]> git.proxmox.com Git - qemu.git/blob - net/tap.c
tap: reset vnet header size on open
[qemu.git] / net / tap.c
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
26 #include "tap.h"
27
28 #include "config-host.h"
29
30 #include <sys/ioctl.h>
31 #include <sys/stat.h>
32 #include <sys/wait.h>
33 #include <sys/socket.h>
34 #include <net/if.h>
35
36 #include "net.h"
37 #include "clients.h"
38 #include "monitor.h"
39 #include "sysemu.h"
40 #include "qemu-char.h"
41 #include "qemu-common.h"
42 #include "qemu-error.h"
43
44 #include "net/tap-linux.h"
45
46 #include "hw/vhost_net.h"
47
48 /* Maximum GSO packet size (64k) plus plenty of room for
49 * the ethernet and virtio_net headers
50 */
51 #define TAP_BUFSIZE (4096 + 65536)
52
53 typedef struct TAPState {
54 NetClientState nc;
55 int fd;
56 char down_script[1024];
57 char down_script_arg[128];
58 uint8_t buf[TAP_BUFSIZE];
59 unsigned int read_poll : 1;
60 unsigned int write_poll : 1;
61 unsigned int using_vnet_hdr : 1;
62 unsigned int has_ufo: 1;
63 VHostNetState *vhost_net;
64 unsigned host_vnet_hdr_len;
65 } TAPState;
66
67 static int launch_script(const char *setup_script, const char *ifname, int fd);
68
69 static int tap_can_send(void *opaque);
70 static void tap_send(void *opaque);
71 static void tap_writable(void *opaque);
72
73 static void tap_update_fd_handler(TAPState *s)
74 {
75 qemu_set_fd_handler2(s->fd,
76 s->read_poll ? tap_can_send : NULL,
77 s->read_poll ? tap_send : NULL,
78 s->write_poll ? tap_writable : NULL,
79 s);
80 }
81
82 static void tap_read_poll(TAPState *s, int enable)
83 {
84 s->read_poll = !!enable;
85 tap_update_fd_handler(s);
86 }
87
88 static void tap_write_poll(TAPState *s, int enable)
89 {
90 s->write_poll = !!enable;
91 tap_update_fd_handler(s);
92 }
93
94 static void tap_writable(void *opaque)
95 {
96 TAPState *s = opaque;
97
98 tap_write_poll(s, 0);
99
100 qemu_flush_queued_packets(&s->nc);
101 }
102
103 static ssize_t tap_write_packet(TAPState *s, const struct iovec *iov, int iovcnt)
104 {
105 ssize_t len;
106
107 do {
108 len = writev(s->fd, iov, iovcnt);
109 } while (len == -1 && errno == EINTR);
110
111 if (len == -1 && errno == EAGAIN) {
112 tap_write_poll(s, 1);
113 return 0;
114 }
115
116 return len;
117 }
118
119 static ssize_t tap_receive_iov(NetClientState *nc, const struct iovec *iov,
120 int iovcnt)
121 {
122 TAPState *s = DO_UPCAST(TAPState, nc, nc);
123 const struct iovec *iovp = iov;
124 struct iovec iov_copy[iovcnt + 1];
125 struct virtio_net_hdr_mrg_rxbuf hdr = { };
126
127 if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
128 iov_copy[0].iov_base = &hdr;
129 iov_copy[0].iov_len = s->host_vnet_hdr_len;
130 memcpy(&iov_copy[1], iov, iovcnt * sizeof(*iov));
131 iovp = iov_copy;
132 iovcnt++;
133 }
134
135 return tap_write_packet(s, iovp, iovcnt);
136 }
137
138 static ssize_t tap_receive_raw(NetClientState *nc, const uint8_t *buf, size_t size)
139 {
140 TAPState *s = DO_UPCAST(TAPState, nc, nc);
141 struct iovec iov[2];
142 int iovcnt = 0;
143 struct virtio_net_hdr_mrg_rxbuf hdr = { };
144
145 if (s->host_vnet_hdr_len) {
146 iov[iovcnt].iov_base = &hdr;
147 iov[iovcnt].iov_len = s->host_vnet_hdr_len;
148 iovcnt++;
149 }
150
151 iov[iovcnt].iov_base = (char *)buf;
152 iov[iovcnt].iov_len = size;
153 iovcnt++;
154
155 return tap_write_packet(s, iov, iovcnt);
156 }
157
158 static ssize_t tap_receive(NetClientState *nc, const uint8_t *buf, size_t size)
159 {
160 TAPState *s = DO_UPCAST(TAPState, nc, nc);
161 struct iovec iov[1];
162
163 if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
164 return tap_receive_raw(nc, buf, size);
165 }
166
167 iov[0].iov_base = (char *)buf;
168 iov[0].iov_len = size;
169
170 return tap_write_packet(s, iov, 1);
171 }
172
173 static int tap_can_send(void *opaque)
174 {
175 TAPState *s = opaque;
176
177 return qemu_can_send_packet(&s->nc);
178 }
179
180 #ifndef __sun__
181 ssize_t tap_read_packet(int tapfd, uint8_t *buf, int maxlen)
182 {
183 return read(tapfd, buf, maxlen);
184 }
185 #endif
186
187 static void tap_send_completed(NetClientState *nc, ssize_t len)
188 {
189 TAPState *s = DO_UPCAST(TAPState, nc, nc);
190 tap_read_poll(s, 1);
191 }
192
193 static void tap_send(void *opaque)
194 {
195 TAPState *s = opaque;
196 int size;
197
198 do {
199 uint8_t *buf = s->buf;
200
201 size = tap_read_packet(s->fd, s->buf, sizeof(s->buf));
202 if (size <= 0) {
203 break;
204 }
205
206 if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
207 buf += s->host_vnet_hdr_len;
208 size -= s->host_vnet_hdr_len;
209 }
210
211 size = qemu_send_packet_async(&s->nc, buf, size, tap_send_completed);
212 if (size == 0) {
213 tap_read_poll(s, 0);
214 }
215 } while (size > 0 && qemu_can_send_packet(&s->nc));
216 }
217
218 int tap_has_ufo(NetClientState *nc)
219 {
220 TAPState *s = DO_UPCAST(TAPState, nc, nc);
221
222 assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
223
224 return s->has_ufo;
225 }
226
227 int tap_has_vnet_hdr(NetClientState *nc)
228 {
229 TAPState *s = DO_UPCAST(TAPState, nc, nc);
230
231 assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
232
233 return !!s->host_vnet_hdr_len;
234 }
235
236 int tap_has_vnet_hdr_len(NetClientState *nc, int len)
237 {
238 TAPState *s = DO_UPCAST(TAPState, nc, nc);
239
240 assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
241
242 return tap_probe_vnet_hdr_len(s->fd, len);
243 }
244
245 void tap_set_vnet_hdr_len(NetClientState *nc, int len)
246 {
247 TAPState *s = DO_UPCAST(TAPState, nc, nc);
248
249 assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
250 assert(len == sizeof(struct virtio_net_hdr_mrg_rxbuf) ||
251 len == sizeof(struct virtio_net_hdr));
252
253 tap_fd_set_vnet_hdr_len(s->fd, len);
254 s->host_vnet_hdr_len = len;
255 }
256
257 void tap_using_vnet_hdr(NetClientState *nc, int using_vnet_hdr)
258 {
259 TAPState *s = DO_UPCAST(TAPState, nc, nc);
260
261 using_vnet_hdr = using_vnet_hdr != 0;
262
263 assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
264 assert(!!s->host_vnet_hdr_len == using_vnet_hdr);
265
266 s->using_vnet_hdr = using_vnet_hdr;
267 }
268
269 void tap_set_offload(NetClientState *nc, int csum, int tso4,
270 int tso6, int ecn, int ufo)
271 {
272 TAPState *s = DO_UPCAST(TAPState, nc, nc);
273 if (s->fd < 0) {
274 return;
275 }
276
277 tap_fd_set_offload(s->fd, csum, tso4, tso6, ecn, ufo);
278 }
279
280 static void tap_cleanup(NetClientState *nc)
281 {
282 TAPState *s = DO_UPCAST(TAPState, nc, nc);
283
284 if (s->vhost_net) {
285 vhost_net_cleanup(s->vhost_net);
286 s->vhost_net = NULL;
287 }
288
289 qemu_purge_queued_packets(nc);
290
291 if (s->down_script[0])
292 launch_script(s->down_script, s->down_script_arg, s->fd);
293
294 tap_read_poll(s, 0);
295 tap_write_poll(s, 0);
296 close(s->fd);
297 s->fd = -1;
298 }
299
300 static void tap_poll(NetClientState *nc, bool enable)
301 {
302 TAPState *s = DO_UPCAST(TAPState, nc, nc);
303 tap_read_poll(s, enable);
304 tap_write_poll(s, enable);
305 }
306
307 int tap_get_fd(NetClientState *nc)
308 {
309 TAPState *s = DO_UPCAST(TAPState, nc, nc);
310 assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
311 return s->fd;
312 }
313
314 /* fd support */
315
316 static NetClientInfo net_tap_info = {
317 .type = NET_CLIENT_OPTIONS_KIND_TAP,
318 .size = sizeof(TAPState),
319 .receive = tap_receive,
320 .receive_raw = tap_receive_raw,
321 .receive_iov = tap_receive_iov,
322 .poll = tap_poll,
323 .cleanup = tap_cleanup,
324 };
325
326 static TAPState *net_tap_fd_init(NetClientState *peer,
327 const char *model,
328 const char *name,
329 int fd,
330 int vnet_hdr)
331 {
332 NetClientState *nc;
333 TAPState *s;
334
335 nc = qemu_new_net_client(&net_tap_info, peer, model, name);
336
337 s = DO_UPCAST(TAPState, nc, nc);
338
339 s->fd = fd;
340 s->host_vnet_hdr_len = vnet_hdr ? sizeof(struct virtio_net_hdr) : 0;
341 s->using_vnet_hdr = 0;
342 s->has_ufo = tap_probe_has_ufo(s->fd);
343 tap_set_offload(&s->nc, 0, 0, 0, 0, 0);
344 /*
345 * Make sure host header length is set correctly in tap:
346 * it might have been modified by another instance of qemu.
347 */
348 if (tap_probe_vnet_hdr_len(s->fd, s->host_vnet_hdr_len)) {
349 tap_fd_set_vnet_hdr_len(s->fd, s->host_vnet_hdr_len);
350 }
351 tap_read_poll(s, 1);
352 s->vhost_net = NULL;
353 return s;
354 }
355
356 static int launch_script(const char *setup_script, const char *ifname, int fd)
357 {
358 int pid, status;
359 char *args[3];
360 char **parg;
361
362 /* try to launch network script */
363 pid = fork();
364 if (pid == 0) {
365 int open_max = sysconf(_SC_OPEN_MAX), i;
366
367 for (i = 0; i < open_max; i++) {
368 if (i != STDIN_FILENO &&
369 i != STDOUT_FILENO &&
370 i != STDERR_FILENO &&
371 i != fd) {
372 close(i);
373 }
374 }
375 parg = args;
376 *parg++ = (char *)setup_script;
377 *parg++ = (char *)ifname;
378 *parg = NULL;
379 execv(setup_script, args);
380 _exit(1);
381 } else if (pid > 0) {
382 while (waitpid(pid, &status, 0) != pid) {
383 /* loop */
384 }
385
386 if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
387 return 0;
388 }
389 }
390 fprintf(stderr, "%s: could not launch network script\n", setup_script);
391 return -1;
392 }
393
394 static int recv_fd(int c)
395 {
396 int fd;
397 uint8_t msgbuf[CMSG_SPACE(sizeof(fd))];
398 struct msghdr msg = {
399 .msg_control = msgbuf,
400 .msg_controllen = sizeof(msgbuf),
401 };
402 struct cmsghdr *cmsg;
403 struct iovec iov;
404 uint8_t req[1];
405 ssize_t len;
406
407 cmsg = CMSG_FIRSTHDR(&msg);
408 cmsg->cmsg_level = SOL_SOCKET;
409 cmsg->cmsg_type = SCM_RIGHTS;
410 cmsg->cmsg_len = CMSG_LEN(sizeof(fd));
411 msg.msg_controllen = cmsg->cmsg_len;
412
413 iov.iov_base = req;
414 iov.iov_len = sizeof(req);
415
416 msg.msg_iov = &iov;
417 msg.msg_iovlen = 1;
418
419 len = recvmsg(c, &msg, 0);
420 if (len > 0) {
421 memcpy(&fd, CMSG_DATA(cmsg), sizeof(fd));
422 return fd;
423 }
424
425 return len;
426 }
427
428 static int net_bridge_run_helper(const char *helper, const char *bridge)
429 {
430 sigset_t oldmask, mask;
431 int pid, status;
432 char *args[5];
433 char **parg;
434 int sv[2];
435
436 sigemptyset(&mask);
437 sigaddset(&mask, SIGCHLD);
438 sigprocmask(SIG_BLOCK, &mask, &oldmask);
439
440 if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) {
441 return -1;
442 }
443
444 /* try to launch bridge helper */
445 pid = fork();
446 if (pid == 0) {
447 int open_max = sysconf(_SC_OPEN_MAX), i;
448 char fd_buf[6+10];
449 char br_buf[6+IFNAMSIZ] = {0};
450 char helper_cmd[PATH_MAX + sizeof(fd_buf) + sizeof(br_buf) + 15];
451
452 for (i = 0; i < open_max; i++) {
453 if (i != STDIN_FILENO &&
454 i != STDOUT_FILENO &&
455 i != STDERR_FILENO &&
456 i != sv[1]) {
457 close(i);
458 }
459 }
460
461 snprintf(fd_buf, sizeof(fd_buf), "%s%d", "--fd=", sv[1]);
462
463 if (strrchr(helper, ' ') || strrchr(helper, '\t')) {
464 /* assume helper is a command */
465
466 if (strstr(helper, "--br=") == NULL) {
467 snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge);
468 }
469
470 snprintf(helper_cmd, sizeof(helper_cmd), "%s %s %s %s",
471 helper, "--use-vnet", fd_buf, br_buf);
472
473 parg = args;
474 *parg++ = (char *)"sh";
475 *parg++ = (char *)"-c";
476 *parg++ = helper_cmd;
477 *parg++ = NULL;
478
479 execv("/bin/sh", args);
480 } else {
481 /* assume helper is just the executable path name */
482
483 snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge);
484
485 parg = args;
486 *parg++ = (char *)helper;
487 *parg++ = (char *)"--use-vnet";
488 *parg++ = fd_buf;
489 *parg++ = br_buf;
490 *parg++ = NULL;
491
492 execv(helper, args);
493 }
494 _exit(1);
495
496 } else if (pid > 0) {
497 int fd;
498
499 close(sv[1]);
500
501 do {
502 fd = recv_fd(sv[0]);
503 } while (fd == -1 && errno == EINTR);
504
505 close(sv[0]);
506
507 while (waitpid(pid, &status, 0) != pid) {
508 /* loop */
509 }
510 sigprocmask(SIG_SETMASK, &oldmask, NULL);
511 if (fd < 0) {
512 fprintf(stderr, "failed to recv file descriptor\n");
513 return -1;
514 }
515
516 if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
517 return fd;
518 }
519 }
520 fprintf(stderr, "failed to launch bridge helper\n");
521 return -1;
522 }
523
524 int net_init_bridge(const NetClientOptions *opts, const char *name,
525 NetClientState *peer)
526 {
527 const NetdevBridgeOptions *bridge;
528 const char *helper, *br;
529
530 TAPState *s;
531 int fd, vnet_hdr;
532
533 assert(opts->kind == NET_CLIENT_OPTIONS_KIND_BRIDGE);
534 bridge = opts->bridge;
535
536 helper = bridge->has_helper ? bridge->helper : DEFAULT_BRIDGE_HELPER;
537 br = bridge->has_br ? bridge->br : DEFAULT_BRIDGE_INTERFACE;
538
539 fd = net_bridge_run_helper(helper, br);
540 if (fd == -1) {
541 return -1;
542 }
543
544 fcntl(fd, F_SETFL, O_NONBLOCK);
545
546 vnet_hdr = tap_probe_vnet_hdr(fd);
547
548 s = net_tap_fd_init(peer, "bridge", name, fd, vnet_hdr);
549 if (!s) {
550 close(fd);
551 return -1;
552 }
553
554 snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s,br=%s", helper,
555 br);
556
557 return 0;
558 }
559
560 static int net_tap_init(const NetdevTapOptions *tap, int *vnet_hdr,
561 const char *setup_script, char *ifname,
562 size_t ifname_sz)
563 {
564 int fd, vnet_hdr_required;
565
566 if (tap->has_ifname) {
567 pstrcpy(ifname, ifname_sz, tap->ifname);
568 } else {
569 assert(ifname_sz > 0);
570 ifname[0] = '\0';
571 }
572
573 if (tap->has_vnet_hdr) {
574 *vnet_hdr = tap->vnet_hdr;
575 vnet_hdr_required = *vnet_hdr;
576 } else {
577 *vnet_hdr = 1;
578 vnet_hdr_required = 0;
579 }
580
581 TFR(fd = tap_open(ifname, ifname_sz, vnet_hdr, vnet_hdr_required));
582 if (fd < 0) {
583 return -1;
584 }
585
586 if (setup_script &&
587 setup_script[0] != '\0' &&
588 strcmp(setup_script, "no") != 0 &&
589 launch_script(setup_script, ifname, fd)) {
590 close(fd);
591 return -1;
592 }
593
594 return fd;
595 }
596
597 int net_init_tap(const NetClientOptions *opts, const char *name,
598 NetClientState *peer)
599 {
600 const NetdevTapOptions *tap;
601
602 int fd, vnet_hdr = 0;
603 const char *model;
604 TAPState *s;
605
606 /* for the no-fd, no-helper case */
607 const char *script = NULL; /* suppress wrong "uninit'd use" gcc warning */
608 char ifname[128];
609
610 assert(opts->kind == NET_CLIENT_OPTIONS_KIND_TAP);
611 tap = opts->tap;
612
613 if (tap->has_fd) {
614 if (tap->has_ifname || tap->has_script || tap->has_downscript ||
615 tap->has_vnet_hdr || tap->has_helper) {
616 error_report("ifname=, script=, downscript=, vnet_hdr=, "
617 "and helper= are invalid with fd=");
618 return -1;
619 }
620
621 fd = monitor_handle_fd_param(cur_mon, tap->fd);
622 if (fd == -1) {
623 return -1;
624 }
625
626 fcntl(fd, F_SETFL, O_NONBLOCK);
627
628 vnet_hdr = tap_probe_vnet_hdr(fd);
629
630 model = "tap";
631
632 } else if (tap->has_helper) {
633 if (tap->has_ifname || tap->has_script || tap->has_downscript ||
634 tap->has_vnet_hdr) {
635 error_report("ifname=, script=, downscript=, and vnet_hdr= "
636 "are invalid with helper=");
637 return -1;
638 }
639
640 fd = net_bridge_run_helper(tap->helper, DEFAULT_BRIDGE_INTERFACE);
641 if (fd == -1) {
642 return -1;
643 }
644
645 fcntl(fd, F_SETFL, O_NONBLOCK);
646
647 vnet_hdr = tap_probe_vnet_hdr(fd);
648
649 model = "bridge";
650
651 } else {
652 script = tap->has_script ? tap->script : DEFAULT_NETWORK_SCRIPT;
653 fd = net_tap_init(tap, &vnet_hdr, script, ifname, sizeof ifname);
654 if (fd == -1) {
655 return -1;
656 }
657
658 model = "tap";
659 }
660
661 s = net_tap_fd_init(peer, model, name, fd, vnet_hdr);
662 if (!s) {
663 close(fd);
664 return -1;
665 }
666
667 if (tap_set_sndbuf(s->fd, tap) < 0) {
668 return -1;
669 }
670
671 if (tap->has_fd) {
672 snprintf(s->nc.info_str, sizeof(s->nc.info_str), "fd=%d", fd);
673 } else if (tap->has_helper) {
674 snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s",
675 tap->helper);
676 } else {
677 const char *downscript;
678
679 downscript = tap->has_downscript ? tap->downscript :
680 DEFAULT_NETWORK_DOWN_SCRIPT;
681
682 snprintf(s->nc.info_str, sizeof(s->nc.info_str),
683 "ifname=%s,script=%s,downscript=%s", ifname, script,
684 downscript);
685
686 if (strcmp(downscript, "no") != 0) {
687 snprintf(s->down_script, sizeof(s->down_script), "%s", downscript);
688 snprintf(s->down_script_arg, sizeof(s->down_script_arg), "%s", ifname);
689 }
690 }
691
692 if (tap->has_vhost ? tap->vhost :
693 tap->has_vhostfd || (tap->has_vhostforce && tap->vhostforce)) {
694 int vhostfd;
695
696 if (tap->has_vhostfd) {
697 vhostfd = monitor_handle_fd_param(cur_mon, tap->vhostfd);
698 if (vhostfd == -1) {
699 return -1;
700 }
701 } else {
702 vhostfd = -1;
703 }
704
705 s->vhost_net = vhost_net_init(&s->nc, vhostfd,
706 tap->has_vhostforce && tap->vhostforce);
707 if (!s->vhost_net) {
708 error_report("vhost-net requested but could not be initialized");
709 return -1;
710 }
711 } else if (tap->has_vhostfd) {
712 error_report("vhostfd= is not valid without vhost");
713 return -1;
714 }
715
716 return 0;
717 }
718
719 VHostNetState *tap_get_vhost_net(NetClientState *nc)
720 {
721 TAPState *s = DO_UPCAST(TAPState, nc, nc);
722 assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
723 return s->vhost_net;
724 }