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)
442 sigset_t oldmask, mask;
449 sigaddset(&mask, SIGCHLD);
450 sigprocmask(SIG_BLOCK, &mask, &oldmask);
452 if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) {
456 /* try to launch bridge helper */
459 int open_max = sysconf(_SC_OPEN_MAX), i;
461 char br_buf[6+IFNAMSIZ] = {0};
462 char helper_cmd[PATH_MAX + sizeof(fd_buf) + sizeof(br_buf) + 15];
464 for (i = 3; i < open_max; i++) {
470 snprintf(fd_buf, sizeof(fd_buf), "%s%d", "--fd=", sv[1]);
472 if (strrchr(helper, ' ') || strrchr(helper, '\t')) {
473 /* assume helper is a command */
475 if (strstr(helper, "--br=") == NULL) {
476 snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge);
479 snprintf(helper_cmd, sizeof(helper_cmd), "%s %s %s %s",
480 helper, "--use-vnet", fd_buf, br_buf);
483 *parg++ = (char *)"sh";
484 *parg++ = (char *)"-c";
485 *parg++ = helper_cmd;
488 execv("/bin/sh", args);
490 /* assume helper is just the executable path name */
492 snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge);
495 *parg++ = (char *)helper;
496 *parg++ = (char *)"--use-vnet";
505 } else if (pid > 0) {
512 } while (fd == -1 && errno == EINTR);
516 while (waitpid(pid, &status, 0) != pid) {
519 sigprocmask(SIG_SETMASK, &oldmask, NULL);
521 fprintf(stderr, "failed to recv file descriptor\n");
525 if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
529 fprintf(stderr, "failed to launch bridge helper\n");
533 int net_init_bridge(const NetClientOptions *opts, const char *name,
534 NetClientState *peer)
536 const NetdevBridgeOptions *bridge;
537 const char *helper, *br;
542 assert(opts->kind == NET_CLIENT_OPTIONS_KIND_BRIDGE);
543 bridge = opts->bridge;
545 helper = bridge->has_helper ? bridge->helper : DEFAULT_BRIDGE_HELPER;
546 br = bridge->has_br ? bridge->br : DEFAULT_BRIDGE_INTERFACE;
548 fd = net_bridge_run_helper(helper, br);
553 fcntl(fd, F_SETFL, O_NONBLOCK);
555 vnet_hdr = tap_probe_vnet_hdr(fd);
557 s = net_tap_fd_init(peer, "bridge", name, fd, vnet_hdr);
563 snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s,br=%s", helper,
569 static int net_tap_init(const NetdevTapOptions *tap, int *vnet_hdr,
570 const char *setup_script, char *ifname,
571 size_t ifname_sz, int mq_required)
573 int fd, vnet_hdr_required;
575 if (tap->has_vnet_hdr) {
576 *vnet_hdr = tap->vnet_hdr;
577 vnet_hdr_required = *vnet_hdr;
580 vnet_hdr_required = 0;
583 TFR(fd = tap_open(ifname, ifname_sz, vnet_hdr, vnet_hdr_required,
590 setup_script[0] != '\0' &&
591 strcmp(setup_script, "no") != 0 &&
592 launch_script(setup_script, ifname, fd)) {
600 #define MAX_TAP_QUEUES 1024
602 static int net_init_tap_one(const NetdevTapOptions *tap, NetClientState *peer,
603 const char *model, const char *name,
604 const char *ifname, const char *script,
605 const char *downscript, const char *vhostfdname,
606 int vnet_hdr, int fd)
612 s = net_tap_fd_init(peer, model, name, fd, vnet_hdr);
617 if (tap_set_sndbuf(s->fd, tap) < 0) {
621 if (tap->has_fd || tap->has_fds) {
622 snprintf(s->nc.info_str, sizeof(s->nc.info_str), "fd=%d", fd);
623 } else if (tap->has_helper) {
624 snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s",
627 snprintf(s->nc.info_str, sizeof(s->nc.info_str),
628 "ifname=%s,script=%s,downscript=%s", ifname, script,
631 if (strcmp(downscript, "no") != 0) {
632 snprintf(s->down_script, sizeof(s->down_script), "%s", downscript);
633 snprintf(s->down_script_arg, sizeof(s->down_script_arg),
638 if (tap->has_vhost ? tap->vhost :
639 vhostfdname || (tap->has_vhostforce && tap->vhostforce)) {
640 VhostNetOptions options;
642 options.backend_type = VHOST_BACKEND_TYPE_KERNEL;
643 options.net_backend = &s->nc;
644 options.force = tap->has_vhostforce && tap->vhostforce;
646 if (tap->has_vhostfd || tap->has_vhostfds) {
647 vhostfd = monitor_fd_param(cur_mon, vhostfdname, &err);
649 error_report_err(err);
653 vhostfd = open("/dev/vhost-net", O_RDWR);
655 error_report("tap: open vhost char device failed: %s",
660 options.opaque = (void *)(uintptr_t)vhostfd;
662 s->vhost_net = vhost_net_init(&options);
664 error_report("vhost-net requested but could not be initialized");
667 } else if (tap->has_vhostfd || tap->has_vhostfds) {
668 error_report("vhostfd= is not valid without vhost");
675 static int get_fds(char *str, char *fds[], int max)
677 char *ptr = str, *this;
678 size_t len = strlen(str);
681 while (i < max && ptr < str + len) {
682 this = strchr(ptr, ':');
685 fds[i] = g_strdup(ptr);
687 fds[i] = g_strndup(ptr, this - ptr);
701 int net_init_tap(const NetClientOptions *opts, const char *name,
702 NetClientState *peer)
704 const NetdevTapOptions *tap;
705 int fd, vnet_hdr = 0, i = 0, queues;
706 /* for the no-fd, no-helper case */
707 const char *script = NULL; /* suppress wrong "uninit'd use" gcc warning */
708 const char *downscript = NULL;
710 const char *vhostfdname;
713 assert(opts->kind == NET_CLIENT_OPTIONS_KIND_TAP);
715 queues = tap->has_queues ? tap->queues : 1;
716 vhostfdname = tap->has_vhostfd ? tap->vhostfd : NULL;
718 /* QEMU vlans does not support multiqueue tap, in this case peer is set.
719 * For -netdev, peer is always NULL. */
720 if (peer && (tap->has_queues || tap->has_fds || tap->has_vhostfds)) {
721 error_report("Multiqueue tap cannot be used with QEMU vlans");
726 if (tap->has_ifname || tap->has_script || tap->has_downscript ||
727 tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
728 tap->has_fds || tap->has_vhostfds) {
729 error_report("ifname=, script=, downscript=, vnet_hdr=, "
730 "helper=, queues=, fds=, and vhostfds= "
731 "are invalid with fd=");
735 fd = monitor_fd_param(cur_mon, tap->fd, &err);
737 error_report_err(err);
741 fcntl(fd, F_SETFL, O_NONBLOCK);
743 vnet_hdr = tap_probe_vnet_hdr(fd);
745 if (net_init_tap_one(tap, peer, "tap", name, NULL,
747 vhostfdname, vnet_hdr, fd)) {
750 } else if (tap->has_fds) {
751 char *fds[MAX_TAP_QUEUES];
752 char *vhost_fds[MAX_TAP_QUEUES];
755 if (tap->has_ifname || tap->has_script || tap->has_downscript ||
756 tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
758 error_report("ifname=, script=, downscript=, vnet_hdr=, "
759 "helper=, queues=, and vhostfd= "
760 "are invalid with fds=");
764 nfds = get_fds(tap->fds, fds, MAX_TAP_QUEUES);
765 if (tap->has_vhostfds) {
766 nvhosts = get_fds(tap->vhostfds, vhost_fds, MAX_TAP_QUEUES);
767 if (nfds != nvhosts) {
768 error_report("The number of fds passed does not match the "
769 "number of vhostfds passed");
774 for (i = 0; i < nfds; i++) {
775 fd = monitor_fd_param(cur_mon, fds[i], &err);
777 error_report_err(err);
781 fcntl(fd, F_SETFL, O_NONBLOCK);
784 vnet_hdr = tap_probe_vnet_hdr(fd);
785 } else if (vnet_hdr != tap_probe_vnet_hdr(fd)) {
786 error_report("vnet_hdr not consistent across given tap fds");
790 if (net_init_tap_one(tap, peer, "tap", name, ifname,
792 tap->has_vhostfds ? vhost_fds[i] : NULL,
797 } else if (tap->has_helper) {
798 if (tap->has_ifname || tap->has_script || tap->has_downscript ||
799 tap->has_vnet_hdr || tap->has_queues || tap->has_vhostfds) {
800 error_report("ifname=, script=, downscript=, and vnet_hdr= "
801 "queues=, and vhostfds= are invalid with helper=");
805 fd = net_bridge_run_helper(tap->helper, DEFAULT_BRIDGE_INTERFACE);
810 fcntl(fd, F_SETFL, O_NONBLOCK);
811 vnet_hdr = tap_probe_vnet_hdr(fd);
813 if (net_init_tap_one(tap, peer, "bridge", name, ifname,
814 script, downscript, vhostfdname,
820 if (tap->has_vhostfds) {
821 error_report("vhostfds= is invalid if fds= wasn't specified");
824 script = tap->has_script ? tap->script : DEFAULT_NETWORK_SCRIPT;
825 downscript = tap->has_downscript ? tap->downscript :
826 DEFAULT_NETWORK_DOWN_SCRIPT;
828 if (tap->has_ifname) {
829 pstrcpy(ifname, sizeof ifname, tap->ifname);
834 for (i = 0; i < queues; i++) {
835 fd = net_tap_init(tap, &vnet_hdr, i >= 1 ? "no" : script,
836 ifname, sizeof ifname, queues > 1);
841 if (queues > 1 && i == 0 && !tap->has_ifname) {
842 if (tap_fd_get_ifname(fd, ifname)) {
843 error_report("Fail to get ifname");
849 if (net_init_tap_one(tap, peer, "tap", name, ifname,
850 i >= 1 ? "no" : script,
851 i >= 1 ? "no" : downscript,
852 vhostfdname, vnet_hdr, fd)) {
862 VHostNetState *tap_get_vhost_net(NetClientState *nc)
864 TAPState *s = DO_UPCAST(TAPState, nc, nc);
865 assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
869 int tap_enable(NetClientState *nc)
871 TAPState *s = DO_UPCAST(TAPState, nc, nc);
877 ret = tap_fd_enable(s->fd);
880 tap_update_fd_handler(s);
886 int tap_disable(NetClientState *nc)
888 TAPState *s = DO_UPCAST(TAPState, nc, nc);
891 if (s->enabled == 0) {
894 ret = tap_fd_disable(s->fd);
896 qemu_purge_queued_packets(nc);
898 tap_update_fd_handler(s);