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 while (qemu_can_send_packet(&s->nc)) {
195 uint8_t *buf = s->buf;
197 size = tap_read_packet(s->fd, s->buf, sizeof(s->buf));
202 if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
203 buf += s->host_vnet_hdr_len;
204 size -= s->host_vnet_hdr_len;
207 size = qemu_send_packet_async(&s->nc, buf, size, tap_send_completed);
209 tap_read_poll(s, false);
211 } else if (size < 0) {
216 * When the host keeps receiving more packets while tap_send() is
217 * running we can hog the QEMU global mutex. Limit the number of
218 * packets that are processed per tap_send() callback to prevent
219 * stalling the guest.
228 static bool tap_has_ufo(NetClientState *nc)
230 TAPState *s = DO_UPCAST(TAPState, nc, nc);
232 assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
237 static bool tap_has_vnet_hdr(NetClientState *nc)
239 TAPState *s = DO_UPCAST(TAPState, nc, nc);
241 assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
243 return !!s->host_vnet_hdr_len;
246 static bool tap_has_vnet_hdr_len(NetClientState *nc, int len)
248 TAPState *s = DO_UPCAST(TAPState, nc, nc);
250 assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
252 return !!tap_probe_vnet_hdr_len(s->fd, len);
255 static void tap_set_vnet_hdr_len(NetClientState *nc, int len)
257 TAPState *s = DO_UPCAST(TAPState, nc, nc);
259 assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
260 assert(len == sizeof(struct virtio_net_hdr_mrg_rxbuf) ||
261 len == sizeof(struct virtio_net_hdr));
263 tap_fd_set_vnet_hdr_len(s->fd, len);
264 s->host_vnet_hdr_len = len;
267 static void tap_using_vnet_hdr(NetClientState *nc, bool using_vnet_hdr)
269 TAPState *s = DO_UPCAST(TAPState, nc, nc);
271 assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
272 assert(!!s->host_vnet_hdr_len == using_vnet_hdr);
274 s->using_vnet_hdr = using_vnet_hdr;
277 static void tap_set_offload(NetClientState *nc, int csum, int tso4,
278 int tso6, int ecn, int ufo)
280 TAPState *s = DO_UPCAST(TAPState, nc, nc);
285 tap_fd_set_offload(s->fd, csum, tso4, tso6, ecn, ufo);
288 static void tap_cleanup(NetClientState *nc)
290 TAPState *s = DO_UPCAST(TAPState, nc, nc);
293 vhost_net_cleanup(s->vhost_net);
297 qemu_purge_queued_packets(nc);
299 if (s->down_script[0])
300 launch_script(s->down_script, s->down_script_arg, s->fd);
302 tap_read_poll(s, false);
303 tap_write_poll(s, false);
308 static void tap_poll(NetClientState *nc, bool enable)
310 TAPState *s = DO_UPCAST(TAPState, nc, nc);
311 tap_read_poll(s, enable);
312 tap_write_poll(s, enable);
315 int tap_get_fd(NetClientState *nc)
317 TAPState *s = DO_UPCAST(TAPState, nc, nc);
318 assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
324 static NetClientInfo net_tap_info = {
325 .type = NET_CLIENT_OPTIONS_KIND_TAP,
326 .size = sizeof(TAPState),
327 .receive = tap_receive,
328 .receive_raw = tap_receive_raw,
329 .receive_iov = tap_receive_iov,
331 .cleanup = tap_cleanup,
332 .has_ufo = tap_has_ufo,
333 .has_vnet_hdr = tap_has_vnet_hdr,
334 .has_vnet_hdr_len = tap_has_vnet_hdr_len,
335 .using_vnet_hdr = tap_using_vnet_hdr,
336 .set_offload = tap_set_offload,
337 .set_vnet_hdr_len = tap_set_vnet_hdr_len,
340 static TAPState *net_tap_fd_init(NetClientState *peer,
349 nc = qemu_new_net_client(&net_tap_info, peer, model, name);
351 s = DO_UPCAST(TAPState, nc, nc);
354 s->host_vnet_hdr_len = vnet_hdr ? sizeof(struct virtio_net_hdr) : 0;
355 s->using_vnet_hdr = false;
356 s->has_ufo = tap_probe_has_ufo(s->fd);
358 tap_set_offload(&s->nc, 0, 0, 0, 0, 0);
360 * Make sure host header length is set correctly in tap:
361 * it might have been modified by another instance of qemu.
363 if (tap_probe_vnet_hdr_len(s->fd, s->host_vnet_hdr_len)) {
364 tap_fd_set_vnet_hdr_len(s->fd, s->host_vnet_hdr_len);
366 tap_read_poll(s, true);
371 static int launch_script(const char *setup_script, const char *ifname, int fd)
377 /* try to launch network script */
380 int open_max = sysconf(_SC_OPEN_MAX), i;
382 for (i = 3; i < open_max; i++) {
388 *parg++ = (char *)setup_script;
389 *parg++ = (char *)ifname;
391 execv(setup_script, args);
393 } else if (pid > 0) {
394 while (waitpid(pid, &status, 0) != pid) {
398 if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
402 fprintf(stderr, "%s: could not launch network script\n", setup_script);
406 static int recv_fd(int c)
409 uint8_t msgbuf[CMSG_SPACE(sizeof(fd))];
410 struct msghdr msg = {
411 .msg_control = msgbuf,
412 .msg_controllen = sizeof(msgbuf),
414 struct cmsghdr *cmsg;
419 cmsg = CMSG_FIRSTHDR(&msg);
420 cmsg->cmsg_level = SOL_SOCKET;
421 cmsg->cmsg_type = SCM_RIGHTS;
422 cmsg->cmsg_len = CMSG_LEN(sizeof(fd));
423 msg.msg_controllen = cmsg->cmsg_len;
426 iov.iov_len = sizeof(req);
431 len = recvmsg(c, &msg, 0);
433 memcpy(&fd, CMSG_DATA(cmsg), sizeof(fd));
440 static int net_bridge_run_helper(const char *helper, const char *bridge,
443 sigset_t oldmask, mask;
450 sigaddset(&mask, SIGCHLD);
451 sigprocmask(SIG_BLOCK, &mask, &oldmask);
453 if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) {
454 error_setg_errno(errp, errno, "socketpair() failed");
458 /* try to launch bridge helper */
461 error_setg_errno(errp, errno, "Can't fork bridge helper");
465 int open_max = sysconf(_SC_OPEN_MAX), i;
467 char br_buf[6+IFNAMSIZ] = {0};
468 char helper_cmd[PATH_MAX + sizeof(fd_buf) + sizeof(br_buf) + 15];
470 for (i = 3; i < open_max; i++) {
476 snprintf(fd_buf, sizeof(fd_buf), "%s%d", "--fd=", sv[1]);
478 if (strrchr(helper, ' ') || strrchr(helper, '\t')) {
479 /* assume helper is a command */
481 if (strstr(helper, "--br=") == NULL) {
482 snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge);
485 snprintf(helper_cmd, sizeof(helper_cmd), "%s %s %s %s",
486 helper, "--use-vnet", fd_buf, br_buf);
489 *parg++ = (char *)"sh";
490 *parg++ = (char *)"-c";
491 *parg++ = helper_cmd;
494 execv("/bin/sh", args);
496 /* assume helper is just the executable path name */
498 snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge);
501 *parg++ = (char *)helper;
502 *parg++ = (char *)"--use-vnet";
519 } while (fd == -1 && errno == EINTR);
524 while (waitpid(pid, &status, 0) != pid) {
527 sigprocmask(SIG_SETMASK, &oldmask, NULL);
529 error_setg_errno(errp, saved_errno,
530 "failed to recv file descriptor");
533 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
534 error_setg(errp, "bridge helper failed");
541 int net_init_bridge(const NetClientOptions *opts, const char *name,
542 NetClientState *peer, Error **errp)
544 const NetdevBridgeOptions *bridge;
545 const char *helper, *br;
549 assert(opts->kind == NET_CLIENT_OPTIONS_KIND_BRIDGE);
550 bridge = opts->bridge;
552 helper = bridge->has_helper ? bridge->helper : DEFAULT_BRIDGE_HELPER;
553 br = bridge->has_br ? bridge->br : DEFAULT_BRIDGE_INTERFACE;
555 fd = net_bridge_run_helper(helper, br, errp);
560 fcntl(fd, F_SETFL, O_NONBLOCK);
561 vnet_hdr = tap_probe_vnet_hdr(fd);
562 s = net_tap_fd_init(peer, "bridge", name, fd, vnet_hdr);
564 snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s,br=%s", helper,
570 static int net_tap_init(const NetdevTapOptions *tap, int *vnet_hdr,
571 const char *setup_script, char *ifname,
572 size_t ifname_sz, int mq_required)
574 int fd, vnet_hdr_required;
576 if (tap->has_vnet_hdr) {
577 *vnet_hdr = tap->vnet_hdr;
578 vnet_hdr_required = *vnet_hdr;
581 vnet_hdr_required = 0;
584 TFR(fd = tap_open(ifname, ifname_sz, vnet_hdr, vnet_hdr_required,
591 setup_script[0] != '\0' &&
592 strcmp(setup_script, "no") != 0 &&
593 launch_script(setup_script, ifname, fd)) {
601 #define MAX_TAP_QUEUES 1024
603 static void net_init_tap_one(const NetdevTapOptions *tap, NetClientState *peer,
604 const char *model, const char *name,
605 const char *ifname, const char *script,
606 const char *downscript, const char *vhostfdname,
607 int vnet_hdr, int fd, Error **errp)
610 TAPState *s = net_tap_fd_init(peer, model, name, fd, vnet_hdr);
613 tap_set_sndbuf(s->fd, tap, &err);
615 error_propagate(errp, err);
619 if (tap->has_fd || tap->has_fds) {
620 snprintf(s->nc.info_str, sizeof(s->nc.info_str), "fd=%d", fd);
621 } else if (tap->has_helper) {
622 snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s",
625 snprintf(s->nc.info_str, sizeof(s->nc.info_str),
626 "ifname=%s,script=%s,downscript=%s", ifname, script,
629 if (strcmp(downscript, "no") != 0) {
630 snprintf(s->down_script, sizeof(s->down_script), "%s", downscript);
631 snprintf(s->down_script_arg, sizeof(s->down_script_arg),
636 if (tap->has_vhost ? tap->vhost :
637 vhostfdname || (tap->has_vhostforce && tap->vhostforce)) {
638 VhostNetOptions options;
640 options.backend_type = VHOST_BACKEND_TYPE_KERNEL;
641 options.net_backend = &s->nc;
642 options.force = tap->has_vhostforce && tap->vhostforce;
644 if (tap->has_vhostfd || tap->has_vhostfds) {
645 vhostfd = monitor_fd_param(cur_mon, vhostfdname, &err);
647 error_propagate(errp, err);
651 vhostfd = open("/dev/vhost-net", O_RDWR);
653 error_setg_errno(errp, errno,
654 "tap: open vhost char device failed");
658 options.opaque = (void *)(uintptr_t)vhostfd;
660 s->vhost_net = vhost_net_init(&options);
663 "vhost-net requested but could not be initialized");
666 } else if (tap->has_vhostfd || tap->has_vhostfds) {
667 error_setg(errp, "vhostfd= is not valid without vhost");
671 static int get_fds(char *str, char *fds[], int max)
673 char *ptr = str, *this;
674 size_t len = strlen(str);
677 while (i < max && ptr < str + len) {
678 this = strchr(ptr, ':');
681 fds[i] = g_strdup(ptr);
683 fds[i] = g_strndup(ptr, this - ptr);
697 int net_init_tap(const NetClientOptions *opts, const char *name,
698 NetClientState *peer, Error **errp)
700 /* FIXME error_setg(errp, ...) on failure */
701 const NetdevTapOptions *tap;
702 int fd, vnet_hdr = 0, i = 0, queues;
703 /* for the no-fd, no-helper case */
704 const char *script = NULL; /* suppress wrong "uninit'd use" gcc warning */
705 const char *downscript = NULL;
707 const char *vhostfdname;
710 assert(opts->kind == NET_CLIENT_OPTIONS_KIND_TAP);
712 queues = tap->has_queues ? tap->queues : 1;
713 vhostfdname = tap->has_vhostfd ? tap->vhostfd : NULL;
715 /* QEMU vlans does not support multiqueue tap, in this case peer is set.
716 * For -netdev, peer is always NULL. */
717 if (peer && (tap->has_queues || tap->has_fds || tap->has_vhostfds)) {
718 error_report("Multiqueue tap cannot be used with QEMU vlans");
723 if (tap->has_ifname || tap->has_script || tap->has_downscript ||
724 tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
725 tap->has_fds || tap->has_vhostfds) {
726 error_report("ifname=, script=, downscript=, vnet_hdr=, "
727 "helper=, queues=, fds=, and vhostfds= "
728 "are invalid with fd=");
732 fd = monitor_fd_param(cur_mon, tap->fd, &err);
734 error_report_err(err);
738 fcntl(fd, F_SETFL, O_NONBLOCK);
740 vnet_hdr = tap_probe_vnet_hdr(fd);
742 net_init_tap_one(tap, peer, "tap", name, NULL,
744 vhostfdname, vnet_hdr, fd, &err);
746 error_report_err(err);
749 } else if (tap->has_fds) {
750 char *fds[MAX_TAP_QUEUES];
751 char *vhost_fds[MAX_TAP_QUEUES];
754 if (tap->has_ifname || tap->has_script || tap->has_downscript ||
755 tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
757 error_report("ifname=, script=, downscript=, vnet_hdr=, "
758 "helper=, queues=, and vhostfd= "
759 "are invalid with fds=");
763 nfds = get_fds(tap->fds, fds, MAX_TAP_QUEUES);
764 if (tap->has_vhostfds) {
765 nvhosts = get_fds(tap->vhostfds, vhost_fds, MAX_TAP_QUEUES);
766 if (nfds != nvhosts) {
767 error_report("The number of fds passed does not match the "
768 "number of vhostfds passed");
773 for (i = 0; i < nfds; i++) {
774 fd = monitor_fd_param(cur_mon, fds[i], &err);
776 error_report_err(err);
780 fcntl(fd, F_SETFL, O_NONBLOCK);
783 vnet_hdr = tap_probe_vnet_hdr(fd);
784 } else if (vnet_hdr != tap_probe_vnet_hdr(fd)) {
785 error_report("vnet_hdr not consistent across given tap fds");
789 net_init_tap_one(tap, peer, "tap", name, ifname,
791 tap->has_vhostfds ? vhost_fds[i] : NULL,
794 error_report_err(err);
798 } else if (tap->has_helper) {
799 if (tap->has_ifname || tap->has_script || tap->has_downscript ||
800 tap->has_vnet_hdr || tap->has_queues || tap->has_vhostfds) {
801 error_report("ifname=, script=, downscript=, and vnet_hdr= "
802 "queues=, and vhostfds= are invalid with helper=");
806 fd = net_bridge_run_helper(tap->helper, DEFAULT_BRIDGE_INTERFACE,
812 fcntl(fd, F_SETFL, O_NONBLOCK);
813 vnet_hdr = tap_probe_vnet_hdr(fd);
815 net_init_tap_one(tap, peer, "bridge", name, ifname,
816 script, downscript, vhostfdname,
819 error_report_err(err);
824 if (tap->has_vhostfds) {
825 error_report("vhostfds= is invalid if fds= wasn't specified");
828 script = tap->has_script ? tap->script : DEFAULT_NETWORK_SCRIPT;
829 downscript = tap->has_downscript ? tap->downscript :
830 DEFAULT_NETWORK_DOWN_SCRIPT;
832 if (tap->has_ifname) {
833 pstrcpy(ifname, sizeof ifname, tap->ifname);
838 for (i = 0; i < queues; i++) {
839 fd = net_tap_init(tap, &vnet_hdr, i >= 1 ? "no" : script,
840 ifname, sizeof ifname, queues > 1);
845 if (queues > 1 && i == 0 && !tap->has_ifname) {
846 if (tap_fd_get_ifname(fd, ifname)) {
847 error_report("Fail to get ifname");
853 net_init_tap_one(tap, peer, "tap", name, ifname,
854 i >= 1 ? "no" : script,
855 i >= 1 ? "no" : downscript,
856 vhostfdname, vnet_hdr, fd, &err);
858 error_report_err(err);
868 VHostNetState *tap_get_vhost_net(NetClientState *nc)
870 TAPState *s = DO_UPCAST(TAPState, nc, nc);
871 assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
875 int tap_enable(NetClientState *nc)
877 TAPState *s = DO_UPCAST(TAPState, nc, nc);
883 ret = tap_fd_enable(s->fd);
886 tap_update_fd_handler(s);
892 int tap_disable(NetClientState *nc)
894 TAPState *s = DO_UPCAST(TAPState, nc, nc);
897 if (s->enabled == 0) {
900 ret = tap_fd_disable(s->fd);
902 qemu_purge_queued_packets(nc);
904 tap_update_fd_handler(s);