]>
Commit | Line | Data |
---|---|---|
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 "qemu/osdep.h" | |
27 | #include "tap_int.h" | |
28 | ||
29 | ||
30 | #include <sys/ioctl.h> | |
31 | #include <sys/wait.h> | |
32 | #include <sys/socket.h> | |
33 | #include <net/if.h> | |
34 | ||
35 | #include "net/net.h" | |
36 | #include "clients.h" | |
37 | #include "monitor/monitor.h" | |
38 | #include "sysemu/sysemu.h" | |
39 | #include "qapi/error.h" | |
40 | #include "qemu-common.h" | |
41 | #include "qemu/cutils.h" | |
42 | #include "qemu/error-report.h" | |
43 | #include "qemu/sockets.h" | |
44 | ||
45 | #include "net/tap.h" | |
46 | ||
47 | #include "net/vhost_net.h" | |
48 | ||
49 | typedef struct TAPState { | |
50 | NetClientState nc; | |
51 | int fd; | |
52 | char down_script[1024]; | |
53 | char down_script_arg[128]; | |
54 | uint8_t buf[NET_BUFSIZE]; | |
55 | bool read_poll; | |
56 | bool write_poll; | |
57 | bool using_vnet_hdr; | |
58 | bool has_ufo; | |
59 | bool enabled; | |
60 | VHostNetState *vhost_net; | |
61 | unsigned host_vnet_hdr_len; | |
62 | Notifier exit; | |
63 | } TAPState; | |
64 | ||
65 | static void launch_script(const char *setup_script, const char *ifname, | |
66 | int fd, Error **errp); | |
67 | ||
68 | static void tap_send(void *opaque); | |
69 | static void tap_writable(void *opaque); | |
70 | ||
71 | static void tap_update_fd_handler(TAPState *s) | |
72 | { | |
73 | qemu_set_fd_handler(s->fd, | |
74 | s->read_poll && s->enabled ? tap_send : NULL, | |
75 | s->write_poll && s->enabled ? tap_writable : NULL, | |
76 | s); | |
77 | } | |
78 | ||
79 | static void tap_read_poll(TAPState *s, bool enable) | |
80 | { | |
81 | s->read_poll = enable; | |
82 | tap_update_fd_handler(s); | |
83 | } | |
84 | ||
85 | static void tap_write_poll(TAPState *s, bool enable) | |
86 | { | |
87 | s->write_poll = enable; | |
88 | tap_update_fd_handler(s); | |
89 | } | |
90 | ||
91 | static void tap_writable(void *opaque) | |
92 | { | |
93 | TAPState *s = opaque; | |
94 | ||
95 | tap_write_poll(s, false); | |
96 | ||
97 | qemu_flush_queued_packets(&s->nc); | |
98 | } | |
99 | ||
100 | static ssize_t tap_write_packet(TAPState *s, const struct iovec *iov, int iovcnt) | |
101 | { | |
102 | ssize_t len; | |
103 | ||
104 | do { | |
105 | len = writev(s->fd, iov, iovcnt); | |
106 | } while (len == -1 && errno == EINTR); | |
107 | ||
108 | if (len == -1 && errno == EAGAIN) { | |
109 | tap_write_poll(s, true); | |
110 | return 0; | |
111 | } | |
112 | ||
113 | return len; | |
114 | } | |
115 | ||
116 | static ssize_t tap_receive_iov(NetClientState *nc, const struct iovec *iov, | |
117 | int iovcnt) | |
118 | { | |
119 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
120 | const struct iovec *iovp = iov; | |
121 | struct iovec iov_copy[iovcnt + 1]; | |
122 | struct virtio_net_hdr_mrg_rxbuf hdr = { }; | |
123 | ||
124 | if (s->host_vnet_hdr_len && !s->using_vnet_hdr) { | |
125 | iov_copy[0].iov_base = &hdr; | |
126 | iov_copy[0].iov_len = s->host_vnet_hdr_len; | |
127 | memcpy(&iov_copy[1], iov, iovcnt * sizeof(*iov)); | |
128 | iovp = iov_copy; | |
129 | iovcnt++; | |
130 | } | |
131 | ||
132 | return tap_write_packet(s, iovp, iovcnt); | |
133 | } | |
134 | ||
135 | static ssize_t tap_receive_raw(NetClientState *nc, const uint8_t *buf, size_t size) | |
136 | { | |
137 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
138 | struct iovec iov[2]; | |
139 | int iovcnt = 0; | |
140 | struct virtio_net_hdr_mrg_rxbuf hdr = { }; | |
141 | ||
142 | if (s->host_vnet_hdr_len) { | |
143 | iov[iovcnt].iov_base = &hdr; | |
144 | iov[iovcnt].iov_len = s->host_vnet_hdr_len; | |
145 | iovcnt++; | |
146 | } | |
147 | ||
148 | iov[iovcnt].iov_base = (char *)buf; | |
149 | iov[iovcnt].iov_len = size; | |
150 | iovcnt++; | |
151 | ||
152 | return tap_write_packet(s, iov, iovcnt); | |
153 | } | |
154 | ||
155 | static ssize_t tap_receive(NetClientState *nc, const uint8_t *buf, size_t size) | |
156 | { | |
157 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
158 | struct iovec iov[1]; | |
159 | ||
160 | if (s->host_vnet_hdr_len && !s->using_vnet_hdr) { | |
161 | return tap_receive_raw(nc, buf, size); | |
162 | } | |
163 | ||
164 | iov[0].iov_base = (char *)buf; | |
165 | iov[0].iov_len = size; | |
166 | ||
167 | return tap_write_packet(s, iov, 1); | |
168 | } | |
169 | ||
170 | #ifndef __sun__ | |
171 | ssize_t tap_read_packet(int tapfd, uint8_t *buf, int maxlen) | |
172 | { | |
173 | return read(tapfd, buf, maxlen); | |
174 | } | |
175 | #endif | |
176 | ||
177 | static void tap_send_completed(NetClientState *nc, ssize_t len) | |
178 | { | |
179 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
180 | tap_read_poll(s, true); | |
181 | } | |
182 | ||
183 | static void tap_send(void *opaque) | |
184 | { | |
185 | TAPState *s = opaque; | |
186 | int size; | |
187 | int packets = 0; | |
188 | ||
189 | while (true) { | |
190 | uint8_t *buf = s->buf; | |
191 | ||
192 | size = tap_read_packet(s->fd, s->buf, sizeof(s->buf)); | |
193 | if (size <= 0) { | |
194 | break; | |
195 | } | |
196 | ||
197 | if (s->host_vnet_hdr_len && !s->using_vnet_hdr) { | |
198 | buf += s->host_vnet_hdr_len; | |
199 | size -= s->host_vnet_hdr_len; | |
200 | } | |
201 | ||
202 | size = qemu_send_packet_async(&s->nc, buf, size, tap_send_completed); | |
203 | if (size == 0) { | |
204 | tap_read_poll(s, false); | |
205 | break; | |
206 | } else if (size < 0) { | |
207 | break; | |
208 | } | |
209 | ||
210 | /* | |
211 | * When the host keeps receiving more packets while tap_send() is | |
212 | * running we can hog the QEMU global mutex. Limit the number of | |
213 | * packets that are processed per tap_send() callback to prevent | |
214 | * stalling the guest. | |
215 | */ | |
216 | packets++; | |
217 | if (packets >= 50) { | |
218 | break; | |
219 | } | |
220 | } | |
221 | } | |
222 | ||
223 | static bool tap_has_ufo(NetClientState *nc) | |
224 | { | |
225 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
226 | ||
227 | assert(nc->info->type == NET_CLIENT_DRIVER_TAP); | |
228 | ||
229 | return s->has_ufo; | |
230 | } | |
231 | ||
232 | static bool tap_has_vnet_hdr(NetClientState *nc) | |
233 | { | |
234 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
235 | ||
236 | assert(nc->info->type == NET_CLIENT_DRIVER_TAP); | |
237 | ||
238 | return !!s->host_vnet_hdr_len; | |
239 | } | |
240 | ||
241 | static bool tap_has_vnet_hdr_len(NetClientState *nc, int len) | |
242 | { | |
243 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
244 | ||
245 | assert(nc->info->type == NET_CLIENT_DRIVER_TAP); | |
246 | ||
247 | return !!tap_probe_vnet_hdr_len(s->fd, len); | |
248 | } | |
249 | ||
250 | static void tap_set_vnet_hdr_len(NetClientState *nc, int len) | |
251 | { | |
252 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
253 | ||
254 | assert(nc->info->type == NET_CLIENT_DRIVER_TAP); | |
255 | assert(len == sizeof(struct virtio_net_hdr_mrg_rxbuf) || | |
256 | len == sizeof(struct virtio_net_hdr)); | |
257 | ||
258 | tap_fd_set_vnet_hdr_len(s->fd, len); | |
259 | s->host_vnet_hdr_len = len; | |
260 | } | |
261 | ||
262 | static void tap_using_vnet_hdr(NetClientState *nc, bool using_vnet_hdr) | |
263 | { | |
264 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
265 | ||
266 | assert(nc->info->type == NET_CLIENT_DRIVER_TAP); | |
267 | assert(!!s->host_vnet_hdr_len == using_vnet_hdr); | |
268 | ||
269 | s->using_vnet_hdr = using_vnet_hdr; | |
270 | } | |
271 | ||
272 | static int tap_set_vnet_le(NetClientState *nc, bool is_le) | |
273 | { | |
274 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
275 | ||
276 | return tap_fd_set_vnet_le(s->fd, is_le); | |
277 | } | |
278 | ||
279 | static int tap_set_vnet_be(NetClientState *nc, bool is_be) | |
280 | { | |
281 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
282 | ||
283 | return tap_fd_set_vnet_be(s->fd, is_be); | |
284 | } | |
285 | ||
286 | static void tap_set_offload(NetClientState *nc, int csum, int tso4, | |
287 | int tso6, int ecn, int ufo) | |
288 | { | |
289 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
290 | if (s->fd < 0) { | |
291 | return; | |
292 | } | |
293 | ||
294 | tap_fd_set_offload(s->fd, csum, tso4, tso6, ecn, ufo); | |
295 | } | |
296 | ||
297 | static void tap_exit_notify(Notifier *notifier, void *data) | |
298 | { | |
299 | TAPState *s = container_of(notifier, TAPState, exit); | |
300 | Error *err = NULL; | |
301 | ||
302 | if (s->down_script[0]) { | |
303 | launch_script(s->down_script, s->down_script_arg, s->fd, &err); | |
304 | if (err) { | |
305 | error_report_err(err); | |
306 | } | |
307 | } | |
308 | } | |
309 | ||
310 | static void tap_cleanup(NetClientState *nc) | |
311 | { | |
312 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
313 | ||
314 | if (s->vhost_net) { | |
315 | vhost_net_cleanup(s->vhost_net); | |
316 | g_free(s->vhost_net); | |
317 | s->vhost_net = NULL; | |
318 | } | |
319 | ||
320 | qemu_purge_queued_packets(nc); | |
321 | ||
322 | tap_exit_notify(&s->exit, NULL); | |
323 | qemu_remove_exit_notifier(&s->exit); | |
324 | ||
325 | tap_read_poll(s, false); | |
326 | tap_write_poll(s, false); | |
327 | close(s->fd); | |
328 | s->fd = -1; | |
329 | } | |
330 | ||
331 | static void tap_poll(NetClientState *nc, bool enable) | |
332 | { | |
333 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
334 | tap_read_poll(s, enable); | |
335 | tap_write_poll(s, enable); | |
336 | } | |
337 | ||
338 | int tap_get_fd(NetClientState *nc) | |
339 | { | |
340 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
341 | assert(nc->info->type == NET_CLIENT_DRIVER_TAP); | |
342 | return s->fd; | |
343 | } | |
344 | ||
345 | /* fd support */ | |
346 | ||
347 | static NetClientInfo net_tap_info = { | |
348 | .type = NET_CLIENT_DRIVER_TAP, | |
349 | .size = sizeof(TAPState), | |
350 | .receive = tap_receive, | |
351 | .receive_raw = tap_receive_raw, | |
352 | .receive_iov = tap_receive_iov, | |
353 | .poll = tap_poll, | |
354 | .cleanup = tap_cleanup, | |
355 | .has_ufo = tap_has_ufo, | |
356 | .has_vnet_hdr = tap_has_vnet_hdr, | |
357 | .has_vnet_hdr_len = tap_has_vnet_hdr_len, | |
358 | .using_vnet_hdr = tap_using_vnet_hdr, | |
359 | .set_offload = tap_set_offload, | |
360 | .set_vnet_hdr_len = tap_set_vnet_hdr_len, | |
361 | .set_vnet_le = tap_set_vnet_le, | |
362 | .set_vnet_be = tap_set_vnet_be, | |
363 | }; | |
364 | ||
365 | static TAPState *net_tap_fd_init(NetClientState *peer, | |
366 | const char *model, | |
367 | const char *name, | |
368 | int fd, | |
369 | int vnet_hdr) | |
370 | { | |
371 | NetClientState *nc; | |
372 | TAPState *s; | |
373 | ||
374 | nc = qemu_new_net_client(&net_tap_info, peer, model, name); | |
375 | ||
376 | s = DO_UPCAST(TAPState, nc, nc); | |
377 | ||
378 | s->fd = fd; | |
379 | s->host_vnet_hdr_len = vnet_hdr ? sizeof(struct virtio_net_hdr) : 0; | |
380 | s->using_vnet_hdr = false; | |
381 | s->has_ufo = tap_probe_has_ufo(s->fd); | |
382 | s->enabled = true; | |
383 | tap_set_offload(&s->nc, 0, 0, 0, 0, 0); | |
384 | /* | |
385 | * Make sure host header length is set correctly in tap: | |
386 | * it might have been modified by another instance of qemu. | |
387 | */ | |
388 | if (tap_probe_vnet_hdr_len(s->fd, s->host_vnet_hdr_len)) { | |
389 | tap_fd_set_vnet_hdr_len(s->fd, s->host_vnet_hdr_len); | |
390 | } | |
391 | tap_read_poll(s, true); | |
392 | s->vhost_net = NULL; | |
393 | ||
394 | s->exit.notify = tap_exit_notify; | |
395 | qemu_add_exit_notifier(&s->exit); | |
396 | ||
397 | return s; | |
398 | } | |
399 | ||
400 | static void launch_script(const char *setup_script, const char *ifname, | |
401 | int fd, Error **errp) | |
402 | { | |
403 | int pid, status; | |
404 | char *args[3]; | |
405 | char **parg; | |
406 | ||
407 | /* try to launch network script */ | |
408 | pid = fork(); | |
409 | if (pid < 0) { | |
410 | error_setg_errno(errp, errno, "could not launch network script %s", | |
411 | setup_script); | |
412 | return; | |
413 | } | |
414 | if (pid == 0) { | |
415 | int open_max = sysconf(_SC_OPEN_MAX), i; | |
416 | ||
417 | for (i = 3; i < open_max; i++) { | |
418 | if (i != fd) { | |
419 | close(i); | |
420 | } | |
421 | } | |
422 | parg = args; | |
423 | *parg++ = (char *)setup_script; | |
424 | *parg++ = (char *)ifname; | |
425 | *parg = NULL; | |
426 | execv(setup_script, args); | |
427 | _exit(1); | |
428 | } else { | |
429 | while (waitpid(pid, &status, 0) != pid) { | |
430 | /* loop */ | |
431 | } | |
432 | ||
433 | if (WIFEXITED(status) && WEXITSTATUS(status) == 0) { | |
434 | return; | |
435 | } | |
436 | error_setg(errp, "network script %s failed with status %d", | |
437 | setup_script, status); | |
438 | } | |
439 | } | |
440 | ||
441 | static int recv_fd(int c) | |
442 | { | |
443 | int fd; | |
444 | uint8_t msgbuf[CMSG_SPACE(sizeof(fd))]; | |
445 | struct msghdr msg = { | |
446 | .msg_control = msgbuf, | |
447 | .msg_controllen = sizeof(msgbuf), | |
448 | }; | |
449 | struct cmsghdr *cmsg; | |
450 | struct iovec iov; | |
451 | uint8_t req[1]; | |
452 | ssize_t len; | |
453 | ||
454 | cmsg = CMSG_FIRSTHDR(&msg); | |
455 | cmsg->cmsg_level = SOL_SOCKET; | |
456 | cmsg->cmsg_type = SCM_RIGHTS; | |
457 | cmsg->cmsg_len = CMSG_LEN(sizeof(fd)); | |
458 | msg.msg_controllen = cmsg->cmsg_len; | |
459 | ||
460 | iov.iov_base = req; | |
461 | iov.iov_len = sizeof(req); | |
462 | ||
463 | msg.msg_iov = &iov; | |
464 | msg.msg_iovlen = 1; | |
465 | ||
466 | len = recvmsg(c, &msg, 0); | |
467 | if (len > 0) { | |
468 | memcpy(&fd, CMSG_DATA(cmsg), sizeof(fd)); | |
469 | return fd; | |
470 | } | |
471 | ||
472 | return len; | |
473 | } | |
474 | ||
475 | static int net_bridge_run_helper(const char *helper, const char *bridge, | |
476 | Error **errp) | |
477 | { | |
478 | sigset_t oldmask, mask; | |
479 | int pid, status; | |
480 | char *args[5]; | |
481 | char **parg; | |
482 | int sv[2]; | |
483 | ||
484 | sigemptyset(&mask); | |
485 | sigaddset(&mask, SIGCHLD); | |
486 | sigprocmask(SIG_BLOCK, &mask, &oldmask); | |
487 | ||
488 | if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) { | |
489 | error_setg_errno(errp, errno, "socketpair() failed"); | |
490 | return -1; | |
491 | } | |
492 | ||
493 | /* try to launch bridge helper */ | |
494 | pid = fork(); | |
495 | if (pid < 0) { | |
496 | error_setg_errno(errp, errno, "Can't fork bridge helper"); | |
497 | return -1; | |
498 | } | |
499 | if (pid == 0) { | |
500 | int open_max = sysconf(_SC_OPEN_MAX), i; | |
501 | char fd_buf[6+10]; | |
502 | char br_buf[6+IFNAMSIZ] = {0}; | |
503 | char helper_cmd[PATH_MAX + sizeof(fd_buf) + sizeof(br_buf) + 15]; | |
504 | ||
505 | for (i = 3; i < open_max; i++) { | |
506 | if (i != sv[1]) { | |
507 | close(i); | |
508 | } | |
509 | } | |
510 | ||
511 | snprintf(fd_buf, sizeof(fd_buf), "%s%d", "--fd=", sv[1]); | |
512 | ||
513 | if (strrchr(helper, ' ') || strrchr(helper, '\t')) { | |
514 | /* assume helper is a command */ | |
515 | ||
516 | if (strstr(helper, "--br=") == NULL) { | |
517 | snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge); | |
518 | } | |
519 | ||
520 | snprintf(helper_cmd, sizeof(helper_cmd), "%s %s %s %s", | |
521 | helper, "--use-vnet", fd_buf, br_buf); | |
522 | ||
523 | parg = args; | |
524 | *parg++ = (char *)"sh"; | |
525 | *parg++ = (char *)"-c"; | |
526 | *parg++ = helper_cmd; | |
527 | *parg++ = NULL; | |
528 | ||
529 | execv("/bin/sh", args); | |
530 | } else { | |
531 | /* assume helper is just the executable path name */ | |
532 | ||
533 | snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge); | |
534 | ||
535 | parg = args; | |
536 | *parg++ = (char *)helper; | |
537 | *parg++ = (char *)"--use-vnet"; | |
538 | *parg++ = fd_buf; | |
539 | *parg++ = br_buf; | |
540 | *parg++ = NULL; | |
541 | ||
542 | execv(helper, args); | |
543 | } | |
544 | _exit(1); | |
545 | ||
546 | } else { | |
547 | int fd; | |
548 | int saved_errno; | |
549 | ||
550 | close(sv[1]); | |
551 | ||
552 | do { | |
553 | fd = recv_fd(sv[0]); | |
554 | } while (fd == -1 && errno == EINTR); | |
555 | saved_errno = errno; | |
556 | ||
557 | close(sv[0]); | |
558 | ||
559 | while (waitpid(pid, &status, 0) != pid) { | |
560 | /* loop */ | |
561 | } | |
562 | sigprocmask(SIG_SETMASK, &oldmask, NULL); | |
563 | if (fd < 0) { | |
564 | error_setg_errno(errp, saved_errno, | |
565 | "failed to recv file descriptor"); | |
566 | return -1; | |
567 | } | |
568 | if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) { | |
569 | error_setg(errp, "bridge helper failed"); | |
570 | return -1; | |
571 | } | |
572 | return fd; | |
573 | } | |
574 | } | |
575 | ||
576 | int net_init_bridge(const Netdev *netdev, const char *name, | |
577 | NetClientState *peer, Error **errp) | |
578 | { | |
579 | const NetdevBridgeOptions *bridge; | |
580 | const char *helper, *br; | |
581 | TAPState *s; | |
582 | int fd, vnet_hdr; | |
583 | ||
584 | assert(netdev->type == NET_CLIENT_DRIVER_BRIDGE); | |
585 | bridge = &netdev->u.bridge; | |
586 | ||
587 | helper = bridge->has_helper ? bridge->helper : DEFAULT_BRIDGE_HELPER; | |
588 | br = bridge->has_br ? bridge->br : DEFAULT_BRIDGE_INTERFACE; | |
589 | ||
590 | fd = net_bridge_run_helper(helper, br, errp); | |
591 | if (fd == -1) { | |
592 | return -1; | |
593 | } | |
594 | ||
595 | qemu_set_nonblock(fd); | |
596 | vnet_hdr = tap_probe_vnet_hdr(fd); | |
597 | s = net_tap_fd_init(peer, "bridge", name, fd, vnet_hdr); | |
598 | ||
599 | snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s,br=%s", helper, | |
600 | br); | |
601 | ||
602 | return 0; | |
603 | } | |
604 | ||
605 | static int net_tap_init(const NetdevTapOptions *tap, int *vnet_hdr, | |
606 | const char *setup_script, char *ifname, | |
607 | size_t ifname_sz, int mq_required, Error **errp) | |
608 | { | |
609 | Error *err = NULL; | |
610 | int fd, vnet_hdr_required; | |
611 | ||
612 | if (tap->has_vnet_hdr) { | |
613 | *vnet_hdr = tap->vnet_hdr; | |
614 | vnet_hdr_required = *vnet_hdr; | |
615 | } else { | |
616 | *vnet_hdr = 1; | |
617 | vnet_hdr_required = 0; | |
618 | } | |
619 | ||
620 | TFR(fd = tap_open(ifname, ifname_sz, vnet_hdr, vnet_hdr_required, | |
621 | mq_required, errp)); | |
622 | if (fd < 0) { | |
623 | return -1; | |
624 | } | |
625 | ||
626 | if (setup_script && | |
627 | setup_script[0] != '\0' && | |
628 | strcmp(setup_script, "no") != 0) { | |
629 | launch_script(setup_script, ifname, fd, &err); | |
630 | if (err) { | |
631 | error_propagate(errp, err); | |
632 | close(fd); | |
633 | return -1; | |
634 | } | |
635 | } | |
636 | ||
637 | return fd; | |
638 | } | |
639 | ||
640 | #define MAX_TAP_QUEUES 1024 | |
641 | ||
642 | static void net_init_tap_one(const NetdevTapOptions *tap, NetClientState *peer, | |
643 | const char *model, const char *name, | |
644 | const char *ifname, const char *script, | |
645 | const char *downscript, const char *vhostfdname, | |
646 | int vnet_hdr, int fd, Error **errp) | |
647 | { | |
648 | Error *err = NULL; | |
649 | TAPState *s = net_tap_fd_init(peer, model, name, fd, vnet_hdr); | |
650 | int vhostfd; | |
651 | ||
652 | tap_set_sndbuf(s->fd, tap, &err); | |
653 | if (err) { | |
654 | error_propagate(errp, err); | |
655 | return; | |
656 | } | |
657 | ||
658 | if (tap->has_fd || tap->has_fds) { | |
659 | snprintf(s->nc.info_str, sizeof(s->nc.info_str), "fd=%d", fd); | |
660 | } else if (tap->has_helper) { | |
661 | snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s", | |
662 | tap->helper); | |
663 | } else { | |
664 | snprintf(s->nc.info_str, sizeof(s->nc.info_str), | |
665 | "ifname=%s,script=%s,downscript=%s", ifname, script, | |
666 | downscript); | |
667 | ||
668 | if (strcmp(downscript, "no") != 0) { | |
669 | snprintf(s->down_script, sizeof(s->down_script), "%s", downscript); | |
670 | snprintf(s->down_script_arg, sizeof(s->down_script_arg), | |
671 | "%s", ifname); | |
672 | } | |
673 | } | |
674 | ||
675 | if (tap->has_vhost ? tap->vhost : | |
676 | vhostfdname || (tap->has_vhostforce && tap->vhostforce)) { | |
677 | VhostNetOptions options; | |
678 | ||
679 | options.backend_type = VHOST_BACKEND_TYPE_KERNEL; | |
680 | options.net_backend = &s->nc; | |
681 | if (tap->has_poll_us) { | |
682 | options.busyloop_timeout = tap->poll_us; | |
683 | } else { | |
684 | options.busyloop_timeout = 0; | |
685 | } | |
686 | ||
687 | if (vhostfdname) { | |
688 | vhostfd = monitor_fd_param(cur_mon, vhostfdname, &err); | |
689 | if (vhostfd == -1) { | |
690 | if (tap->has_vhostforce && tap->vhostforce) { | |
691 | error_propagate(errp, err); | |
692 | } else { | |
693 | warn_report_err(err); | |
694 | } | |
695 | return; | |
696 | } | |
697 | qemu_set_nonblock(vhostfd); | |
698 | } else { | |
699 | vhostfd = open("/dev/vhost-net", O_RDWR); | |
700 | if (vhostfd < 0) { | |
701 | if (tap->has_vhostforce && tap->vhostforce) { | |
702 | error_setg_errno(errp, errno, | |
703 | "tap: open vhost char device failed"); | |
704 | } else { | |
705 | warn_report("tap: open vhost char device failed: %s", | |
706 | strerror(errno)); | |
707 | } | |
708 | return; | |
709 | } | |
710 | qemu_set_nonblock(vhostfd); | |
711 | } | |
712 | options.opaque = (void *)(uintptr_t)vhostfd; | |
713 | ||
714 | s->vhost_net = vhost_net_init(&options); | |
715 | if (!s->vhost_net) { | |
716 | if (tap->has_vhostforce && tap->vhostforce) { | |
717 | error_setg(errp, VHOST_NET_INIT_FAILED); | |
718 | } else { | |
719 | warn_report(VHOST_NET_INIT_FAILED); | |
720 | } | |
721 | return; | |
722 | } | |
723 | } else if (vhostfdname) { | |
724 | error_setg(errp, "vhostfd(s)= is not valid without vhost"); | |
725 | } | |
726 | } | |
727 | ||
728 | static int get_fds(char *str, char *fds[], int max) | |
729 | { | |
730 | char *ptr = str, *this; | |
731 | size_t len = strlen(str); | |
732 | int i = 0; | |
733 | ||
734 | while (i < max && ptr < str + len) { | |
735 | this = strchr(ptr, ':'); | |
736 | ||
737 | if (this == NULL) { | |
738 | fds[i] = g_strdup(ptr); | |
739 | } else { | |
740 | fds[i] = g_strndup(ptr, this - ptr); | |
741 | } | |
742 | ||
743 | i++; | |
744 | if (this == NULL) { | |
745 | break; | |
746 | } else { | |
747 | ptr = this + 1; | |
748 | } | |
749 | } | |
750 | ||
751 | return i; | |
752 | } | |
753 | ||
754 | int net_init_tap(const Netdev *netdev, const char *name, | |
755 | NetClientState *peer, Error **errp) | |
756 | { | |
757 | const NetdevTapOptions *tap; | |
758 | int fd, vnet_hdr = 0, i = 0, queues; | |
759 | /* for the no-fd, no-helper case */ | |
760 | const char *script = NULL; /* suppress wrong "uninit'd use" gcc warning */ | |
761 | const char *downscript = NULL; | |
762 | Error *err = NULL; | |
763 | const char *vhostfdname; | |
764 | char ifname[128]; | |
765 | ||
766 | assert(netdev->type == NET_CLIENT_DRIVER_TAP); | |
767 | tap = &netdev->u.tap; | |
768 | queues = tap->has_queues ? tap->queues : 1; | |
769 | vhostfdname = tap->has_vhostfd ? tap->vhostfd : NULL; | |
770 | ||
771 | /* QEMU hubs do not support multiqueue tap, in this case peer is set. | |
772 | * For -netdev, peer is always NULL. */ | |
773 | if (peer && (tap->has_queues || tap->has_fds || tap->has_vhostfds)) { | |
774 | error_setg(errp, "Multiqueue tap cannot be used with hubs"); | |
775 | return -1; | |
776 | } | |
777 | ||
778 | if (tap->has_fd) { | |
779 | if (tap->has_ifname || tap->has_script || tap->has_downscript || | |
780 | tap->has_vnet_hdr || tap->has_helper || tap->has_queues || | |
781 | tap->has_fds || tap->has_vhostfds) { | |
782 | error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, " | |
783 | "helper=, queues=, fds=, and vhostfds= " | |
784 | "are invalid with fd="); | |
785 | return -1; | |
786 | } | |
787 | ||
788 | fd = monitor_fd_param(cur_mon, tap->fd, &err); | |
789 | if (fd == -1) { | |
790 | error_propagate(errp, err); | |
791 | return -1; | |
792 | } | |
793 | ||
794 | qemu_set_nonblock(fd); | |
795 | ||
796 | vnet_hdr = tap_probe_vnet_hdr(fd); | |
797 | ||
798 | net_init_tap_one(tap, peer, "tap", name, NULL, | |
799 | script, downscript, | |
800 | vhostfdname, vnet_hdr, fd, &err); | |
801 | if (err) { | |
802 | error_propagate(errp, err); | |
803 | return -1; | |
804 | } | |
805 | } else if (tap->has_fds) { | |
806 | char **fds; | |
807 | char **vhost_fds; | |
808 | int nfds = 0, nvhosts = 0; | |
809 | int ret = 0; | |
810 | ||
811 | if (tap->has_ifname || tap->has_script || tap->has_downscript || | |
812 | tap->has_vnet_hdr || tap->has_helper || tap->has_queues || | |
813 | tap->has_vhostfd) { | |
814 | error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, " | |
815 | "helper=, queues=, and vhostfd= " | |
816 | "are invalid with fds="); | |
817 | return -1; | |
818 | } | |
819 | ||
820 | fds = g_new0(char *, MAX_TAP_QUEUES); | |
821 | vhost_fds = g_new0(char *, MAX_TAP_QUEUES); | |
822 | ||
823 | nfds = get_fds(tap->fds, fds, MAX_TAP_QUEUES); | |
824 | if (tap->has_vhostfds) { | |
825 | nvhosts = get_fds(tap->vhostfds, vhost_fds, MAX_TAP_QUEUES); | |
826 | if (nfds != nvhosts) { | |
827 | error_setg(errp, "The number of fds passed does not match " | |
828 | "the number of vhostfds passed"); | |
829 | ret = -1; | |
830 | goto free_fail; | |
831 | } | |
832 | } | |
833 | ||
834 | for (i = 0; i < nfds; i++) { | |
835 | fd = monitor_fd_param(cur_mon, fds[i], &err); | |
836 | if (fd == -1) { | |
837 | error_propagate(errp, err); | |
838 | ret = -1; | |
839 | goto free_fail; | |
840 | } | |
841 | ||
842 | qemu_set_nonblock(fd); | |
843 | ||
844 | if (i == 0) { | |
845 | vnet_hdr = tap_probe_vnet_hdr(fd); | |
846 | } else if (vnet_hdr != tap_probe_vnet_hdr(fd)) { | |
847 | error_setg(errp, | |
848 | "vnet_hdr not consistent across given tap fds"); | |
849 | ret = -1; | |
850 | goto free_fail; | |
851 | } | |
852 | ||
853 | net_init_tap_one(tap, peer, "tap", name, ifname, | |
854 | script, downscript, | |
855 | tap->has_vhostfds ? vhost_fds[i] : NULL, | |
856 | vnet_hdr, fd, &err); | |
857 | if (err) { | |
858 | error_propagate(errp, err); | |
859 | ret = -1; | |
860 | goto free_fail; | |
861 | } | |
862 | } | |
863 | ||
864 | free_fail: | |
865 | for (i = 0; i < nvhosts; i++) { | |
866 | g_free(vhost_fds[i]); | |
867 | } | |
868 | for (i = 0; i < nfds; i++) { | |
869 | g_free(fds[i]); | |
870 | } | |
871 | g_free(fds); | |
872 | g_free(vhost_fds); | |
873 | return ret; | |
874 | } else if (tap->has_helper) { | |
875 | if (tap->has_ifname || tap->has_script || tap->has_downscript || | |
876 | tap->has_vnet_hdr || tap->has_queues || tap->has_vhostfds) { | |
877 | error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, " | |
878 | "queues=, and vhostfds= are invalid with helper="); | |
879 | return -1; | |
880 | } | |
881 | ||
882 | fd = net_bridge_run_helper(tap->helper, | |
883 | tap->has_br ? | |
884 | tap->br : DEFAULT_BRIDGE_INTERFACE, | |
885 | errp); | |
886 | if (fd == -1) { | |
887 | return -1; | |
888 | } | |
889 | ||
890 | qemu_set_nonblock(fd); | |
891 | vnet_hdr = tap_probe_vnet_hdr(fd); | |
892 | ||
893 | net_init_tap_one(tap, peer, "bridge", name, ifname, | |
894 | script, downscript, vhostfdname, | |
895 | vnet_hdr, fd, &err); | |
896 | if (err) { | |
897 | error_propagate(errp, err); | |
898 | close(fd); | |
899 | return -1; | |
900 | } | |
901 | } else { | |
902 | if (tap->has_vhostfds) { | |
903 | error_setg(errp, "vhostfds= is invalid if fds= wasn't specified"); | |
904 | return -1; | |
905 | } | |
906 | script = tap->has_script ? tap->script : DEFAULT_NETWORK_SCRIPT; | |
907 | downscript = tap->has_downscript ? tap->downscript : | |
908 | DEFAULT_NETWORK_DOWN_SCRIPT; | |
909 | ||
910 | if (tap->has_ifname) { | |
911 | pstrcpy(ifname, sizeof ifname, tap->ifname); | |
912 | } else { | |
913 | ifname[0] = '\0'; | |
914 | } | |
915 | ||
916 | for (i = 0; i < queues; i++) { | |
917 | fd = net_tap_init(tap, &vnet_hdr, i >= 1 ? "no" : script, | |
918 | ifname, sizeof ifname, queues > 1, errp); | |
919 | if (fd == -1) { | |
920 | return -1; | |
921 | } | |
922 | ||
923 | if (queues > 1 && i == 0 && !tap->has_ifname) { | |
924 | if (tap_fd_get_ifname(fd, ifname)) { | |
925 | error_setg(errp, "Fail to get ifname"); | |
926 | close(fd); | |
927 | return -1; | |
928 | } | |
929 | } | |
930 | ||
931 | net_init_tap_one(tap, peer, "tap", name, ifname, | |
932 | i >= 1 ? "no" : script, | |
933 | i >= 1 ? "no" : downscript, | |
934 | vhostfdname, vnet_hdr, fd, &err); | |
935 | if (err) { | |
936 | error_propagate(errp, err); | |
937 | close(fd); | |
938 | return -1; | |
939 | } | |
940 | } | |
941 | } | |
942 | ||
943 | return 0; | |
944 | } | |
945 | ||
946 | VHostNetState *tap_get_vhost_net(NetClientState *nc) | |
947 | { | |
948 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
949 | assert(nc->info->type == NET_CLIENT_DRIVER_TAP); | |
950 | return s->vhost_net; | |
951 | } | |
952 | ||
953 | int tap_enable(NetClientState *nc) | |
954 | { | |
955 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
956 | int ret; | |
957 | ||
958 | if (s->enabled) { | |
959 | return 0; | |
960 | } else { | |
961 | ret = tap_fd_enable(s->fd); | |
962 | if (ret == 0) { | |
963 | s->enabled = true; | |
964 | tap_update_fd_handler(s); | |
965 | } | |
966 | return ret; | |
967 | } | |
968 | } | |
969 | ||
970 | int tap_disable(NetClientState *nc) | |
971 | { | |
972 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
973 | int ret; | |
974 | ||
975 | if (s->enabled == 0) { | |
976 | return 0; | |
977 | } else { | |
978 | ret = tap_fd_disable(s->fd); | |
979 | if (ret == 0) { | |
980 | qemu_purge_queued_packets(nc); | |
981 | s->enabled = false; | |
982 | tap_update_fd_handler(s); | |
983 | } | |
984 | return ret; | |
985 | } | |
986 | } |