]>
Commit | Line | Data |
---|---|---|
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 | 48 | typedef 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; |
5281d757 MM |
61 | } TAPState; |
62 | ||
ac4fcf56 MA |
63 | static void launch_script(const char *setup_script, const char *ifname, |
64 | int fd, Error **errp); | |
5281d757 | 65 | |
5281d757 MM |
66 | static void tap_send(void *opaque); |
67 | static void tap_writable(void *opaque); | |
68 | ||
69 | static void tap_update_fd_handler(TAPState *s) | |
70 | { | |
82e1cc4b FZ |
71 | qemu_set_fd_handler(s->fd, |
72 | s->read_poll && s->enabled ? tap_send : NULL, | |
73 | s->write_poll && s->enabled ? tap_writable : NULL, | |
74 | s); | |
5281d757 MM |
75 | } |
76 | ||
ec45f083 | 77 | static void tap_read_poll(TAPState *s, bool enable) |
5281d757 | 78 | { |
ec45f083 | 79 | s->read_poll = enable; |
5281d757 MM |
80 | tap_update_fd_handler(s); |
81 | } | |
82 | ||
ec45f083 | 83 | static void tap_write_poll(TAPState *s, bool enable) |
5281d757 | 84 | { |
ec45f083 | 85 | s->write_poll = enable; |
5281d757 MM |
86 | tap_update_fd_handler(s); |
87 | } | |
88 | ||
89 | static void tap_writable(void *opaque) | |
90 | { | |
91 | TAPState *s = opaque; | |
92 | ||
ec45f083 | 93 | tap_write_poll(s, false); |
5281d757 | 94 | |
3e35ba93 | 95 | qemu_flush_queued_packets(&s->nc); |
5281d757 MM |
96 | } |
97 | ||
98 | static ssize_t tap_write_packet(TAPState *s, const struct iovec *iov, int iovcnt) | |
99 | { | |
100 | ssize_t len; | |
101 | ||
102 | do { | |
103 | len = writev(s->fd, iov, iovcnt); | |
104 | } while (len == -1 && errno == EINTR); | |
105 | ||
106 | if (len == -1 && errno == EAGAIN) { | |
ec45f083 | 107 | tap_write_poll(s, true); |
5281d757 MM |
108 | return 0; |
109 | } | |
110 | ||
111 | return len; | |
112 | } | |
113 | ||
4e68f7a0 | 114 | static ssize_t tap_receive_iov(NetClientState *nc, const struct iovec *iov, |
5281d757 MM |
115 | int iovcnt) |
116 | { | |
3e35ba93 | 117 | TAPState *s = DO_UPCAST(TAPState, nc, nc); |
5281d757 MM |
118 | const struct iovec *iovp = iov; |
119 | struct iovec iov_copy[iovcnt + 1]; | |
ef4252b1 | 120 | struct virtio_net_hdr_mrg_rxbuf hdr = { }; |
5281d757 | 121 | |
ef4252b1 | 122 | if (s->host_vnet_hdr_len && !s->using_vnet_hdr) { |
5281d757 | 123 | iov_copy[0].iov_base = &hdr; |
ef4252b1 | 124 | iov_copy[0].iov_len = s->host_vnet_hdr_len; |
5281d757 MM |
125 | memcpy(&iov_copy[1], iov, iovcnt * sizeof(*iov)); |
126 | iovp = iov_copy; | |
127 | iovcnt++; | |
128 | } | |
129 | ||
130 | return tap_write_packet(s, iovp, iovcnt); | |
131 | } | |
132 | ||
4e68f7a0 | 133 | static ssize_t tap_receive_raw(NetClientState *nc, const uint8_t *buf, size_t size) |
5281d757 | 134 | { |
3e35ba93 | 135 | TAPState *s = DO_UPCAST(TAPState, nc, nc); |
5281d757 MM |
136 | struct iovec iov[2]; |
137 | int iovcnt = 0; | |
ef4252b1 | 138 | struct virtio_net_hdr_mrg_rxbuf hdr = { }; |
5281d757 | 139 | |
ef4252b1 | 140 | if (s->host_vnet_hdr_len) { |
5281d757 | 141 | iov[iovcnt].iov_base = &hdr; |
ef4252b1 | 142 | iov[iovcnt].iov_len = s->host_vnet_hdr_len; |
5281d757 MM |
143 | iovcnt++; |
144 | } | |
145 | ||
146 | iov[iovcnt].iov_base = (char *)buf; | |
147 | iov[iovcnt].iov_len = size; | |
148 | iovcnt++; | |
149 | ||
150 | return tap_write_packet(s, iov, iovcnt); | |
151 | } | |
152 | ||
4e68f7a0 | 153 | static ssize_t tap_receive(NetClientState *nc, const uint8_t *buf, size_t size) |
5281d757 | 154 | { |
3e35ba93 | 155 | TAPState *s = DO_UPCAST(TAPState, nc, nc); |
5281d757 MM |
156 | struct iovec iov[1]; |
157 | ||
ef4252b1 | 158 | if (s->host_vnet_hdr_len && !s->using_vnet_hdr) { |
3e35ba93 | 159 | return tap_receive_raw(nc, buf, size); |
5281d757 MM |
160 | } |
161 | ||
162 | iov[0].iov_base = (char *)buf; | |
163 | iov[0].iov_len = size; | |
164 | ||
165 | return tap_write_packet(s, iov, 1); | |
166 | } | |
167 | ||
966ea5ec MM |
168 | #ifndef __sun__ |
169 | ssize_t tap_read_packet(int tapfd, uint8_t *buf, int maxlen) | |
5281d757 MM |
170 | { |
171 | return read(tapfd, buf, maxlen); | |
172 | } | |
173 | #endif | |
174 | ||
4e68f7a0 | 175 | static void tap_send_completed(NetClientState *nc, ssize_t len) |
5281d757 | 176 | { |
3e35ba93 | 177 | TAPState *s = DO_UPCAST(TAPState, nc, nc); |
ec45f083 | 178 | tap_read_poll(s, true); |
5281d757 MM |
179 | } |
180 | ||
181 | static void tap_send(void *opaque) | |
182 | { | |
183 | TAPState *s = opaque; | |
184 | int size; | |
756ae78b | 185 | int packets = 0; |
5281d757 | 186 | |
a90a7425 | 187 | while (true) { |
5819c918 MM |
188 | uint8_t *buf = s->buf; |
189 | ||
190 | size = tap_read_packet(s->fd, s->buf, sizeof(s->buf)); | |
191 | if (size <= 0) { | |
192 | break; | |
193 | } | |
194 | ||
ef4252b1 MT |
195 | if (s->host_vnet_hdr_len && !s->using_vnet_hdr) { |
196 | buf += s->host_vnet_hdr_len; | |
197 | size -= s->host_vnet_hdr_len; | |
5819c918 MM |
198 | } |
199 | ||
3e35ba93 | 200 | size = qemu_send_packet_async(&s->nc, buf, size, tap_send_completed); |
5819c918 | 201 | if (size == 0) { |
ec45f083 | 202 | tap_read_poll(s, false); |
68e5ec64 SH |
203 | break; |
204 | } else if (size < 0) { | |
205 | break; | |
5819c918 | 206 | } |
756ae78b WK |
207 | |
208 | /* | |
209 | * When the host keeps receiving more packets while tap_send() is | |
210 | * running we can hog the QEMU global mutex. Limit the number of | |
211 | * packets that are processed per tap_send() callback to prevent | |
212 | * stalling the guest. | |
213 | */ | |
214 | packets++; | |
215 | if (packets >= 50) { | |
216 | break; | |
217 | } | |
68e5ec64 | 218 | } |
5281d757 MM |
219 | } |
220 | ||
3bac80d3 | 221 | static bool tap_has_ufo(NetClientState *nc) |
5281d757 | 222 | { |
3e35ba93 | 223 | TAPState *s = DO_UPCAST(TAPState, nc, nc); |
5281d757 | 224 | |
2be64a68 | 225 | assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP); |
5281d757 MM |
226 | |
227 | return s->has_ufo; | |
228 | } | |
229 | ||
3bac80d3 | 230 | static bool tap_has_vnet_hdr(NetClientState *nc) |
5281d757 | 231 | { |
3e35ba93 | 232 | TAPState *s = DO_UPCAST(TAPState, nc, nc); |
5281d757 | 233 | |
2be64a68 | 234 | assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP); |
5281d757 | 235 | |
ef4252b1 | 236 | return !!s->host_vnet_hdr_len; |
5281d757 MM |
237 | } |
238 | ||
3bac80d3 | 239 | static bool tap_has_vnet_hdr_len(NetClientState *nc, int len) |
445d892f MT |
240 | { |
241 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
242 | ||
2be64a68 | 243 | assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP); |
445d892f | 244 | |
e96dfd11 | 245 | return !!tap_probe_vnet_hdr_len(s->fd, len); |
445d892f MT |
246 | } |
247 | ||
3bac80d3 | 248 | static void tap_set_vnet_hdr_len(NetClientState *nc, int len) |
445d892f MT |
249 | { |
250 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
251 | ||
2be64a68 | 252 | assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP); |
445d892f MT |
253 | assert(len == sizeof(struct virtio_net_hdr_mrg_rxbuf) || |
254 | len == sizeof(struct virtio_net_hdr)); | |
255 | ||
256 | tap_fd_set_vnet_hdr_len(s->fd, len); | |
257 | s->host_vnet_hdr_len = len; | |
258 | } | |
259 | ||
3bac80d3 | 260 | static void tap_using_vnet_hdr(NetClientState *nc, bool using_vnet_hdr) |
5281d757 | 261 | { |
3e35ba93 | 262 | TAPState *s = DO_UPCAST(TAPState, nc, nc); |
5281d757 | 263 | |
2be64a68 | 264 | assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP); |
ef4252b1 | 265 | assert(!!s->host_vnet_hdr_len == using_vnet_hdr); |
5281d757 MM |
266 | |
267 | s->using_vnet_hdr = using_vnet_hdr; | |
268 | } | |
269 | ||
c80cd6bb GK |
270 | static int tap_set_vnet_le(NetClientState *nc, bool is_le) |
271 | { | |
272 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
273 | ||
274 | return tap_fd_set_vnet_le(s->fd, is_le); | |
275 | } | |
276 | ||
277 | static int tap_set_vnet_be(NetClientState *nc, bool is_be) | |
278 | { | |
279 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
280 | ||
281 | return tap_fd_set_vnet_be(s->fd, is_be); | |
282 | } | |
283 | ||
3bac80d3 | 284 | static void tap_set_offload(NetClientState *nc, int csum, int tso4, |
5281d757 MM |
285 | int tso6, int ecn, int ufo) |
286 | { | |
3e35ba93 | 287 | TAPState *s = DO_UPCAST(TAPState, nc, nc); |
27a6375d MT |
288 | if (s->fd < 0) { |
289 | return; | |
290 | } | |
5281d757 | 291 | |
27a6375d | 292 | tap_fd_set_offload(s->fd, csum, tso4, tso6, ecn, ufo); |
5281d757 MM |
293 | } |
294 | ||
4e68f7a0 | 295 | static void tap_cleanup(NetClientState *nc) |
5281d757 | 296 | { |
3e35ba93 | 297 | TAPState *s = DO_UPCAST(TAPState, nc, nc); |
ac4fcf56 | 298 | Error *err = NULL; |
5281d757 | 299 | |
82b0d80e MT |
300 | if (s->vhost_net) { |
301 | vhost_net_cleanup(s->vhost_net); | |
43849424 | 302 | s->vhost_net = NULL; |
82b0d80e MT |
303 | } |
304 | ||
3e35ba93 | 305 | qemu_purge_queued_packets(nc); |
5281d757 | 306 | |
ac4fcf56 MA |
307 | if (s->down_script[0]) { |
308 | launch_script(s->down_script, s->down_script_arg, s->fd, &err); | |
309 | if (err) { | |
310 | error_report_err(err); | |
311 | } | |
312 | } | |
5281d757 | 313 | |
ec45f083 JW |
314 | tap_read_poll(s, false); |
315 | tap_write_poll(s, false); | |
5281d757 | 316 | close(s->fd); |
27a6375d | 317 | s->fd = -1; |
5281d757 MM |
318 | } |
319 | ||
4e68f7a0 | 320 | static void tap_poll(NetClientState *nc, bool enable) |
ceb69615 MT |
321 | { |
322 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
323 | tap_read_poll(s, enable); | |
324 | tap_write_poll(s, enable); | |
325 | } | |
326 | ||
4e68f7a0 | 327 | int tap_get_fd(NetClientState *nc) |
95d528a2 MT |
328 | { |
329 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
2be64a68 | 330 | assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP); |
95d528a2 MT |
331 | return s->fd; |
332 | } | |
333 | ||
5281d757 MM |
334 | /* fd support */ |
335 | ||
3e35ba93 | 336 | static NetClientInfo net_tap_info = { |
2be64a68 | 337 | .type = NET_CLIENT_OPTIONS_KIND_TAP, |
3e35ba93 MM |
338 | .size = sizeof(TAPState), |
339 | .receive = tap_receive, | |
340 | .receive_raw = tap_receive_raw, | |
341 | .receive_iov = tap_receive_iov, | |
ceb69615 | 342 | .poll = tap_poll, |
3e35ba93 | 343 | .cleanup = tap_cleanup, |
2e753bcc VM |
344 | .has_ufo = tap_has_ufo, |
345 | .has_vnet_hdr = tap_has_vnet_hdr, | |
346 | .has_vnet_hdr_len = tap_has_vnet_hdr_len, | |
347 | .using_vnet_hdr = tap_using_vnet_hdr, | |
348 | .set_offload = tap_set_offload, | |
349 | .set_vnet_hdr_len = tap_set_vnet_hdr_len, | |
c80cd6bb GK |
350 | .set_vnet_le = tap_set_vnet_le, |
351 | .set_vnet_be = tap_set_vnet_be, | |
3e35ba93 MM |
352 | }; |
353 | ||
4e68f7a0 | 354 | static TAPState *net_tap_fd_init(NetClientState *peer, |
5281d757 MM |
355 | const char *model, |
356 | const char *name, | |
357 | int fd, | |
358 | int vnet_hdr) | |
359 | { | |
4e68f7a0 | 360 | NetClientState *nc; |
5281d757 | 361 | TAPState *s; |
5281d757 | 362 | |
ab5f3f84 | 363 | nc = qemu_new_net_client(&net_tap_info, peer, model, name); |
3e35ba93 MM |
364 | |
365 | s = DO_UPCAST(TAPState, nc, nc); | |
366 | ||
5281d757 | 367 | s->fd = fd; |
ef4252b1 | 368 | s->host_vnet_hdr_len = vnet_hdr ? sizeof(struct virtio_net_hdr) : 0; |
ec45f083 | 369 | s->using_vnet_hdr = false; |
9c282718 | 370 | s->has_ufo = tap_probe_has_ufo(s->fd); |
16dbaf90 | 371 | s->enabled = true; |
3e35ba93 | 372 | tap_set_offload(&s->nc, 0, 0, 0, 0, 0); |
58ddcd50 MT |
373 | /* |
374 | * Make sure host header length is set correctly in tap: | |
375 | * it might have been modified by another instance of qemu. | |
376 | */ | |
377 | if (tap_probe_vnet_hdr_len(s->fd, s->host_vnet_hdr_len)) { | |
378 | tap_fd_set_vnet_hdr_len(s->fd, s->host_vnet_hdr_len); | |
379 | } | |
ec45f083 | 380 | tap_read_poll(s, true); |
82b0d80e | 381 | s->vhost_net = NULL; |
5281d757 MM |
382 | return s; |
383 | } | |
384 | ||
ac4fcf56 MA |
385 | static void launch_script(const char *setup_script, const char *ifname, |
386 | int fd, Error **errp) | |
5281d757 | 387 | { |
5281d757 MM |
388 | int pid, status; |
389 | char *args[3]; | |
390 | char **parg; | |
391 | ||
5281d757 MM |
392 | /* try to launch network script */ |
393 | pid = fork(); | |
ac4fcf56 MA |
394 | if (pid < 0) { |
395 | error_setg_errno(errp, errno, "could not launch network script %s", | |
396 | setup_script); | |
397 | return; | |
398 | } | |
5281d757 MM |
399 | if (pid == 0) { |
400 | int open_max = sysconf(_SC_OPEN_MAX), i; | |
401 | ||
13a12f86 PG |
402 | for (i = 3; i < open_max; i++) { |
403 | if (i != fd) { | |
5281d757 MM |
404 | close(i); |
405 | } | |
406 | } | |
407 | parg = args; | |
408 | *parg++ = (char *)setup_script; | |
409 | *parg++ = (char *)ifname; | |
9678d950 | 410 | *parg = NULL; |
5281d757 MM |
411 | execv(setup_script, args); |
412 | _exit(1); | |
ac4fcf56 | 413 | } else { |
5281d757 MM |
414 | while (waitpid(pid, &status, 0) != pid) { |
415 | /* loop */ | |
416 | } | |
5281d757 MM |
417 | |
418 | if (WIFEXITED(status) && WEXITSTATUS(status) == 0) { | |
ac4fcf56 | 419 | return; |
5281d757 | 420 | } |
ac4fcf56 MA |
421 | error_setg(errp, "network script %s failed with status %d", |
422 | setup_script, status); | |
5281d757 | 423 | } |
5281d757 MM |
424 | } |
425 | ||
a7c36ee4 CB |
426 | static int recv_fd(int c) |
427 | { | |
428 | int fd; | |
429 | uint8_t msgbuf[CMSG_SPACE(sizeof(fd))]; | |
430 | struct msghdr msg = { | |
431 | .msg_control = msgbuf, | |
432 | .msg_controllen = sizeof(msgbuf), | |
433 | }; | |
434 | struct cmsghdr *cmsg; | |
435 | struct iovec iov; | |
436 | uint8_t req[1]; | |
437 | ssize_t len; | |
438 | ||
439 | cmsg = CMSG_FIRSTHDR(&msg); | |
440 | cmsg->cmsg_level = SOL_SOCKET; | |
441 | cmsg->cmsg_type = SCM_RIGHTS; | |
442 | cmsg->cmsg_len = CMSG_LEN(sizeof(fd)); | |
443 | msg.msg_controllen = cmsg->cmsg_len; | |
444 | ||
445 | iov.iov_base = req; | |
446 | iov.iov_len = sizeof(req); | |
447 | ||
448 | msg.msg_iov = &iov; | |
449 | msg.msg_iovlen = 1; | |
450 | ||
451 | len = recvmsg(c, &msg, 0); | |
452 | if (len > 0) { | |
453 | memcpy(&fd, CMSG_DATA(cmsg), sizeof(fd)); | |
454 | return fd; | |
455 | } | |
456 | ||
457 | return len; | |
458 | } | |
459 | ||
a8a21be9 MA |
460 | static int net_bridge_run_helper(const char *helper, const char *bridge, |
461 | Error **errp) | |
a7c36ee4 CB |
462 | { |
463 | sigset_t oldmask, mask; | |
464 | int pid, status; | |
465 | char *args[5]; | |
466 | char **parg; | |
467 | int sv[2]; | |
468 | ||
469 | sigemptyset(&mask); | |
470 | sigaddset(&mask, SIGCHLD); | |
471 | sigprocmask(SIG_BLOCK, &mask, &oldmask); | |
472 | ||
473 | if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) { | |
a8a21be9 | 474 | error_setg_errno(errp, errno, "socketpair() failed"); |
a7c36ee4 CB |
475 | return -1; |
476 | } | |
477 | ||
478 | /* try to launch bridge helper */ | |
479 | pid = fork(); | |
a8a21be9 MA |
480 | if (pid < 0) { |
481 | error_setg_errno(errp, errno, "Can't fork bridge helper"); | |
482 | return -1; | |
483 | } | |
a7c36ee4 CB |
484 | if (pid == 0) { |
485 | int open_max = sysconf(_SC_OPEN_MAX), i; | |
486 | char fd_buf[6+10]; | |
487 | char br_buf[6+IFNAMSIZ] = {0}; | |
488 | char helper_cmd[PATH_MAX + sizeof(fd_buf) + sizeof(br_buf) + 15]; | |
489 | ||
13a12f86 PG |
490 | for (i = 3; i < open_max; i++) { |
491 | if (i != sv[1]) { | |
a7c36ee4 CB |
492 | close(i); |
493 | } | |
494 | } | |
495 | ||
496 | snprintf(fd_buf, sizeof(fd_buf), "%s%d", "--fd=", sv[1]); | |
497 | ||
498 | if (strrchr(helper, ' ') || strrchr(helper, '\t')) { | |
499 | /* assume helper is a command */ | |
500 | ||
501 | if (strstr(helper, "--br=") == NULL) { | |
502 | snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge); | |
503 | } | |
504 | ||
505 | snprintf(helper_cmd, sizeof(helper_cmd), "%s %s %s %s", | |
506 | helper, "--use-vnet", fd_buf, br_buf); | |
507 | ||
508 | parg = args; | |
509 | *parg++ = (char *)"sh"; | |
510 | *parg++ = (char *)"-c"; | |
511 | *parg++ = helper_cmd; | |
512 | *parg++ = NULL; | |
513 | ||
514 | execv("/bin/sh", args); | |
515 | } else { | |
516 | /* assume helper is just the executable path name */ | |
517 | ||
518 | snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge); | |
519 | ||
520 | parg = args; | |
521 | *parg++ = (char *)helper; | |
522 | *parg++ = (char *)"--use-vnet"; | |
523 | *parg++ = fd_buf; | |
524 | *parg++ = br_buf; | |
525 | *parg++ = NULL; | |
526 | ||
527 | execv(helper, args); | |
528 | } | |
529 | _exit(1); | |
530 | ||
a8a21be9 | 531 | } else { |
a7c36ee4 | 532 | int fd; |
a8a21be9 | 533 | int saved_errno; |
a7c36ee4 CB |
534 | |
535 | close(sv[1]); | |
536 | ||
537 | do { | |
538 | fd = recv_fd(sv[0]); | |
539 | } while (fd == -1 && errno == EINTR); | |
a8a21be9 | 540 | saved_errno = errno; |
a7c36ee4 CB |
541 | |
542 | close(sv[0]); | |
543 | ||
544 | while (waitpid(pid, &status, 0) != pid) { | |
545 | /* loop */ | |
546 | } | |
547 | sigprocmask(SIG_SETMASK, &oldmask, NULL); | |
548 | if (fd < 0) { | |
a8a21be9 MA |
549 | error_setg_errno(errp, saved_errno, |
550 | "failed to recv file descriptor"); | |
a7c36ee4 CB |
551 | return -1; |
552 | } | |
a8a21be9 MA |
553 | if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) { |
554 | error_setg(errp, "bridge helper failed"); | |
555 | return -1; | |
a7c36ee4 | 556 | } |
a8a21be9 | 557 | return fd; |
a7c36ee4 | 558 | } |
a7c36ee4 CB |
559 | } |
560 | ||
1a0c0958 | 561 | int net_init_bridge(const NetClientOptions *opts, const char *name, |
a30ecde6 | 562 | NetClientState *peer, Error **errp) |
a7c36ee4 | 563 | { |
f79b51b0 LE |
564 | const NetdevBridgeOptions *bridge; |
565 | const char *helper, *br; | |
a7c36ee4 CB |
566 | TAPState *s; |
567 | int fd, vnet_hdr; | |
568 | ||
8d0bcba8 | 569 | assert(opts->type == NET_CLIENT_OPTIONS_KIND_BRIDGE); |
32bafa8f | 570 | bridge = opts->u.bridge.data; |
f79b51b0 LE |
571 | |
572 | helper = bridge->has_helper ? bridge->helper : DEFAULT_BRIDGE_HELPER; | |
573 | br = bridge->has_br ? bridge->br : DEFAULT_BRIDGE_INTERFACE; | |
a7c36ee4 | 574 | |
a8a21be9 | 575 | fd = net_bridge_run_helper(helper, br, errp); |
a7c36ee4 CB |
576 | if (fd == -1) { |
577 | return -1; | |
578 | } | |
579 | ||
580 | fcntl(fd, F_SETFL, O_NONBLOCK); | |
a7c36ee4 | 581 | vnet_hdr = tap_probe_vnet_hdr(fd); |
d33d93b2 | 582 | s = net_tap_fd_init(peer, "bridge", name, fd, vnet_hdr); |
a7c36ee4 | 583 | |
f79b51b0 LE |
584 | snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s,br=%s", helper, |
585 | br); | |
a7c36ee4 CB |
586 | |
587 | return 0; | |
588 | } | |
589 | ||
08c573a8 LE |
590 | static int net_tap_init(const NetdevTapOptions *tap, int *vnet_hdr, |
591 | const char *setup_script, char *ifname, | |
468dd824 | 592 | size_t ifname_sz, int mq_required, Error **errp) |
5281d757 | 593 | { |
ac4fcf56 | 594 | Error *err = NULL; |
5281d757 | 595 | int fd, vnet_hdr_required; |
5281d757 | 596 | |
08c573a8 LE |
597 | if (tap->has_vnet_hdr) { |
598 | *vnet_hdr = tap->vnet_hdr; | |
5281d757 MM |
599 | vnet_hdr_required = *vnet_hdr; |
600 | } else { | |
08c573a8 | 601 | *vnet_hdr = 1; |
5281d757 MM |
602 | vnet_hdr_required = 0; |
603 | } | |
604 | ||
264986e2 | 605 | TFR(fd = tap_open(ifname, ifname_sz, vnet_hdr, vnet_hdr_required, |
468dd824 | 606 | mq_required, errp)); |
5281d757 MM |
607 | if (fd < 0) { |
608 | return -1; | |
609 | } | |
610 | ||
5281d757 MM |
611 | if (setup_script && |
612 | setup_script[0] != '\0' && | |
ac4fcf56 MA |
613 | strcmp(setup_script, "no") != 0) { |
614 | launch_script(setup_script, ifname, fd, &err); | |
615 | if (err) { | |
468dd824 | 616 | error_propagate(errp, err); |
ac4fcf56 MA |
617 | close(fd); |
618 | return -1; | |
619 | } | |
5281d757 MM |
620 | } |
621 | ||
5281d757 MM |
622 | return fd; |
623 | } | |
624 | ||
264986e2 JW |
625 | #define MAX_TAP_QUEUES 1024 |
626 | ||
445f116c MA |
627 | static void net_init_tap_one(const NetdevTapOptions *tap, NetClientState *peer, |
628 | const char *model, const char *name, | |
629 | const char *ifname, const char *script, | |
630 | const char *downscript, const char *vhostfdname, | |
631 | int vnet_hdr, int fd, Error **errp) | |
5193e5fb | 632 | { |
1677f4c6 | 633 | Error *err = NULL; |
da4a4eac | 634 | TAPState *s = net_tap_fd_init(peer, model, name, fd, vnet_hdr); |
81647a65 | 635 | int vhostfd; |
5193e5fb | 636 | |
80b832c3 MA |
637 | tap_set_sndbuf(s->fd, tap, &err); |
638 | if (err) { | |
445f116c MA |
639 | error_propagate(errp, err); |
640 | return; | |
5193e5fb JW |
641 | } |
642 | ||
264986e2 | 643 | if (tap->has_fd || tap->has_fds) { |
5193e5fb JW |
644 | snprintf(s->nc.info_str, sizeof(s->nc.info_str), "fd=%d", fd); |
645 | } else if (tap->has_helper) { | |
646 | snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s", | |
647 | tap->helper); | |
648 | } else { | |
5193e5fb JW |
649 | snprintf(s->nc.info_str, sizeof(s->nc.info_str), |
650 | "ifname=%s,script=%s,downscript=%s", ifname, script, | |
651 | downscript); | |
652 | ||
653 | if (strcmp(downscript, "no") != 0) { | |
654 | snprintf(s->down_script, sizeof(s->down_script), "%s", downscript); | |
655 | snprintf(s->down_script_arg, sizeof(s->down_script_arg), | |
656 | "%s", ifname); | |
657 | } | |
658 | } | |
659 | ||
660 | if (tap->has_vhost ? tap->vhost : | |
661 | vhostfdname || (tap->has_vhostforce && tap->vhostforce)) { | |
81647a65 NN |
662 | VhostNetOptions options; |
663 | ||
1a1bfac9 | 664 | options.backend_type = VHOST_BACKEND_TYPE_KERNEL; |
81647a65 | 665 | options.net_backend = &s->nc; |
5193e5fb | 666 | |
3a2d44f6 | 667 | if (vhostfdname) { |
1677f4c6 | 668 | vhostfd = monitor_fd_param(cur_mon, vhostfdname, &err); |
5193e5fb | 669 | if (vhostfd == -1) { |
445f116c MA |
670 | error_propagate(errp, err); |
671 | return; | |
5193e5fb JW |
672 | } |
673 | } else { | |
81647a65 NN |
674 | vhostfd = open("/dev/vhost-net", O_RDWR); |
675 | if (vhostfd < 0) { | |
445f116c MA |
676 | error_setg_errno(errp, errno, |
677 | "tap: open vhost char device failed"); | |
678 | return; | |
81647a65 | 679 | } |
5193e5fb | 680 | } |
81647a65 | 681 | options.opaque = (void *)(uintptr_t)vhostfd; |
5193e5fb | 682 | |
81647a65 | 683 | s->vhost_net = vhost_net_init(&options); |
5193e5fb | 684 | if (!s->vhost_net) { |
445f116c MA |
685 | error_setg(errp, |
686 | "vhost-net requested but could not be initialized"); | |
687 | return; | |
5193e5fb | 688 | } |
3a2d44f6 | 689 | } else if (vhostfdname) { |
445f116c | 690 | error_setg(errp, "vhostfd= is not valid without vhost"); |
5193e5fb | 691 | } |
5193e5fb JW |
692 | } |
693 | ||
264986e2 JW |
694 | static int get_fds(char *str, char *fds[], int max) |
695 | { | |
696 | char *ptr = str, *this; | |
697 | size_t len = strlen(str); | |
698 | int i = 0; | |
699 | ||
700 | while (i < max && ptr < str + len) { | |
701 | this = strchr(ptr, ':'); | |
702 | ||
703 | if (this == NULL) { | |
704 | fds[i] = g_strdup(ptr); | |
705 | } else { | |
706 | fds[i] = g_strndup(ptr, this - ptr); | |
707 | } | |
708 | ||
709 | i++; | |
710 | if (this == NULL) { | |
711 | break; | |
712 | } else { | |
713 | ptr = this + 1; | |
714 | } | |
715 | } | |
716 | ||
717 | return i; | |
718 | } | |
719 | ||
1a0c0958 | 720 | int net_init_tap(const NetClientOptions *opts, const char *name, |
a30ecde6 | 721 | NetClientState *peer, Error **errp) |
5281d757 | 722 | { |
08c573a8 | 723 | const NetdevTapOptions *tap; |
264986e2 | 724 | int fd, vnet_hdr = 0, i = 0, queues; |
08c573a8 LE |
725 | /* for the no-fd, no-helper case */ |
726 | const char *script = NULL; /* suppress wrong "uninit'd use" gcc warning */ | |
5193e5fb | 727 | const char *downscript = NULL; |
1677f4c6 | 728 | Error *err = NULL; |
264986e2 | 729 | const char *vhostfdname; |
08c573a8 LE |
730 | char ifname[128]; |
731 | ||
8d0bcba8 | 732 | assert(opts->type == NET_CLIENT_OPTIONS_KIND_TAP); |
32bafa8f | 733 | tap = opts->u.tap.data; |
264986e2 JW |
734 | queues = tap->has_queues ? tap->queues : 1; |
735 | vhostfdname = tap->has_vhostfd ? tap->vhostfd : NULL; | |
5281d757 | 736 | |
ce675a75 JW |
737 | /* QEMU vlans does not support multiqueue tap, in this case peer is set. |
738 | * For -netdev, peer is always NULL. */ | |
739 | if (peer && (tap->has_queues || tap->has_fds || tap->has_vhostfds)) { | |
a3088177 | 740 | error_setg(errp, "Multiqueue tap cannot be used with QEMU vlans"); |
ce675a75 JW |
741 | return -1; |
742 | } | |
743 | ||
08c573a8 LE |
744 | if (tap->has_fd) { |
745 | if (tap->has_ifname || tap->has_script || tap->has_downscript || | |
264986e2 | 746 | tap->has_vnet_hdr || tap->has_helper || tap->has_queues || |
c87826a8 | 747 | tap->has_fds || tap->has_vhostfds) { |
a3088177 MA |
748 | error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, " |
749 | "helper=, queues=, fds=, and vhostfds= " | |
750 | "are invalid with fd="); | |
5281d757 MM |
751 | return -1; |
752 | } | |
753 | ||
1677f4c6 | 754 | fd = monitor_fd_param(cur_mon, tap->fd, &err); |
5281d757 | 755 | if (fd == -1) { |
a3088177 | 756 | error_propagate(errp, err); |
5281d757 MM |
757 | return -1; |
758 | } | |
759 | ||
760 | fcntl(fd, F_SETFL, O_NONBLOCK); | |
761 | ||
762 | vnet_hdr = tap_probe_vnet_hdr(fd); | |
a7c36ee4 | 763 | |
445f116c MA |
764 | net_init_tap_one(tap, peer, "tap", name, NULL, |
765 | script, downscript, | |
766 | vhostfdname, vnet_hdr, fd, &err); | |
767 | if (err) { | |
a3088177 | 768 | error_propagate(errp, err); |
264986e2 JW |
769 | return -1; |
770 | } | |
771 | } else if (tap->has_fds) { | |
11196e95 ZJ |
772 | char **fds = g_new(char *, MAX_TAP_QUEUES); |
773 | char **vhost_fds = g_new(char *, MAX_TAP_QUEUES); | |
264986e2 JW |
774 | int nfds, nvhosts; |
775 | ||
776 | if (tap->has_ifname || tap->has_script || tap->has_downscript || | |
777 | tap->has_vnet_hdr || tap->has_helper || tap->has_queues || | |
c87826a8 | 778 | tap->has_vhostfd) { |
a3088177 MA |
779 | error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, " |
780 | "helper=, queues=, and vhostfd= " | |
781 | "are invalid with fds="); | |
264986e2 JW |
782 | return -1; |
783 | } | |
784 | ||
785 | nfds = get_fds(tap->fds, fds, MAX_TAP_QUEUES); | |
786 | if (tap->has_vhostfds) { | |
787 | nvhosts = get_fds(tap->vhostfds, vhost_fds, MAX_TAP_QUEUES); | |
788 | if (nfds != nvhosts) { | |
a3088177 MA |
789 | error_setg(errp, "The number of fds passed does not match " |
790 | "the number of vhostfds passed"); | |
264986e2 JW |
791 | return -1; |
792 | } | |
793 | } | |
794 | ||
795 | for (i = 0; i < nfds; i++) { | |
1677f4c6 | 796 | fd = monitor_fd_param(cur_mon, fds[i], &err); |
264986e2 | 797 | if (fd == -1) { |
a3088177 | 798 | error_propagate(errp, err); |
264986e2 JW |
799 | return -1; |
800 | } | |
801 | ||
802 | fcntl(fd, F_SETFL, O_NONBLOCK); | |
a7c36ee4 | 803 | |
264986e2 JW |
804 | if (i == 0) { |
805 | vnet_hdr = tap_probe_vnet_hdr(fd); | |
806 | } else if (vnet_hdr != tap_probe_vnet_hdr(fd)) { | |
a3088177 MA |
807 | error_setg(errp, |
808 | "vnet_hdr not consistent across given tap fds"); | |
264986e2 JW |
809 | return -1; |
810 | } | |
811 | ||
445f116c MA |
812 | net_init_tap_one(tap, peer, "tap", name, ifname, |
813 | script, downscript, | |
814 | tap->has_vhostfds ? vhost_fds[i] : NULL, | |
815 | vnet_hdr, fd, &err); | |
816 | if (err) { | |
a3088177 | 817 | error_propagate(errp, err); |
264986e2 JW |
818 | return -1; |
819 | } | |
820 | } | |
11196e95 ZJ |
821 | g_free(fds); |
822 | g_free(vhost_fds); | |
08c573a8 LE |
823 | } else if (tap->has_helper) { |
824 | if (tap->has_ifname || tap->has_script || tap->has_downscript || | |
c87826a8 | 825 | tap->has_vnet_hdr || tap->has_queues || tap->has_vhostfds) { |
a3088177 MA |
826 | error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, " |
827 | "queues=, and vhostfds= are invalid with helper="); | |
a7c36ee4 CB |
828 | return -1; |
829 | } | |
830 | ||
a8a21be9 MA |
831 | fd = net_bridge_run_helper(tap->helper, DEFAULT_BRIDGE_INTERFACE, |
832 | errp); | |
a7c36ee4 CB |
833 | if (fd == -1) { |
834 | return -1; | |
835 | } | |
836 | ||
837 | fcntl(fd, F_SETFL, O_NONBLOCK); | |
a7c36ee4 CB |
838 | vnet_hdr = tap_probe_vnet_hdr(fd); |
839 | ||
445f116c MA |
840 | net_init_tap_one(tap, peer, "bridge", name, ifname, |
841 | script, downscript, vhostfdname, | |
842 | vnet_hdr, fd, &err); | |
843 | if (err) { | |
a3088177 | 844 | error_propagate(errp, err); |
84f8f3da | 845 | close(fd); |
264986e2 JW |
846 | return -1; |
847 | } | |
5281d757 | 848 | } else { |
c87826a8 | 849 | if (tap->has_vhostfds) { |
a3088177 | 850 | error_setg(errp, "vhostfds= is invalid if fds= wasn't specified"); |
c87826a8 JW |
851 | return -1; |
852 | } | |
08c573a8 | 853 | script = tap->has_script ? tap->script : DEFAULT_NETWORK_SCRIPT; |
5193e5fb JW |
854 | downscript = tap->has_downscript ? tap->downscript : |
855 | DEFAULT_NETWORK_DOWN_SCRIPT; | |
264986e2 JW |
856 | |
857 | if (tap->has_ifname) { | |
858 | pstrcpy(ifname, sizeof ifname, tap->ifname); | |
859 | } else { | |
860 | ifname[0] = '\0'; | |
929fe497 | 861 | } |
a7c36ee4 | 862 | |
264986e2 JW |
863 | for (i = 0; i < queues; i++) { |
864 | fd = net_tap_init(tap, &vnet_hdr, i >= 1 ? "no" : script, | |
a3088177 | 865 | ifname, sizeof ifname, queues > 1, errp); |
264986e2 JW |
866 | if (fd == -1) { |
867 | return -1; | |
868 | } | |
869 | ||
870 | if (queues > 1 && i == 0 && !tap->has_ifname) { | |
871 | if (tap_fd_get_ifname(fd, ifname)) { | |
a3088177 | 872 | error_setg(errp, "Fail to get ifname"); |
84f8f3da | 873 | close(fd); |
264986e2 JW |
874 | return -1; |
875 | } | |
876 | } | |
877 | ||
445f116c MA |
878 | net_init_tap_one(tap, peer, "tap", name, ifname, |
879 | i >= 1 ? "no" : script, | |
880 | i >= 1 ? "no" : downscript, | |
881 | vhostfdname, vnet_hdr, fd, &err); | |
882 | if (err) { | |
a3088177 | 883 | error_propagate(errp, err); |
84f8f3da | 884 | close(fd); |
264986e2 JW |
885 | return -1; |
886 | } | |
887 | } | |
5281d757 MM |
888 | } |
889 | ||
264986e2 | 890 | return 0; |
5281d757 | 891 | } |
b202554c | 892 | |
4e68f7a0 | 893 | VHostNetState *tap_get_vhost_net(NetClientState *nc) |
b202554c MT |
894 | { |
895 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
2be64a68 | 896 | assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP); |
b202554c MT |
897 | return s->vhost_net; |
898 | } | |
16dbaf90 JW |
899 | |
900 | int tap_enable(NetClientState *nc) | |
901 | { | |
902 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
903 | int ret; | |
904 | ||
905 | if (s->enabled) { | |
906 | return 0; | |
907 | } else { | |
908 | ret = tap_fd_enable(s->fd); | |
909 | if (ret == 0) { | |
910 | s->enabled = true; | |
911 | tap_update_fd_handler(s); | |
912 | } | |
913 | return ret; | |
914 | } | |
915 | } | |
916 | ||
917 | int tap_disable(NetClientState *nc) | |
918 | { | |
919 | TAPState *s = DO_UPCAST(TAPState, nc, nc); | |
920 | int ret; | |
921 | ||
922 | if (s->enabled == 0) { | |
923 | return 0; | |
924 | } else { | |
925 | ret = tap_fd_disable(s->fd); | |
926 | if (ret == 0) { | |
927 | qemu_purge_queued_packets(nc); | |
928 | s->enabled = false; | |
929 | tap_update_fd_handler(s); | |
930 | } | |
931 | return ret; | |
932 | } | |
933 | } |