLOCK);
a7c36ee4 826
264986e2
JW
827 if (i == 0) {
828 vnet_hdr = tap_probe_vnet_hdr(fd);
829 } else if (vnet_hdr != tap_probe_vnet_hdr(fd)) {
a3088177
MA
830 error_setg(errp,
831 "vnet_hdr not consistent across given tap fds");
091a6b2a 832 goto free_fail;
264986e2
JW
833 }
834
445f116c
MA
835 net_init_tap_one(tap, peer, "tap", name, ifname,
836 script, downscript,
837 tap->has_vhostfds ? vhost_fds[i] : NULL,
838 vnet_hdr, fd, &err);
839 if (err) {
a3088177 840 error_propagate(errp, err);
091a6b2a 841 goto free_fail;
264986e2
JW
842 }
843 }
11196e95
ZJ
844 g_free(fds);
845 g_free(vhost_fds);
091a6b2a
PB
846 return 0;
847
848free_fail:
849 for (i = 0; i < nfds; i++) {
850 g_free(fds[i]);
851 g_free(vhost_fds[i]);
852 }
853 g_free(fds);
854 g_free(vhost_fds);
855 return -1;
08c573a8
LE
856 } else if (tap->has_helper) {
857 if (tap->has_ifname || tap->has_script || tap->has_downscript ||
c87826a8 858 tap->has_vnet_hdr || tap->has_queues || tap->has_vhostfds) {
a3088177
MA
859 error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
860 "queues=, and vhostfds= are invalid with helper=");
a7c36ee4
CB
861 return -1;
862 }
863
584613ea
AK
864 fd = net_bridge_run_helper(tap->helper,
865 tap->has_br ?
866 tap->br : DEFAULT_BRIDGE_INTERFACE,
a8a21be9 867 errp);
a7c36ee4
CB
868 if (fd == -1) {
869 return -1;
870 }
871
872 fcntl(fd, F_SETFL, O_NONBLOCK);
a7c36ee4
CB
873 vnet_hdr = tap_probe_vnet_hdr(fd);
874
445f116c
MA
875 net_init_tap_one(tap, peer, "bridge", name, ifname,
876 script, downscript, vhostfdname,
877 vnet_hdr, fd, &err);
878 if (err) {
a3088177 879 error_propagate(errp, err);
84f8f3da 880 close(fd);
264986e2
JW
881 return -1;
882 }
5281d757 883 } else {
c87826a8 884 if (tap->has_vhostfds) {
a3088177 885 error_setg(errp, "vhostfds= is invalid if fds= wasn't specified");
c87826a8
JW
886 return -1;
887 }
08c573a8 888 script = tap->has_script ? tap->script : DEFAULT_NETWORK_SCRIPT;
5193e5fb
JW
889 downscript = tap->has_downscript ? tap->downscript :
890 DEFAULT_NETWORK_DOWN_SCRIPT;
264986e2
JW
891
892 if (tap->has_ifname) {
893 pstrcpy(ifname, sizeof ifname, tap->ifname);
894 } else {
895 ifname[0] = '\0';
929fe497 896 }
a7c36ee4 897
264986e2
JW
898 for (i = 0; i < queues; i++) {
899 fd = net_tap_init(tap, &vnet_hdr, i >= 1 ? "no" : script,
a3088177 900 ifname, sizeof ifname, queues > 1, errp);
264986e2
JW
901 if (fd == -1) {
902 return -1;
903 }
904
905 if (queues > 1 && i == 0 && !tap->has_ifname) {
906 if (tap_fd_get_ifname(fd, ifname)) {
a3088177 907 error_setg(errp, "Fail to get ifname");
84f8f3da 908 close(fd);
264986e2
JW
909 return -1;
910 }
911 }
912
445f116c
MA
913 net_init_tap_one(tap, peer, "tap", name, ifname,
914 i >= 1 ? "no" : script,
915 i >= 1 ? "no" : downscript,
916 vhostfdname, vnet_hdr, fd, &err);
917 if (err) {
a3088177 918 error_propagate(errp, err);
84f8f3da 919 close(fd);
264986e2
JW
920 return -1;
921 }
922 }
5281d757
MM
923 }
924
264986e2 925 return 0;
5281d757 926}
b202554c 927
4e68f7a0 928VHostNetState *tap_get_vhost_net(NetClientState *nc)
b202554c
MT
929{
930 TAPState *s = DO_UPCAST(TAPState, nc, nc);
f394b2e2 931 assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
b202554c
MT
932 return s->vhost_net;
933}
16dbaf90
JW
934
935int tap_enable(NetClientState *nc)
936{
937 TAPState *s = DO_UPCAST(TAPState, nc, nc);
938 int ret;
939
940 if (s->enabled) {
941 return 0;
942 } else {
943 ret = tap_fd_enable(s->fd);
944 if (ret == 0) {
945 s->enabled = true;
946 tap_update_fd_handler(s);
947 }
948 return ret;
949 }
950}
951
952int tap_disable(NetClientState *nc)
953{
954 TAPState *s = DO_UPCAST(TAPState, nc, nc);
955 int ret;
956
957 if (s->enabled == 0) {
958 return 0;
959 } else {
960 ret = tap_fd_disable(s->fd);
961 if (ret == 0) {
962 qemu_purge_queued_packets(nc);
963 s->enabled = false;
964 tap_update_fd_handler(s);
965 }
966 return ret;
967 }
968}