]>
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/eth.h" | |
36 | #include "net/net.h" | |
37 | #include "clients.h" | |
38 | #include "monitor/monitor.h" | |
39 | #include "sysemu/sysemu.h" | |
40 | #include "qapi/error.h" | |
41 | #include "qemu/cutils.h" | |
42 | #include "qemu/error-report.h" | |
43 | #include "qemu/main-loop.h" | |
44 | #include "qemu/sockets.h" | |
45 | ||
46 | #include "net/tap.h" | |
47 | ||
48 | #include "net/vhost_net.h" | |
49 | ||
50 | typedef struct TAPState { | |
51 | NetClientState nc; | |
52 | int fd; | |
53 | char down_script[1024]; | |
54 | char down_script_arg[128]; | |
55 | uint8_t buf[NET_BUFSIZE]; | |
56 | bool read_poll; | |
57 | bool write_poll; | |
58 | bool using_vnet_hdr; | |
59 | bool has_ufo; | |
60 | bool has_uso; | |
61 | bool enabled; | |
62 | VHostNetState *vhost_net; | |
63 | unsigned host_vnet_hdr_len; | |
64 | Notifier exit; | |
65 | } TAPState; | |
66 | ||
67 | static void launch_script(const char *setup_script, const char *ifname, | |
68 | int fd, Error **errp); | |
69 | ||
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_handler(s->fd, | |
76 | s->read_poll && s->enabled ? tap_send : NULL, | |
77 | s->write_poll && s->enabled ? tap_writable : NULL, | |
78 | s); | |
79 | } | |
80 | ||
81 | static void tap_read_poll(TAPState *s, bool enable) | |
82 | { | |
83 | s->read_poll = enable; | |
84 | tap_update_fd_handler(s); | |
85 | } | |
86 | ||
87 | static void tap_write_poll(TAPState *s, bool enable) | |
88 | { | |
89 | s->write_poll = enable; | |
90 | tap_update_fd_handler(s); | |
91 | } | |
92 | ||
93 | static void tap_writable(void *opaque) | |
94 | { | |
95 | TAPState *s = opaque; | |
96 | ||
97 | tap_write_poll(s, false); | |
98 | ||
99 | qemu_flush_queued_packets(&s->nc); | |
100 | } | |
101 | ||
102 | static ssize_t tap_write_packet(TAPState *s, const struct iovec *iov, int iovcnt) | |
103 | { | |
104 | ssize_t len; | |
105 | ||
106 | len = RETRY_ON_EINTR(writev(s->fd, iov, iovcnt)); | |
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 | g_autofree struct iovec *iov_copy = NULL; | |
122 | struct virtio_net_hdr_mrg_rxbuf hdr = { }; | |
123 | ||
124 | if (s->host_vnet_hdr_len && !s->using_vnet_hdr) { | |
125 | iov_copy = g_new(struct iovec, iovcnt + 1); | |
126 | iov_copy[0].iov_base = &hdr; | |
127 | iov_copy[0].iov_len = s->host_vnet_hdr_len; | |
128 | memcpy(&iov_copy[1], iov, iovcnt * sizeof(*iov)); | |
129 | iovp = iov_copy; | |
130 | iovcnt++; | |
131 | } | |
132 | ||
133 | return tap_write_packet(s, iovp, iovcnt); | |
134 | } | |
135 | ||
136 | static ssize_t tap_receive_raw(NetClientState *nc, const uint8_t *buf, size_t size) | |
137 | { | |
138 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
139 | struct iovec iov[2]; | |
140 | int iovcnt = 0; | |
141 | struct virtio_net_hdr_mrg_rxbuf hdr = { }; | |
142 | ||
143 | if (s->host_vnet_hdr_len) { | |
144 | iov[iovcnt].iov_base = &hdr; | |
145 | iov[iovcnt].iov_len = s->host_vnet_hdr_len; | |
146 | iovcnt++; | |
147 | } | |
148 | ||
149 | iov[iovcnt].iov_base = (char *)buf; | |
150 | iov[iovcnt].iov_len = size; | |
151 | iovcnt++; | |
152 | ||
153 | return tap_write_packet(s, iov, iovcnt); | |
154 | } | |
155 | ||
156 | static ssize_t tap_receive(NetClientState *nc, const uint8_t *buf, size_t size) | |
157 | { | |
158 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
159 | struct iovec iov[1]; | |
160 | ||
161 | if (s->host_vnet_hdr_len && !s->using_vnet_hdr) { | |
162 | return tap_receive_raw(nc, buf, size); | |
163 | } | |
164 | ||
165 | iov[0].iov_base = (char *)buf; | |
166 | iov[0].iov_len = size; | |
167 | ||
168 | return tap_write_packet(s, iov, 1); | |
169 | } | |
170 | ||
171 | #ifndef __sun__ | |
172 | ssize_t tap_read_packet(int tapfd, uint8_t *buf, int maxlen) | |
173 | { | |
174 | return read(tapfd, buf, maxlen); | |
175 | } | |
176 | #endif | |
177 | ||
178 | static void tap_send_completed(NetClientState *nc, ssize_t len) | |
179 | { | |
180 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
181 | tap_read_poll(s, true); | |
182 | } | |
183 | ||
184 | static void tap_send(void *opaque) | |
185 | { | |
186 | TAPState *s = opaque; | |
187 | int size; | |
188 | int packets = 0; | |
189 | ||
190 | while (true) { | |
191 | uint8_t *buf = s->buf; | |
192 | uint8_t min_pkt[ETH_ZLEN]; | |
193 | size_t min_pktsz = sizeof(min_pkt); | |
194 | ||
195 | size = tap_read_packet(s->fd, s->buf, sizeof(s->buf)); | |
196 | if (size <= 0) { | |
197 | break; | |
198 | } | |
199 | ||
200 | if (s->host_vnet_hdr_len && !s->using_vnet_hdr) { | |
201 | buf += s->host_vnet_hdr_len; | |
202 | size -= s->host_vnet_hdr_len; | |
203 | } | |
204 | ||
205 | if (net_peer_needs_padding(&s->nc)) { | |
206 | if (eth_pad_short_frame(min_pkt, &min_pktsz, buf, size)) { | |
207 | buf = min_pkt; | |
208 | size = min_pktsz; | |
209 | } | |
210 | } | |
211 | ||
212 | size = qemu_send_packet_async(&s->nc, buf, size, tap_send_completed); | |
213 | if (size == 0) { | |
214 | tap_read_poll(s, false); | |
215 | break; | |
216 | } else if (size < 0) { | |
217 | break; | |
218 | } | |
219 | ||
220 | /* | |
221 | * When the host keeps receiving more packets while tap_send() is | |
222 | * running we can hog the QEMU global mutex. Limit the number of | |
223 | * packets that are processed per tap_send() callback to prevent | |
224 | * stalling the guest. | |
225 | */ | |
226 | packets++; | |
227 | if (packets >= 50) { | |
228 | break; | |
229 | } | |
230 | } | |
231 | } | |
232 | ||
233 | static bool tap_has_ufo(NetClientState *nc) | |
234 | { | |
235 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
236 | ||
237 | assert(nc->info->type == NET_CLIENT_DRIVER_TAP); | |
238 | ||
239 | return s->has_ufo; | |
240 | } | |
241 | ||
242 | static bool tap_has_uso(NetClientState *nc) | |
243 | { | |
244 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
245 | ||
246 | assert(nc->info->type == NET_CLIENT_DRIVER_TAP); | |
247 | ||
248 | return s->has_uso; | |
249 | } | |
250 | ||
251 | static bool tap_has_vnet_hdr(NetClientState *nc) | |
252 | { | |
253 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
254 | ||
255 | assert(nc->info->type == NET_CLIENT_DRIVER_TAP); | |
256 | ||
257 | return !!s->host_vnet_hdr_len; | |
258 | } | |
259 | ||
260 | static bool tap_has_vnet_hdr_len(NetClientState *nc, int len) | |
261 | { | |
262 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
263 | ||
264 | assert(nc->info->type == NET_CLIENT_DRIVER_TAP); | |
265 | ||
266 | return !!tap_probe_vnet_hdr_len(s->fd, len); | |
267 | } | |
268 | ||
269 | static int tap_get_vnet_hdr_len(NetClientState *nc) | |
270 | { | |
271 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
272 | ||
273 | return s->host_vnet_hdr_len; | |
274 | } | |
275 | ||
276 | static void tap_set_vnet_hdr_len(NetClientState *nc, int len) | |
277 | { | |
278 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
279 | ||
280 | assert(nc->info->type == NET_CLIENT_DRIVER_TAP); | |
281 | assert(len == sizeof(struct virtio_net_hdr_mrg_rxbuf) || | |
282 | len == sizeof(struct virtio_net_hdr) || | |
283 | len == sizeof(struct virtio_net_hdr_v1_hash)); | |
284 | ||
285 | tap_fd_set_vnet_hdr_len(s->fd, len); | |
286 | s->host_vnet_hdr_len = len; | |
287 | } | |
288 | ||
289 | static bool tap_get_using_vnet_hdr(NetClientState *nc) | |
290 | { | |
291 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
292 | ||
293 | return s->using_vnet_hdr; | |
294 | } | |
295 | ||
296 | static void tap_using_vnet_hdr(NetClientState *nc, bool using_vnet_hdr) | |
297 | { | |
298 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
299 | ||
300 | assert(nc->info->type == NET_CLIENT_DRIVER_TAP); | |
301 | assert(!!s->host_vnet_hdr_len == using_vnet_hdr); | |
302 | ||
303 | s->using_vnet_hdr = using_vnet_hdr; | |
304 | } | |
305 | ||
306 | static int tap_set_vnet_le(NetClientState *nc, bool is_le) | |
307 | { | |
308 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
309 | ||
310 | return tap_fd_set_vnet_le(s->fd, is_le); | |
311 | } | |
312 | ||
313 | static int tap_set_vnet_be(NetClientState *nc, bool is_be) | |
314 | { | |
315 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
316 | ||
317 | return tap_fd_set_vnet_be(s->fd, is_be); | |
318 | } | |
319 | ||
320 | static void tap_set_offload(NetClientState *nc, int csum, int tso4, | |
321 | int tso6, int ecn, int ufo, int uso4, int uso6) | |
322 | { | |
323 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
324 | if (s->fd < 0) { | |
325 | return; | |
326 | } | |
327 | ||
328 | tap_fd_set_offload(s->fd, csum, tso4, tso6, ecn, ufo, uso4, uso6); | |
329 | } | |
330 | ||
331 | static void tap_exit_notify(Notifier *notifier, void *data) | |
332 | { | |
333 | TAPState *s = container_of(notifier, TAPState, exit); | |
334 | Error *err = NULL; | |
335 | ||
336 | if (s->down_script[0]) { | |
337 | launch_script(s->down_script, s->down_script_arg, s->fd, &err); | |
338 | if (err) { | |
339 | error_report_err(err); | |
340 | } | |
341 | } | |
342 | } | |
343 | ||
344 | static void tap_cleanup(NetClientState *nc) | |
345 | { | |
346 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
347 | ||
348 | if (s->vhost_net) { | |
349 | vhost_net_cleanup(s->vhost_net); | |
350 | g_free(s->vhost_net); | |
351 | s->vhost_net = NULL; | |
352 | } | |
353 | ||
354 | qemu_purge_queued_packets(nc); | |
355 | ||
356 | tap_exit_notify(&s->exit, NULL); | |
357 | qemu_remove_exit_notifier(&s->exit); | |
358 | ||
359 | tap_read_poll(s, false); | |
360 | tap_write_poll(s, false); | |
361 | close(s->fd); | |
362 | s->fd = -1; | |
363 | } | |
364 | ||
365 | static void tap_poll(NetClientState *nc, bool enable) | |
366 | { | |
367 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
368 | tap_read_poll(s, enable); | |
369 | tap_write_poll(s, enable); | |
370 | } | |
371 | ||
372 | static bool tap_set_steering_ebpf(NetClientState *nc, int prog_fd) | |
373 | { | |
374 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
375 | assert(nc->info->type == NET_CLIENT_DRIVER_TAP); | |
376 | ||
377 | return tap_fd_set_steering_ebpf(s->fd, prog_fd) == 0; | |
378 | } | |
379 | ||
380 | int tap_get_fd(NetClientState *nc) | |
381 | { | |
382 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
383 | assert(nc->info->type == NET_CLIENT_DRIVER_TAP); | |
384 | return s->fd; | |
385 | } | |
386 | ||
387 | /* fd support */ | |
388 | ||
389 | static NetClientInfo net_tap_info = { | |
390 | .type = NET_CLIENT_DRIVER_TAP, | |
391 | .size = sizeof(TAPState), | |
392 | .receive = tap_receive, | |
393 | .receive_raw = tap_receive_raw, | |
394 | .receive_iov = tap_receive_iov, | |
395 | .poll = tap_poll, | |
396 | .cleanup = tap_cleanup, | |
397 | .has_ufo = tap_has_ufo, | |
398 | .has_uso = tap_has_uso, | |
399 | .has_vnet_hdr = tap_has_vnet_hdr, | |
400 | .has_vnet_hdr_len = tap_has_vnet_hdr_len, | |
401 | .get_using_vnet_hdr = tap_get_using_vnet_hdr, | |
402 | .using_vnet_hdr = tap_using_vnet_hdr, | |
403 | .set_offload = tap_set_offload, | |
404 | .get_vnet_hdr_len = tap_get_vnet_hdr_len, | |
405 | .set_vnet_hdr_len = tap_set_vnet_hdr_len, | |
406 | .set_vnet_le = tap_set_vnet_le, | |
407 | .set_vnet_be = tap_set_vnet_be, | |
408 | .set_steering_ebpf = tap_set_steering_ebpf, | |
409 | }; | |
410 | ||
411 | static TAPState *net_tap_fd_init(NetClientState *peer, | |
412 | const char *model, | |
413 | const char *name, | |
414 | int fd, | |
415 | int vnet_hdr) | |
416 | { | |
417 | NetClientState *nc; | |
418 | TAPState *s; | |
419 | ||
420 | nc = qemu_new_net_client(&net_tap_info, peer, model, name); | |
421 | ||
422 | s = DO_UPCAST(TAPState, nc, nc); | |
423 | ||
424 | s->fd = fd; | |
425 | s->host_vnet_hdr_len = vnet_hdr ? sizeof(struct virtio_net_hdr) : 0; | |
426 | s->using_vnet_hdr = false; | |
427 | s->has_ufo = tap_probe_has_ufo(s->fd); | |
428 | s->has_uso = tap_probe_has_uso(s->fd); | |
429 | s->enabled = true; | |
430 | tap_set_offload(&s->nc, 0, 0, 0, 0, 0, 0, 0); | |
431 | /* | |
432 | * Make sure host header length is set correctly in tap: | |
433 | * it might have been modified by another instance of qemu. | |
434 | */ | |
435 | if (tap_probe_vnet_hdr_len(s->fd, s->host_vnet_hdr_len)) { | |
436 | tap_fd_set_vnet_hdr_len(s->fd, s->host_vnet_hdr_len); | |
437 | } | |
438 | tap_read_poll(s, true); | |
439 | s->vhost_net = NULL; | |
440 | ||
441 | s->exit.notify = tap_exit_notify; | |
442 | qemu_add_exit_notifier(&s->exit); | |
443 | ||
444 | return s; | |
445 | } | |
446 | ||
447 | static void launch_script(const char *setup_script, const char *ifname, | |
448 | int fd, Error **errp) | |
449 | { | |
450 | int pid, status; | |
451 | char *args[3]; | |
452 | char **parg; | |
453 | ||
454 | /* try to launch network script */ | |
455 | pid = fork(); | |
456 | if (pid < 0) { | |
457 | error_setg_errno(errp, errno, "could not launch network script %s", | |
458 | setup_script); | |
459 | return; | |
460 | } | |
461 | if (pid == 0) { | |
462 | int open_max = sysconf(_SC_OPEN_MAX), i; | |
463 | ||
464 | for (i = 3; i < open_max; i++) { | |
465 | if (i != fd) { | |
466 | close(i); | |
467 | } | |
468 | } | |
469 | parg = args; | |
470 | *parg++ = (char *)setup_script; | |
471 | *parg++ = (char *)ifname; | |
472 | *parg = NULL; | |
473 | execv(setup_script, args); | |
474 | _exit(1); | |
475 | } else { | |
476 | while (waitpid(pid, &status, 0) != pid) { | |
477 | /* loop */ | |
478 | } | |
479 | ||
480 | if (WIFEXITED(status) && WEXITSTATUS(status) == 0) { | |
481 | return; | |
482 | } | |
483 | error_setg(errp, "network script %s failed with status %d", | |
484 | setup_script, status); | |
485 | } | |
486 | } | |
487 | ||
488 | static int recv_fd(int c) | |
489 | { | |
490 | int fd; | |
491 | uint8_t msgbuf[CMSG_SPACE(sizeof(fd))]; | |
492 | struct msghdr msg = { | |
493 | .msg_control = msgbuf, | |
494 | .msg_controllen = sizeof(msgbuf), | |
495 | }; | |
496 | struct cmsghdr *cmsg; | |
497 | struct iovec iov; | |
498 | uint8_t req[1]; | |
499 | ssize_t len; | |
500 | ||
501 | cmsg = CMSG_FIRSTHDR(&msg); | |
502 | cmsg->cmsg_level = SOL_SOCKET; | |
503 | cmsg->cmsg_type = SCM_RIGHTS; | |
504 | cmsg->cmsg_len = CMSG_LEN(sizeof(fd)); | |
505 | msg.msg_controllen = cmsg->cmsg_len; | |
506 | ||
507 | iov.iov_base = req; | |
508 | iov.iov_len = sizeof(req); | |
509 | ||
510 | msg.msg_iov = &iov; | |
511 | msg.msg_iovlen = 1; | |
512 | ||
513 | len = recvmsg(c, &msg, 0); | |
514 | if (len > 0) { | |
515 | memcpy(&fd, CMSG_DATA(cmsg), sizeof(fd)); | |
516 | return fd; | |
517 | } | |
518 | ||
519 | return len; | |
520 | } | |
521 | ||
522 | static int net_bridge_run_helper(const char *helper, const char *bridge, | |
523 | Error **errp) | |
524 | { | |
525 | sigset_t oldmask, mask; | |
526 | g_autofree char *default_helper = NULL; | |
527 | int pid, status; | |
528 | char *args[5]; | |
529 | char **parg; | |
530 | int sv[2]; | |
531 | ||
532 | sigemptyset(&mask); | |
533 | sigaddset(&mask, SIGCHLD); | |
534 | sigprocmask(SIG_BLOCK, &mask, &oldmask); | |
535 | ||
536 | if (!helper) { | |
537 | helper = default_helper = get_relocated_path(DEFAULT_BRIDGE_HELPER); | |
538 | } | |
539 | ||
540 | if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) { | |
541 | error_setg_errno(errp, errno, "socketpair() failed"); | |
542 | return -1; | |
543 | } | |
544 | ||
545 | /* try to launch bridge helper */ | |
546 | pid = fork(); | |
547 | if (pid < 0) { | |
548 | error_setg_errno(errp, errno, "Can't fork bridge helper"); | |
549 | return -1; | |
550 | } | |
551 | if (pid == 0) { | |
552 | int open_max = sysconf(_SC_OPEN_MAX), i; | |
553 | char *fd_buf = NULL; | |
554 | char *br_buf = NULL; | |
555 | char *helper_cmd = NULL; | |
556 | ||
557 | for (i = 3; i < open_max; i++) { | |
558 | if (i != sv[1]) { | |
559 | close(i); | |
560 | } | |
561 | } | |
562 | ||
563 | fd_buf = g_strdup_printf("%s%d", "--fd=", sv[1]); | |
564 | ||
565 | if (strrchr(helper, ' ') || strrchr(helper, '\t')) { | |
566 | /* assume helper is a command */ | |
567 | ||
568 | if (strstr(helper, "--br=") == NULL) { | |
569 | br_buf = g_strdup_printf("%s%s", "--br=", bridge); | |
570 | } | |
571 | ||
572 | helper_cmd = g_strdup_printf("%s %s %s %s", helper, | |
573 | "--use-vnet", fd_buf, br_buf ? br_buf : ""); | |
574 | ||
575 | parg = args; | |
576 | *parg++ = (char *)"sh"; | |
577 | *parg++ = (char *)"-c"; | |
578 | *parg++ = helper_cmd; | |
579 | *parg++ = NULL; | |
580 | ||
581 | execv("/bin/sh", args); | |
582 | g_free(helper_cmd); | |
583 | } else { | |
584 | /* assume helper is just the executable path name */ | |
585 | ||
586 | br_buf = g_strdup_printf("%s%s", "--br=", bridge); | |
587 | ||
588 | parg = args; | |
589 | *parg++ = (char *)helper; | |
590 | *parg++ = (char *)"--use-vnet"; | |
591 | *parg++ = fd_buf; | |
592 | *parg++ = br_buf; | |
593 | *parg++ = NULL; | |
594 | ||
595 | execv(helper, args); | |
596 | } | |
597 | g_free(fd_buf); | |
598 | g_free(br_buf); | |
599 | _exit(1); | |
600 | ||
601 | } else { | |
602 | int fd; | |
603 | int saved_errno; | |
604 | ||
605 | close(sv[1]); | |
606 | ||
607 | fd = RETRY_ON_EINTR(recv_fd(sv[0])); | |
608 | saved_errno = errno; | |
609 | ||
610 | close(sv[0]); | |
611 | ||
612 | while (waitpid(pid, &status, 0) != pid) { | |
613 | /* loop */ | |
614 | } | |
615 | sigprocmask(SIG_SETMASK, &oldmask, NULL); | |
616 | if (fd < 0) { | |
617 | error_setg_errno(errp, saved_errno, | |
618 | "failed to recv file descriptor"); | |
619 | return -1; | |
620 | } | |
621 | if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) { | |
622 | error_setg(errp, "bridge helper failed"); | |
623 | return -1; | |
624 | } | |
625 | return fd; | |
626 | } | |
627 | } | |
628 | ||
629 | int net_init_bridge(const Netdev *netdev, const char *name, | |
630 | NetClientState *peer, Error **errp) | |
631 | { | |
632 | const NetdevBridgeOptions *bridge; | |
633 | const char *helper, *br; | |
634 | TAPState *s; | |
635 | int fd, vnet_hdr; | |
636 | ||
637 | assert(netdev->type == NET_CLIENT_DRIVER_BRIDGE); | |
638 | bridge = &netdev->u.bridge; | |
639 | helper = bridge->helper; | |
640 | br = bridge->br ?: DEFAULT_BRIDGE_INTERFACE; | |
641 | ||
642 | fd = net_bridge_run_helper(helper, br, errp); | |
643 | if (fd == -1) { | |
644 | return -1; | |
645 | } | |
646 | ||
647 | if (!g_unix_set_fd_nonblocking(fd, true, NULL)) { | |
648 | error_setg_errno(errp, errno, "Failed to set FD nonblocking"); | |
649 | return -1; | |
650 | } | |
651 | vnet_hdr = tap_probe_vnet_hdr(fd, errp); | |
652 | if (vnet_hdr < 0) { | |
653 | close(fd); | |
654 | return -1; | |
655 | } | |
656 | s = net_tap_fd_init(peer, "bridge", name, fd, vnet_hdr); | |
657 | ||
658 | qemu_set_info_str(&s->nc, "helper=%s,br=%s", helper, br); | |
659 | ||
660 | return 0; | |
661 | } | |
662 | ||
663 | static int net_tap_init(const NetdevTapOptions *tap, int *vnet_hdr, | |
664 | const char *setup_script, char *ifname, | |
665 | size_t ifname_sz, int mq_required, Error **errp) | |
666 | { | |
667 | Error *err = NULL; | |
668 | int fd, vnet_hdr_required; | |
669 | ||
670 | if (tap->has_vnet_hdr) { | |
671 | *vnet_hdr = tap->vnet_hdr; | |
672 | vnet_hdr_required = *vnet_hdr; | |
673 | } else { | |
674 | *vnet_hdr = 1; | |
675 | vnet_hdr_required = 0; | |
676 | } | |
677 | ||
678 | fd = RETRY_ON_EINTR(tap_open(ifname, ifname_sz, vnet_hdr, vnet_hdr_required, | |
679 | mq_required, errp)); | |
680 | if (fd < 0) { | |
681 | return -1; | |
682 | } | |
683 | ||
684 | if (setup_script && | |
685 | setup_script[0] != '\0' && | |
686 | strcmp(setup_script, "no") != 0) { | |
687 | launch_script(setup_script, ifname, fd, &err); | |
688 | if (err) { | |
689 | error_propagate(errp, err); | |
690 | close(fd); | |
691 | return -1; | |
692 | } | |
693 | } | |
694 | ||
695 | return fd; | |
696 | } | |
697 | ||
698 | #define MAX_TAP_QUEUES 1024 | |
699 | ||
700 | static void net_init_tap_one(const NetdevTapOptions *tap, NetClientState *peer, | |
701 | const char *model, const char *name, | |
702 | const char *ifname, const char *script, | |
703 | const char *downscript, const char *vhostfdname, | |
704 | int vnet_hdr, int fd, Error **errp) | |
705 | { | |
706 | Error *err = NULL; | |
707 | TAPState *s = net_tap_fd_init(peer, model, name, fd, vnet_hdr); | |
708 | int vhostfd; | |
709 | ||
710 | tap_set_sndbuf(s->fd, tap, &err); | |
711 | if (err) { | |
712 | error_propagate(errp, err); | |
713 | goto failed; | |
714 | } | |
715 | ||
716 | if (tap->fd || tap->fds) { | |
717 | qemu_set_info_str(&s->nc, "fd=%d", fd); | |
718 | } else if (tap->helper) { | |
719 | qemu_set_info_str(&s->nc, "helper=%s", tap->helper); | |
720 | } else { | |
721 | qemu_set_info_str(&s->nc, "ifname=%s,script=%s,downscript=%s", ifname, | |
722 | script, downscript); | |
723 | ||
724 | if (strcmp(downscript, "no") != 0) { | |
725 | snprintf(s->down_script, sizeof(s->down_script), "%s", downscript); | |
726 | snprintf(s->down_script_arg, sizeof(s->down_script_arg), | |
727 | "%s", ifname); | |
728 | } | |
729 | } | |
730 | ||
731 | if (tap->has_vhost ? tap->vhost : | |
732 | vhostfdname || (tap->has_vhostforce && tap->vhostforce)) { | |
733 | VhostNetOptions options; | |
734 | ||
735 | options.backend_type = VHOST_BACKEND_TYPE_KERNEL; | |
736 | options.net_backend = &s->nc; | |
737 | if (tap->has_poll_us) { | |
738 | options.busyloop_timeout = tap->poll_us; | |
739 | } else { | |
740 | options.busyloop_timeout = 0; | |
741 | } | |
742 | ||
743 | if (vhostfdname) { | |
744 | vhostfd = monitor_fd_param(monitor_cur(), vhostfdname, &err); | |
745 | if (vhostfd == -1) { | |
746 | if (tap->has_vhostforce && tap->vhostforce) { | |
747 | error_propagate(errp, err); | |
748 | } else { | |
749 | warn_report_err(err); | |
750 | } | |
751 | goto failed; | |
752 | } | |
753 | if (!g_unix_set_fd_nonblocking(vhostfd, true, NULL)) { | |
754 | error_setg_errno(errp, errno, "%s: Can't use file descriptor %d", | |
755 | name, fd); | |
756 | goto failed; | |
757 | } | |
758 | } else { | |
759 | vhostfd = open("/dev/vhost-net", O_RDWR); | |
760 | if (vhostfd < 0) { | |
761 | if (tap->has_vhostforce && tap->vhostforce) { | |
762 | error_setg_errno(errp, errno, | |
763 | "tap: open vhost char device failed"); | |
764 | } else { | |
765 | warn_report("tap: open vhost char device failed: %s", | |
766 | strerror(errno)); | |
767 | } | |
768 | goto failed; | |
769 | } | |
770 | if (!g_unix_set_fd_nonblocking(vhostfd, true, NULL)) { | |
771 | error_setg_errno(errp, errno, "Failed to set FD nonblocking"); | |
772 | goto failed; | |
773 | } | |
774 | } | |
775 | options.opaque = (void *)(uintptr_t)vhostfd; | |
776 | options.nvqs = 2; | |
777 | ||
778 | s->vhost_net = vhost_net_init(&options); | |
779 | if (!s->vhost_net) { | |
780 | if (tap->has_vhostforce && tap->vhostforce) { | |
781 | error_setg(errp, VHOST_NET_INIT_FAILED); | |
782 | } else { | |
783 | warn_report(VHOST_NET_INIT_FAILED); | |
784 | } | |
785 | goto failed; | |
786 | } | |
787 | } else if (vhostfdname) { | |
788 | error_setg(errp, "vhostfd(s)= is not valid without vhost"); | |
789 | goto failed; | |
790 | } | |
791 | ||
792 | return; | |
793 | ||
794 | failed: | |
795 | qemu_del_net_client(&s->nc); | |
796 | } | |
797 | ||
798 | static int get_fds(char *str, char *fds[], int max) | |
799 | { | |
800 | char *ptr = str, *this; | |
801 | size_t len = strlen(str); | |
802 | int i = 0; | |
803 | ||
804 | while (i < max && ptr < str + len) { | |
805 | this = strchr(ptr, ':'); | |
806 | ||
807 | if (this == NULL) { | |
808 | fds[i] = g_strdup(ptr); | |
809 | } else { | |
810 | fds[i] = g_strndup(ptr, this - ptr); | |
811 | } | |
812 | ||
813 | i++; | |
814 | if (this == NULL) { | |
815 | break; | |
816 | } else { | |
817 | ptr = this + 1; | |
818 | } | |
819 | } | |
820 | ||
821 | return i; | |
822 | } | |
823 | ||
824 | int net_init_tap(const Netdev *netdev, const char *name, | |
825 | NetClientState *peer, Error **errp) | |
826 | { | |
827 | const NetdevTapOptions *tap; | |
828 | int fd, vnet_hdr = 0, i = 0, queues; | |
829 | /* for the no-fd, no-helper case */ | |
830 | const char *script; | |
831 | const char *downscript; | |
832 | Error *err = NULL; | |
833 | const char *vhostfdname; | |
834 | char ifname[128]; | |
835 | int ret = 0; | |
836 | ||
837 | assert(netdev->type == NET_CLIENT_DRIVER_TAP); | |
838 | tap = &netdev->u.tap; | |
839 | queues = tap->has_queues ? tap->queues : 1; | |
840 | vhostfdname = tap->vhostfd; | |
841 | script = tap->script; | |
842 | downscript = tap->downscript; | |
843 | ||
844 | /* QEMU hubs do not support multiqueue tap, in this case peer is set. | |
845 | * For -netdev, peer is always NULL. */ | |
846 | if (peer && (tap->has_queues || tap->fds || tap->vhostfds)) { | |
847 | error_setg(errp, "Multiqueue tap cannot be used with hubs"); | |
848 | return -1; | |
849 | } | |
850 | ||
851 | if (tap->fd) { | |
852 | if (tap->ifname || tap->script || tap->downscript || | |
853 | tap->has_vnet_hdr || tap->helper || tap->has_queues || | |
854 | tap->fds || tap->vhostfds) { | |
855 | error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, " | |
856 | "helper=, queues=, fds=, and vhostfds= " | |
857 | "are invalid with fd="); | |
858 | return -1; | |
859 | } | |
860 | ||
861 | fd = monitor_fd_param(monitor_cur(), tap->fd, errp); | |
862 | if (fd == -1) { | |
863 | return -1; | |
864 | } | |
865 | ||
866 | if (!g_unix_set_fd_nonblocking(fd, true, NULL)) { | |
867 | error_setg_errno(errp, errno, "%s: Can't use file descriptor %d", | |
868 | name, fd); | |
869 | close(fd); | |
870 | return -1; | |
871 | } | |
872 | ||
873 | vnet_hdr = tap_probe_vnet_hdr(fd, errp); | |
874 | if (vnet_hdr < 0) { | |
875 | close(fd); | |
876 | return -1; | |
877 | } | |
878 | ||
879 | net_init_tap_one(tap, peer, "tap", name, NULL, | |
880 | script, downscript, | |
881 | vhostfdname, vnet_hdr, fd, &err); | |
882 | if (err) { | |
883 | error_propagate(errp, err); | |
884 | close(fd); | |
885 | return -1; | |
886 | } | |
887 | } else if (tap->fds) { | |
888 | char **fds; | |
889 | char **vhost_fds; | |
890 | int nfds = 0, nvhosts = 0; | |
891 | ||
892 | if (tap->ifname || tap->script || tap->downscript || | |
893 | tap->has_vnet_hdr || tap->helper || tap->has_queues || | |
894 | tap->vhostfd) { | |
895 | error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, " | |
896 | "helper=, queues=, and vhostfd= " | |
897 | "are invalid with fds="); | |
898 | return -1; | |
899 | } | |
900 | ||
901 | fds = g_new0(char *, MAX_TAP_QUEUES); | |
902 | vhost_fds = g_new0(char *, MAX_TAP_QUEUES); | |
903 | ||
904 | nfds = get_fds(tap->fds, fds, MAX_TAP_QUEUES); | |
905 | if (tap->vhostfds) { | |
906 | nvhosts = get_fds(tap->vhostfds, vhost_fds, MAX_TAP_QUEUES); | |
907 | if (nfds != nvhosts) { | |
908 | error_setg(errp, "The number of fds passed does not match " | |
909 | "the number of vhostfds passed"); | |
910 | ret = -1; | |
911 | goto free_fail; | |
912 | } | |
913 | } | |
914 | ||
915 | for (i = 0; i < nfds; i++) { | |
916 | fd = monitor_fd_param(monitor_cur(), fds[i], errp); | |
917 | if (fd == -1) { | |
918 | ret = -1; | |
919 | goto free_fail; | |
920 | } | |
921 | ||
922 | ret = g_unix_set_fd_nonblocking(fd, true, NULL); | |
923 | if (!ret) { | |
924 | error_setg_errno(errp, errno, "%s: Can't use file descriptor %d", | |
925 | name, fd); | |
926 | goto free_fail; | |
927 | } | |
928 | ||
929 | if (i == 0) { | |
930 | vnet_hdr = tap_probe_vnet_hdr(fd, errp); | |
931 | if (vnet_hdr < 0) { | |
932 | ret = -1; | |
933 | goto free_fail; | |
934 | } | |
935 | } else if (vnet_hdr != tap_probe_vnet_hdr(fd, NULL)) { | |
936 | error_setg(errp, | |
937 | "vnet_hdr not consistent across given tap fds"); | |
938 | ret = -1; | |
939 | goto free_fail; | |
940 | } | |
941 | ||
942 | net_init_tap_one(tap, peer, "tap", name, ifname, | |
943 | script, downscript, | |
944 | tap->vhostfds ? vhost_fds[i] : NULL, | |
945 | vnet_hdr, fd, &err); | |
946 | if (err) { | |
947 | error_propagate(errp, err); | |
948 | ret = -1; | |
949 | goto free_fail; | |
950 | } | |
951 | } | |
952 | ||
953 | free_fail: | |
954 | for (i = 0; i < nvhosts; i++) { | |
955 | g_free(vhost_fds[i]); | |
956 | } | |
957 | for (i = 0; i < nfds; i++) { | |
958 | g_free(fds[i]); | |
959 | } | |
960 | g_free(fds); | |
961 | g_free(vhost_fds); | |
962 | return ret; | |
963 | } else if (tap->helper) { | |
964 | if (tap->ifname || tap->script || tap->downscript || | |
965 | tap->has_vnet_hdr || tap->has_queues || tap->vhostfds) { | |
966 | error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, " | |
967 | "queues=, and vhostfds= are invalid with helper="); | |
968 | return -1; | |
969 | } | |
970 | ||
971 | fd = net_bridge_run_helper(tap->helper, | |
972 | tap->br ?: DEFAULT_BRIDGE_INTERFACE, | |
973 | errp); | |
974 | if (fd == -1) { | |
975 | return -1; | |
976 | } | |
977 | ||
978 | if (!g_unix_set_fd_nonblocking(fd, true, NULL)) { | |
979 | error_setg_errno(errp, errno, "Failed to set FD nonblocking"); | |
980 | return -1; | |
981 | } | |
982 | vnet_hdr = tap_probe_vnet_hdr(fd, errp); | |
983 | if (vnet_hdr < 0) { | |
984 | close(fd); | |
985 | return -1; | |
986 | } | |
987 | ||
988 | net_init_tap_one(tap, peer, "bridge", name, ifname, | |
989 | script, downscript, vhostfdname, | |
990 | vnet_hdr, fd, &err); | |
991 | if (err) { | |
992 | error_propagate(errp, err); | |
993 | close(fd); | |
994 | return -1; | |
995 | } | |
996 | } else { | |
997 | g_autofree char *default_script = NULL; | |
998 | g_autofree char *default_downscript = NULL; | |
999 | if (tap->vhostfds) { | |
1000 | error_setg(errp, "vhostfds= is invalid if fds= wasn't specified"); | |
1001 | return -1; | |
1002 | } | |
1003 | ||
1004 | if (!script) { | |
1005 | script = default_script = get_relocated_path(DEFAULT_NETWORK_SCRIPT); | |
1006 | } | |
1007 | if (!downscript) { | |
1008 | downscript = default_downscript = | |
1009 | get_relocated_path(DEFAULT_NETWORK_DOWN_SCRIPT); | |
1010 | } | |
1011 | ||
1012 | if (tap->ifname) { | |
1013 | pstrcpy(ifname, sizeof ifname, tap->ifname); | |
1014 | } else { | |
1015 | ifname[0] = '\0'; | |
1016 | } | |
1017 | ||
1018 | for (i = 0; i < queues; i++) { | |
1019 | fd = net_tap_init(tap, &vnet_hdr, i >= 1 ? "no" : script, | |
1020 | ifname, sizeof ifname, queues > 1, errp); | |
1021 | if (fd == -1) { | |
1022 | return -1; | |
1023 | } | |
1024 | ||
1025 | if (queues > 1 && i == 0 && !tap->ifname) { | |
1026 | if (tap_fd_get_ifname(fd, ifname)) { | |
1027 | error_setg(errp, "Fail to get ifname"); | |
1028 | close(fd); | |
1029 | return -1; | |
1030 | } | |
1031 | } | |
1032 | ||
1033 | net_init_tap_one(tap, peer, "tap", name, ifname, | |
1034 | i >= 1 ? "no" : script, | |
1035 | i >= 1 ? "no" : downscript, | |
1036 | vhostfdname, vnet_hdr, fd, &err); | |
1037 | if (err) { | |
1038 | error_propagate(errp, err); | |
1039 | close(fd); | |
1040 | return -1; | |
1041 | } | |
1042 | } | |
1043 | } | |
1044 | ||
1045 | return 0; | |
1046 | } | |
1047 | ||
1048 | VHostNetState *tap_get_vhost_net(NetClientState *nc) | |
1049 | { | |
1050 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
1051 | assert(nc->info->type == NET_CLIENT_DRIVER_TAP); | |
1052 | return s->vhost_net; | |
1053 | } | |
1054 | ||
1055 | int tap_enable(NetClientState *nc) | |
1056 | { | |
1057 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
1058 | int ret; | |
1059 | ||
1060 | if (s->enabled) { | |
1061 | return 0; | |
1062 | } else { | |
1063 | ret = tap_fd_enable(s->fd); | |
1064 | if (ret == 0) { | |
1065 | s->enabled = true; | |
1066 | tap_update_fd_handler(s); | |
1067 | } | |
1068 | return ret; | |
1069 | } | |
1070 | } | |
1071 | ||
1072 | int tap_disable(NetClientState *nc) | |
1073 | { | |
1074 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
1075 | int ret; | |
1076 | ||
1077 | if (s->enabled == 0) { | |
1078 | return 0; | |
1079 | } else { | |
1080 | ret = tap_fd_disable(s->fd); | |
1081 | if (ret == 0) { | |
1082 | qemu_purge_queued_packets(nc); | |
1083 | s->enabled = false; | |
1084 | tap_update_fd_handler(s); | |
1085 | } | |
1086 | return ret; | |
1087 | } | |
1088 | } |