4 * Copyright (c) 2003-2008 Fabrice Bellard
5 * Copyright (c) 2009 Red Hat, Inc.
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:
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
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
28 #include "config-host.h"
30 #include <sys/ioctl.h>
33 #include <sys/socket.h>
38 #include "monitor/monitor.h"
39 #include "sysemu/sysemu.h"
40 #include "qemu-common.h"
41 #include "qemu/error-report.h"
45 #include "net/vhost_net.h"
47 typedef struct TAPState {
50 char down_script[1024];
51 char down_script_arg[128];
52 uint8_t buf[NET_BUFSIZE];
58 VHostNetState *vhost_net;
59 unsigned host_vnet_hdr_len;
62 static int launch_script(const char *setup_script, const char *ifname, int fd);
64 static int tap_can_send(void *opaque);
65 static void tap_send(void *opaque);
66 static void tap_writable(void *opaque);
68 static void tap_update_fd_handler(TAPState *s)
70 qemu_set_fd_handler2(s->fd,
71 s->read_poll && s->enabled ? tap_can_send : NULL,
72 s->read_poll && s->enabled ? tap_send : NULL,
73 s->write_poll && s->enabled ? tap_writable : NULL,
77 static void tap_read_poll(TAPState *s, bool enable)
79 s->read_poll = enable;
80 tap_update_fd_handler(s);
83 static void tap_write_poll(TAPState *s, bool enable)
85 s->write_poll = enable;
86 tap_update_fd_handler(s);
89 static void tap_writable(void *opaque)
93 tap_write_poll(s, false);
95 qemu_flush_queued_packets(&s->nc);
98 static ssize_t tap_write_packet(TAPState *s, const struct iovec *iov, int iovcnt)
103 len = writev(s->fd, iov, iovcnt);
104 } while (len == -1 && errno == EINTR);
106 if (len == -1 && errno == EAGAIN) {
107 tap_write_poll(s, true);
114 static ssize_t tap_receive_iov(NetClientState *nc, const struct iovec *iov,
117 TAPState *s = DO_UPCAST(TAPState, nc, nc);
118 const struct iovec *iovp = iov;
119 struct iovec iov_copy[iovcnt + 1];
120 struct virtio_net_hdr_mrg_rxbuf hdr = { };
122 if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
123 iov_copy[0].iov_base = &hdr;
124 iov_copy[0].iov_len = s->host_vnet_hdr_len;
125 memcpy(&iov_copy[1], iov, iovcnt * sizeof(*iov));
130 return tap_write_packet(s, iovp, iovcnt);
133 static ssize_t tap_receive_raw(NetClientState *nc, const uint8_t *buf, size_t size)
135 TAPState *s = DO_UPCAST(TAPState, nc, nc);
138 struct virtio_net_hdr_mrg_rxbuf hdr = { };
140 if (s->host_vnet_hdr_len) {
141 iov[iovcnt].iov_base = &hdr;
142 iov[iovcnt].iov_len = s->host_vnet_hdr_len;
146 iov[iovcnt].iov_base = (char *)buf;
147 iov[iovcnt].iov_len = size;
150 return tap_write_packet(s, iov, iovcnt);
153 static ssize_t tap_receive(NetClientState *nc, const uint8_t *buf, size_t size)
155 TAPState *s = DO_UPCAST(TAPState, nc, nc);
158 if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
159 return tap_receive_raw(nc, buf, size);
162 iov[0].iov_base = (char *)buf;
163 iov[0].iov_len = size;
165 return tap_write_packet(s, iov, 1);
168 static int tap_can_send(void *opaque)
170 TAPState *s = opaque;
172 return qemu_can_send_packet(&s->nc);
176 ssize_t tap_read_packet(int tapfd, uint8_t *buf, int maxlen)
178 return read(tapfd, buf, maxlen);
182 static void tap_send_completed(NetClientState *nc, ssize_t len)
184 TAPState *s = DO_UPCAST(TAPState, nc, nc);
185 tap_read_poll(s, true);
188 static void tap_send(void *opaque)
190 TAPState *s = opaque;
194 uint8_t *buf = s->buf;
196 size = tap_read_packet(s->fd, s->buf, sizeof(s->buf));
201 if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
202 buf += s->host_vnet_hdr_len;
203 size -= s->host_vnet_hdr_len;
206 size = qemu_send_packet_async(&s->nc, buf, size, tap_send_completed);
208 tap_read_poll(s, false);
210 } while (size > 0 && qemu_can_send_packet(&s->nc));
213 bool tap_has_ufo(NetClientState *nc)
215 TAPState *s = DO_UPCAST(TAPState, nc, nc);
217 assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
222 int tap_has_vnet_hdr(NetClientState *nc)
224 TAPState *s = DO_UPCAST(TAPState, nc, nc);
226 assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
228 return !!s->host_vnet_hdr_len;
231 int tap_has_vnet_hdr_len(NetClientState *nc, int len)
233 TAPState *s = DO_UPCAST(TAPState, nc, nc);
235 assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
237 return tap_probe_vnet_hdr_len(s->fd, len);
240 void tap_set_vnet_hdr_len(NetClientState *nc, int len)
242 TAPState *s = DO_UPCAST(TAPState, nc, nc);
244 assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
245 assert(len == sizeof(struct virtio_net_hdr_mrg_rxbuf) ||
246 len == sizeof(struct virtio_net_hdr));
248 tap_fd_set_vnet_hdr_len(s->fd, len);
249 s->host_vnet_hdr_len = len;
252 void tap_using_vnet_hdr(NetClientState *nc, bool using_vnet_hdr)
254 TAPState *s = DO_UPCAST(TAPState, nc, nc);
256 assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
257 assert(!!s->host_vnet_hdr_len == using_vnet_hdr);
259 s->using_vnet_hdr = using_vnet_hdr;
262 void tap_set_offload(NetClientState *nc, int csum, int tso4,
263 int tso6, int ecn, int ufo)
265 TAPState *s = DO_UPCAST(TAPState, nc, nc);
270 tap_fd_set_offload(s->fd, csum, tso4, tso6, ecn, ufo);
273 static void tap_cleanup(NetClientState *nc)
275 TAPState *s = DO_UPCAST(TAPState, nc, nc);
278 vhost_net_cleanup(s->vhost_net);
282 qemu_purge_queued_packets(nc);
284 if (s->down_script[0])
285 launch_script(s->down_script, s->down_script_arg, s->fd);
287 tap_read_poll(s, false);
288 tap_write_poll(s, false);
293 static void tap_poll(NetClientState *nc, bool enable)
295 TAPState *s = DO_UPCAST(TAPState, nc, nc);
296 tap_read_poll(s, enable);
297 tap_write_poll(s, enable);
300 int tap_get_fd(NetClientState *nc)
302 TAPState *s = DO_UPCAST(TAPState, nc, nc);
303 assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
309 static NetClientInfo net_tap_info = {
310 .type = NET_CLIENT_OPTIONS_KIND_TAP,
311 .size = sizeof(TAPState),
312 .receive = tap_receive,
313 .receive_raw = tap_receive_raw,
314 .receive_iov = tap_receive_iov,
316 .cleanup = tap_cleanup,
319 static TAPState *net_tap_fd_init(NetClientState *peer,
328 nc = qemu_new_net_client(&net_tap_info, peer, model, name);
330 s = DO_UPCAST(TAPState, nc, nc);
333 s->host_vnet_hdr_len = vnet_hdr ? sizeof(struct virtio_net_hdr) : 0;
334 s->using_vnet_hdr = false;
335 s->has_ufo = tap_probe_has_ufo(s->fd);
337 tap_set_offload(&s->nc, 0, 0, 0, 0, 0);
339 * Make sure host header length is set correctly in tap:
340 * it might have been modified by another instance of qemu.
342 if (tap_probe_vnet_hdr_len(s->fd, s->host_vnet_hdr_len)) {
343 tap_fd_set_vnet_hdr_len(s->fd, s->host_vnet_hdr_len);
345 tap_read_poll(s, true);
350 static int launch_script(const char *setup_script, const char *ifname, int fd)
356 /* try to launch network script */
359 int open_max = sysconf(_SC_OPEN_MAX), i;
361 for (i = 0; i < open_max; i++) {
362 if (i != STDIN_FILENO &&
363 i != STDOUT_FILENO &&
364 i != STDERR_FILENO &&
370 *parg++ = (char *)setup_script;
371 *parg++ = (char *)ifname;
373 execv(setup_script, args);
375 } else if (pid > 0) {
376 while (waitpid(pid, &status, 0) != pid) {
380 if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
384 fprintf(stderr, "%s: could not launch network script\n", setup_script);
388 static int recv_fd(int c)
391 uint8_t msgbuf[CMSG_SPACE(sizeof(fd))];
392 struct msghdr msg = {
393 .msg_control = msgbuf,
394 .msg_controllen = sizeof(msgbuf),
396 struct cmsghdr *cmsg;
401 cmsg = CMSG_FIRSTHDR(&msg);
402 cmsg->cmsg_level = SOL_SOCKET;
403 cmsg->cmsg_type = SCM_RIGHTS;
404 cmsg->cmsg_len = CMSG_LEN(sizeof(fd));
405 msg.msg_controllen = cmsg->cmsg_len;
408 iov.iov_len = sizeof(req);
413 len = recvmsg(c, &msg, 0);
415 memcpy(&fd, CMSG_DATA(cmsg), sizeof(fd));
422 static int net_bridge_run_helper(const char *helper, const char *bridge)
424 sigset_t oldmask, mask;
431 sigaddset(&mask, SIGCHLD);
432 sigprocmask(SIG_BLOCK, &mask, &oldmask);
434 if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) {
438 /* try to launch bridge helper */
441 int open_max = sysconf(_SC_OPEN_MAX), i;
443 char br_buf[6+IFNAMSIZ] = {0};
444 char helper_cmd[PATH_MAX + sizeof(fd_buf) + sizeof(br_buf) + 15];
446 for (i = 0; i < open_max; i++) {
447 if (i != STDIN_FILENO &&
448 i != STDOUT_FILENO &&
449 i != STDERR_FILENO &&
455 snprintf(fd_buf, sizeof(fd_buf), "%s%d", "--fd=", sv[1]);
457 if (strrchr(helper, ' ') || strrchr(helper, '\t')) {
458 /* assume helper is a command */
460 if (strstr(helper, "--br=") == NULL) {
461 snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge);
464 snprintf(helper_cmd, sizeof(helper_cmd), "%s %s %s %s",
465 helper, "--use-vnet", fd_buf, br_buf);
468 *parg++ = (char *)"sh";
469 *parg++ = (char *)"-c";
470 *parg++ = helper_cmd;
473 execv("/bin/sh", args);
475 /* assume helper is just the executable path name */
477 snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge);
480 *parg++ = (char *)helper;
481 *parg++ = (char *)"--use-vnet";
490 } else if (pid > 0) {
497 } while (fd == -1 && errno == EINTR);
501 while (waitpid(pid, &status, 0) != pid) {
504 sigprocmask(SIG_SETMASK, &oldmask, NULL);
506 fprintf(stderr, "failed to recv file descriptor\n");
510 if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
514 fprintf(stderr, "failed to launch bridge helper\n");
518 int net_init_bridge(const NetClientOptions *opts, const char *name,
519 NetClientState *peer)
521 const NetdevBridgeOptions *bridge;
522 const char *helper, *br;
527 assert(opts->kind == NET_CLIENT_OPTIONS_KIND_BRIDGE);
528 bridge = opts->bridge;
530 helper = bridge->has_helper ? bridge->helper : DEFAULT_BRIDGE_HELPER;
531 br = bridge->has_br ? bridge->br : DEFAULT_BRIDGE_INTERFACE;
533 fd = net_bridge_run_helper(helper, br);
538 fcntl(fd, F_SETFL, O_NONBLOCK);
540 vnet_hdr = tap_probe_vnet_hdr(fd);
542 s = net_tap_fd_init(peer, "bridge", name, fd, vnet_hdr);
548 snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s,br=%s", helper,
554 static int net_tap_init(const NetdevTapOptions *tap, int *vnet_hdr,
555 const char *setup_script, char *ifname,
556 size_t ifname_sz, int mq_required)
558 int fd, vnet_hdr_required;
560 if (tap->has_vnet_hdr) {
561 *vnet_hdr = tap->vnet_hdr;
562 vnet_hdr_required = *vnet_hdr;
565 vnet_hdr_required = 0;
568 TFR(fd = tap_open(ifname, ifname_sz, vnet_hdr, vnet_hdr_required,
575 setup_script[0] != '\0' &&
576 strcmp(setup_script, "no") != 0 &&
577 launch_script(setup_script, ifname, fd)) {
585 #define MAX_TAP_QUEUES 1024
587 static int net_init_tap_one(const NetdevTapOptions *tap, NetClientState *peer,
588 const char *model, const char *name,
589 const char *ifname, const char *script,
590 const char *downscript, const char *vhostfdname,
591 int vnet_hdr, int fd)
595 s = net_tap_fd_init(peer, model, name, fd, vnet_hdr);
601 if (tap_set_sndbuf(s->fd, tap) < 0) {
605 if (tap->has_fd || tap->has_fds) {
606 snprintf(s->nc.info_str, sizeof(s->nc.info_str), "fd=%d", fd);
607 } else if (tap->has_helper) {
608 snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s",
611 snprintf(s->nc.info_str, sizeof(s->nc.info_str),
612 "ifname=%s,script=%s,downscript=%s", ifname, script,
615 if (strcmp(downscript, "no") != 0) {
616 snprintf(s->down_script, sizeof(s->down_script), "%s", downscript);
617 snprintf(s->down_script_arg, sizeof(s->down_script_arg),
622 if (tap->has_vhost ? tap->vhost :
623 vhostfdname || (tap->has_vhostforce && tap->vhostforce)) {
626 if (tap->has_vhostfd || tap->has_vhostfds) {
627 vhostfd = monitor_handle_fd_param(cur_mon, vhostfdname);
635 s->vhost_net = vhost_net_init(&s->nc, vhostfd,
636 tap->has_vhostforce && tap->vhostforce);
638 error_report("vhost-net requested but could not be initialized");
641 } else if (tap->has_vhostfd || tap->has_vhostfds) {
642 error_report("vhostfd= is not valid without vhost");
649 static int get_fds(char *str, char *fds[], int max)
651 char *ptr = str, *this;
652 size_t len = strlen(str);
655 while (i < max && ptr < str + len) {
656 this = strchr(ptr, ':');
659 fds[i] = g_strdup(ptr);
661 fds[i] = g_strndup(ptr, this - ptr);
675 int net_init_tap(const NetClientOptions *opts, const char *name,
676 NetClientState *peer)
678 const NetdevTapOptions *tap;
679 int fd, vnet_hdr = 0, i = 0, queues;
680 /* for the no-fd, no-helper case */
681 const char *script = NULL; /* suppress wrong "uninit'd use" gcc warning */
682 const char *downscript = NULL;
683 const char *vhostfdname;
686 assert(opts->kind == NET_CLIENT_OPTIONS_KIND_TAP);
688 queues = tap->has_queues ? tap->queues : 1;
689 vhostfdname = tap->has_vhostfd ? tap->vhostfd : NULL;
691 /* QEMU vlans does not support multiqueue tap, in this case peer is set.
692 * For -netdev, peer is always NULL. */
693 if (peer && (tap->has_queues || tap->has_fds || tap->has_vhostfds)) {
694 error_report("Multiqueue tap cannot be used with QEMU vlans");
699 if (tap->has_ifname || tap->has_script || tap->has_downscript ||
700 tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
702 error_report("ifname=, script=, downscript=, vnet_hdr=, "
703 "helper=, queues=, and fds= are invalid with fd=");
707 fd = monitor_handle_fd_param(cur_mon, tap->fd);
712 fcntl(fd, F_SETFL, O_NONBLOCK);
714 vnet_hdr = tap_probe_vnet_hdr(fd);
716 if (net_init_tap_one(tap, peer, "tap", name, NULL,
718 vhostfdname, vnet_hdr, fd)) {
721 } else if (tap->has_fds) {
722 char *fds[MAX_TAP_QUEUES];
723 char *vhost_fds[MAX_TAP_QUEUES];
726 if (tap->has_ifname || tap->has_script || tap->has_downscript ||
727 tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
729 error_report("ifname=, script=, downscript=, vnet_hdr=, "
730 "helper=, queues=, and fd= are invalid with fds=");
734 nfds = get_fds(tap->fds, fds, MAX_TAP_QUEUES);
735 if (tap->has_vhostfds) {
736 nvhosts = get_fds(tap->vhostfds, vhost_fds, MAX_TAP_QUEUES);
737 if (nfds != nvhosts) {
738 error_report("The number of fds passed does not match the "
739 "number of vhostfds passed");
744 for (i = 0; i < nfds; i++) {
745 fd = monitor_handle_fd_param(cur_mon, fds[i]);
750 fcntl(fd, F_SETFL, O_NONBLOCK);
753 vnet_hdr = tap_probe_vnet_hdr(fd);
754 } else if (vnet_hdr != tap_probe_vnet_hdr(fd)) {
755 error_report("vnet_hdr not consistent across given tap fds");
759 if (net_init_tap_one(tap, peer, "tap", name, ifname,
761 tap->has_vhostfds ? vhost_fds[i] : NULL,
766 } else if (tap->has_helper) {
767 if (tap->has_ifname || tap->has_script || tap->has_downscript ||
768 tap->has_vnet_hdr || tap->has_queues || tap->has_fds) {
769 error_report("ifname=, script=, downscript=, and vnet_hdr= "
770 "queues=, and fds= are invalid with helper=");
774 fd = net_bridge_run_helper(tap->helper, DEFAULT_BRIDGE_INTERFACE);
779 fcntl(fd, F_SETFL, O_NONBLOCK);
780 vnet_hdr = tap_probe_vnet_hdr(fd);
782 if (net_init_tap_one(tap, peer, "bridge", name, ifname,
783 script, downscript, vhostfdname,
788 script = tap->has_script ? tap->script : DEFAULT_NETWORK_SCRIPT;
789 downscript = tap->has_downscript ? tap->downscript :
790 DEFAULT_NETWORK_DOWN_SCRIPT;
792 if (tap->has_ifname) {
793 pstrcpy(ifname, sizeof ifname, tap->ifname);
798 for (i = 0; i < queues; i++) {
799 fd = net_tap_init(tap, &vnet_hdr, i >= 1 ? "no" : script,
800 ifname, sizeof ifname, queues > 1);
805 if (queues > 1 && i == 0 && !tap->has_ifname) {
806 if (tap_fd_get_ifname(fd, ifname)) {
807 error_report("Fail to get ifname");
812 if (net_init_tap_one(tap, peer, "tap", name, ifname,
813 i >= 1 ? "no" : script,
814 i >= 1 ? "no" : downscript,
815 vhostfdname, vnet_hdr, fd)) {
824 VHostNetState *tap_get_vhost_net(NetClientState *nc)
826 TAPState *s = DO_UPCAST(TAPState, nc, nc);
827 assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
831 int tap_enable(NetClientState *nc)
833 TAPState *s = DO_UPCAST(TAPState, nc, nc);
839 ret = tap_fd_enable(s->fd);
842 tap_update_fd_handler(s);
848 int tap_disable(NetClientState *nc)
850 TAPState *s = DO_UPCAST(TAPState, nc, nc);
853 if (s->enabled == 0) {
856 ret = tap_fd_disable(s->fd);
858 qemu_purge_queued_packets(nc);
860 tap_update_fd_handler(s);