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 "hw/vhost_net.h"
47 /* Maximum GSO packet size (64k) plus plenty of room for
48 * the ethernet and virtio_net headers
50 #define TAP_BUFSIZE (4096 + 65536)
52 typedef struct TAPState
{
55 char down_script
[1024];
56 char down_script_arg
[128];
57 uint8_t buf
[TAP_BUFSIZE
];
63 VHostNetState
*vhost_net
;
64 unsigned host_vnet_hdr_len
;
67 static int launch_script(const char *setup_script
, const char *ifname
, int fd
);
69 static int tap_can_send(void *opaque
);
70 static void tap_send(void *opaque
);
71 static void tap_writable(void *opaque
);
73 static void tap_update_fd_handler(TAPState
*s
)
75 qemu_set_fd_handler2(s
->fd
,
76 s
->read_poll
&& s
->enabled
? tap_can_send
: NULL
,
77 s
->read_poll
&& s
->enabled
? tap_send
: NULL
,
78 s
->write_poll
&& s
->enabled
? tap_writable
: NULL
,
82 static void tap_read_poll(TAPState
*s
, bool enable
)
84 s
->read_poll
= enable
;
85 tap_update_fd_handler(s
);
88 static void tap_write_poll(TAPState
*s
, bool enable
)
90 s
->write_poll
= enable
;
91 tap_update_fd_handler(s
);
94 static void tap_writable(void *opaque
)
98 tap_write_poll(s
, false);
100 qemu_flush_queued_packets(&s
->nc
);
103 static ssize_t
tap_write_packet(TAPState
*s
, const struct iovec
*iov
, int iovcnt
)
108 len
= writev(s
->fd
, iov
, iovcnt
);
109 } while (len
== -1 && errno
== EINTR
);
111 if (len
== -1 && errno
== EAGAIN
) {
112 tap_write_poll(s
, true);
119 static ssize_t
tap_receive_iov(NetClientState
*nc
, const struct iovec
*iov
,
122 TAPState
*s
= DO_UPCAST(TAPState
, nc
, nc
);
123 const struct iovec
*iovp
= iov
;
124 struct iovec iov_copy
[iovcnt
+ 1];
125 struct virtio_net_hdr_mrg_rxbuf hdr
= { };
127 if (s
->host_vnet_hdr_len
&& !s
->using_vnet_hdr
) {
128 iov_copy
[0].iov_base
= &hdr
;
129 iov_copy
[0].iov_len
= s
->host_vnet_hdr_len
;
130 memcpy(&iov_copy
[1], iov
, iovcnt
* sizeof(*iov
));
135 return tap_write_packet(s
, iovp
, iovcnt
);
138 static ssize_t
tap_receive_raw(NetClientState
*nc
, const uint8_t *buf
, size_t size
)
140 TAPState
*s
= DO_UPCAST(TAPState
, nc
, nc
);
143 struct virtio_net_hdr_mrg_rxbuf hdr
= { };
145 if (s
->host_vnet_hdr_len
) {
146 iov
[iovcnt
].iov_base
= &hdr
;
147 iov
[iovcnt
].iov_len
= s
->host_vnet_hdr_len
;
151 iov
[iovcnt
].iov_base
= (char *)buf
;
152 iov
[iovcnt
].iov_len
= size
;
155 return tap_write_packet(s
, iov
, iovcnt
);
158 static ssize_t
tap_receive(NetClientState
*nc
, const uint8_t *buf
, size_t size
)
160 TAPState
*s
= DO_UPCAST(TAPState
, nc
, nc
);
163 if (s
->host_vnet_hdr_len
&& !s
->using_vnet_hdr
) {
164 return tap_receive_raw(nc
, buf
, size
);
167 iov
[0].iov_base
= (char *)buf
;
168 iov
[0].iov_len
= size
;
170 return tap_write_packet(s
, iov
, 1);
173 static int tap_can_send(void *opaque
)
175 TAPState
*s
= opaque
;
177 return qemu_can_send_packet(&s
->nc
);
181 ssize_t
tap_read_packet(int tapfd
, uint8_t *buf
, int maxlen
)
183 return read(tapfd
, buf
, maxlen
);
187 static void tap_send_completed(NetClientState
*nc
, ssize_t len
)
189 TAPState
*s
= DO_UPCAST(TAPState
, nc
, nc
);
190 tap_read_poll(s
, true);
193 static void tap_send(void *opaque
)
195 TAPState
*s
= opaque
;
199 uint8_t *buf
= s
->buf
;
201 size
= tap_read_packet(s
->fd
, s
->buf
, sizeof(s
->buf
));
206 if (s
->host_vnet_hdr_len
&& !s
->using_vnet_hdr
) {
207 buf
+= s
->host_vnet_hdr_len
;
208 size
-= s
->host_vnet_hdr_len
;
211 size
= qemu_send_packet_async(&s
->nc
, buf
, size
, tap_send_completed
);
213 tap_read_poll(s
, false);
215 } while (size
> 0 && qemu_can_send_packet(&s
->nc
));
218 bool tap_has_ufo(NetClientState
*nc
)
220 TAPState
*s
= DO_UPCAST(TAPState
, nc
, nc
);
222 assert(nc
->info
->type
== NET_CLIENT_OPTIONS_KIND_TAP
);
227 int tap_has_vnet_hdr(NetClientState
*nc
)
229 TAPState
*s
= DO_UPCAST(TAPState
, nc
, nc
);
231 assert(nc
->info
->type
== NET_CLIENT_OPTIONS_KIND_TAP
);
233 return !!s
->host_vnet_hdr_len
;
236 int tap_has_vnet_hdr_len(NetClientState
*nc
, int len
)
238 TAPState
*s
= DO_UPCAST(TAPState
, nc
, nc
);
240 assert(nc
->info
->type
== NET_CLIENT_OPTIONS_KIND_TAP
);
242 return tap_probe_vnet_hdr_len(s
->fd
, len
);
245 void tap_set_vnet_hdr_len(NetClientState
*nc
, int len
)
247 TAPState
*s
= DO_UPCAST(TAPState
, nc
, nc
);
249 assert(nc
->info
->type
== NET_CLIENT_OPTIONS_KIND_TAP
);
250 assert(len
== sizeof(struct virtio_net_hdr_mrg_rxbuf
) ||
251 len
== sizeof(struct virtio_net_hdr
));
253 tap_fd_set_vnet_hdr_len(s
->fd
, len
);
254 s
->host_vnet_hdr_len
= len
;
257 void tap_using_vnet_hdr(NetClientState
*nc
, bool using_vnet_hdr
)
259 TAPState
*s
= DO_UPCAST(TAPState
, nc
, nc
);
261 assert(nc
->info
->type
== NET_CLIENT_OPTIONS_KIND_TAP
);
262 assert(!!s
->host_vnet_hdr_len
== using_vnet_hdr
);
264 s
->using_vnet_hdr
= using_vnet_hdr
;
267 void tap_set_offload(NetClientState
*nc
, int csum
, int tso4
,
268 int tso6
, int ecn
, int ufo
)
270 TAPState
*s
= DO_UPCAST(TAPState
, nc
, nc
);
275 tap_fd_set_offload(s
->fd
, csum
, tso4
, tso6
, ecn
, ufo
);
278 static void tap_cleanup(NetClientState
*nc
)
280 TAPState
*s
= DO_UPCAST(TAPState
, nc
, nc
);
283 vhost_net_cleanup(s
->vhost_net
);
287 qemu_purge_queued_packets(nc
);
289 if (s
->down_script
[0])
290 launch_script(s
->down_script
, s
->down_script_arg
, s
->fd
);
292 tap_read_poll(s
, false);
293 tap_write_poll(s
, false);
298 static void tap_poll(NetClientState
*nc
, bool enable
)
300 TAPState
*s
= DO_UPCAST(TAPState
, nc
, nc
);
301 tap_read_poll(s
, enable
);
302 tap_write_poll(s
, enable
);
305 int tap_get_fd(NetClientState
*nc
)
307 TAPState
*s
= DO_UPCAST(TAPState
, nc
, nc
);
308 assert(nc
->info
->type
== NET_CLIENT_OPTIONS_KIND_TAP
);
314 static NetClientInfo net_tap_info
= {
315 .type
= NET_CLIENT_OPTIONS_KIND_TAP
,
316 .size
= sizeof(TAPState
),
317 .receive
= tap_receive
,
318 .receive_raw
= tap_receive_raw
,
319 .receive_iov
= tap_receive_iov
,
321 .cleanup
= tap_cleanup
,
324 static TAPState
*net_tap_fd_init(NetClientState
*peer
,
333 nc
= qemu_new_net_client(&net_tap_info
, peer
, model
, name
);
335 s
= DO_UPCAST(TAPState
, nc
, nc
);
338 s
->host_vnet_hdr_len
= vnet_hdr
? sizeof(struct virtio_net_hdr
) : 0;
339 s
->using_vnet_hdr
= false;
340 s
->has_ufo
= tap_probe_has_ufo(s
->fd
);
342 tap_set_offload(&s
->nc
, 0, 0, 0, 0, 0);
344 * Make sure host header length is set correctly in tap:
345 * it might have been modified by another instance of qemu.
347 if (tap_probe_vnet_hdr_len(s
->fd
, s
->host_vnet_hdr_len
)) {
348 tap_fd_set_vnet_hdr_len(s
->fd
, s
->host_vnet_hdr_len
);
350 tap_read_poll(s
, true);
355 static int launch_script(const char *setup_script
, const char *ifname
, int fd
)
361 /* try to launch network script */
364 int open_max
= sysconf(_SC_OPEN_MAX
), i
;
366 for (i
= 0; i
< open_max
; i
++) {
367 if (i
!= STDIN_FILENO
&&
368 i
!= STDOUT_FILENO
&&
369 i
!= STDERR_FILENO
&&
375 *parg
++ = (char *)setup_script
;
376 *parg
++ = (char *)ifname
;
378 execv(setup_script
, args
);
380 } else if (pid
> 0) {
381 while (waitpid(pid
, &status
, 0) != pid
) {
385 if (WIFEXITED(status
) && WEXITSTATUS(status
) == 0) {
389 fprintf(stderr
, "%s: could not launch network script\n", setup_script
);
393 static int recv_fd(int c
)
396 uint8_t msgbuf
[CMSG_SPACE(sizeof(fd
))];
397 struct msghdr msg
= {
398 .msg_control
= msgbuf
,
399 .msg_controllen
= sizeof(msgbuf
),
401 struct cmsghdr
*cmsg
;
406 cmsg
= CMSG_FIRSTHDR(&msg
);
407 cmsg
->cmsg_level
= SOL_SOCKET
;
408 cmsg
->cmsg_type
= SCM_RIGHTS
;
409 cmsg
->cmsg_len
= CMSG_LEN(sizeof(fd
));
410 msg
.msg_controllen
= cmsg
->cmsg_len
;
413 iov
.iov_len
= sizeof(req
);
418 len
= recvmsg(c
, &msg
, 0);
420 memcpy(&fd
, CMSG_DATA(cmsg
), sizeof(fd
));
427 static int net_bridge_run_helper(const char *helper
, const char *bridge
)
429 sigset_t oldmask
, mask
;
436 sigaddset(&mask
, SIGCHLD
);
437 sigprocmask(SIG_BLOCK
, &mask
, &oldmask
);
439 if (socketpair(PF_UNIX
, SOCK_STREAM
, 0, sv
) == -1) {
443 /* try to launch bridge helper */
446 int open_max
= sysconf(_SC_OPEN_MAX
), i
;
448 char br_buf
[6+IFNAMSIZ
] = {0};
449 char helper_cmd
[PATH_MAX
+ sizeof(fd_buf
) + sizeof(br_buf
) + 15];
451 for (i
= 0; i
< open_max
; i
++) {
452 if (i
!= STDIN_FILENO
&&
453 i
!= STDOUT_FILENO
&&
454 i
!= STDERR_FILENO
&&
460 snprintf(fd_buf
, sizeof(fd_buf
), "%s%d", "--fd=", sv
[1]);
462 if (strrchr(helper
, ' ') || strrchr(helper
, '\t')) {
463 /* assume helper is a command */
465 if (strstr(helper
, "--br=") == NULL
) {
466 snprintf(br_buf
, sizeof(br_buf
), "%s%s", "--br=", bridge
);
469 snprintf(helper_cmd
, sizeof(helper_cmd
), "%s %s %s %s",
470 helper
, "--use-vnet", fd_buf
, br_buf
);
473 *parg
++ = (char *)"sh";
474 *parg
++ = (char *)"-c";
475 *parg
++ = helper_cmd
;
478 execv("/bin/sh", args
);
480 /* assume helper is just the executable path name */
482 snprintf(br_buf
, sizeof(br_buf
), "%s%s", "--br=", bridge
);
485 *parg
++ = (char *)helper
;
486 *parg
++ = (char *)"--use-vnet";
495 } else if (pid
> 0) {
502 } while (fd
== -1 && errno
== EINTR
);
506 while (waitpid(pid
, &status
, 0) != pid
) {
509 sigprocmask(SIG_SETMASK
, &oldmask
, NULL
);
511 fprintf(stderr
, "failed to recv file descriptor\n");
515 if (WIFEXITED(status
) && WEXITSTATUS(status
) == 0) {
519 fprintf(stderr
, "failed to launch bridge helper\n");
523 int net_init_bridge(const NetClientOptions
*opts
, const char *name
,
524 NetClientState
*peer
)
526 const NetdevBridgeOptions
*bridge
;
527 const char *helper
, *br
;
532 assert(opts
->kind
== NET_CLIENT_OPTIONS_KIND_BRIDGE
);
533 bridge
= opts
->bridge
;
535 helper
= bridge
->has_helper
? bridge
->helper
: DEFAULT_BRIDGE_HELPER
;
536 br
= bridge
->has_br
? bridge
->br
: DEFAULT_BRIDGE_INTERFACE
;
538 fd
= net_bridge_run_helper(helper
, br
);
543 fcntl(fd
, F_SETFL
, O_NONBLOCK
);
545 vnet_hdr
= tap_probe_vnet_hdr(fd
);
547 s
= net_tap_fd_init(peer
, "bridge", name
, fd
, vnet_hdr
);
553 snprintf(s
->nc
.info_str
, sizeof(s
->nc
.info_str
), "helper=%s,br=%s", helper
,
559 static int net_tap_init(const NetdevTapOptions
*tap
, int *vnet_hdr
,
560 const char *setup_script
, char *ifname
,
561 size_t ifname_sz
, int mq_required
)
563 int fd
, vnet_hdr_required
;
565 if (tap
->has_vnet_hdr
) {
566 *vnet_hdr
= tap
->vnet_hdr
;
567 vnet_hdr_required
= *vnet_hdr
;
570 vnet_hdr_required
= 0;
573 TFR(fd
= tap_open(ifname
, ifname_sz
, vnet_hdr
, vnet_hdr_required
,
580 setup_script
[0] != '\0' &&
581 strcmp(setup_script
, "no") != 0 &&
582 launch_script(setup_script
, ifname
, fd
)) {
590 #define MAX_TAP_QUEUES 1024
592 static int net_init_tap_one(const NetdevTapOptions
*tap
, NetClientState
*peer
,
593 const char *model
, const char *name
,
594 const char *ifname
, const char *script
,
595 const char *downscript
, const char *vhostfdname
,
596 int vnet_hdr
, int fd
)
600 s
= net_tap_fd_init(peer
, model
, name
, fd
, vnet_hdr
);
606 if (tap_set_sndbuf(s
->fd
, tap
) < 0) {
610 if (tap
->has_fd
|| tap
->has_fds
) {
611 snprintf(s
->nc
.info_str
, sizeof(s
->nc
.info_str
), "fd=%d", fd
);
612 } else if (tap
->has_helper
) {
613 snprintf(s
->nc
.info_str
, sizeof(s
->nc
.info_str
), "helper=%s",
616 snprintf(s
->nc
.info_str
, sizeof(s
->nc
.info_str
),
617 "ifname=%s,script=%s,downscript=%s", ifname
, script
,
620 if (strcmp(downscript
, "no") != 0) {
621 snprintf(s
->down_script
, sizeof(s
->down_script
), "%s", downscript
);
622 snprintf(s
->down_script_arg
, sizeof(s
->down_script_arg
),
627 if (tap
->has_vhost
? tap
->vhost
:
628 vhostfdname
|| (tap
->has_vhostforce
&& tap
->vhostforce
)) {
631 if (tap
->has_vhostfd
) {
632 vhostfd
= monitor_handle_fd_param(cur_mon
, vhostfdname
);
640 s
->vhost_net
= vhost_net_init(&s
->nc
, vhostfd
,
641 tap
->has_vhostforce
&& tap
->vhostforce
);
643 error_report("vhost-net requested but could not be initialized");
646 } else if (tap
->has_vhostfd
|| tap
->has_vhostfds
) {
647 error_report("vhostfd= is not valid without vhost");
654 static int get_fds(char *str
, char *fds
[], int max
)
656 char *ptr
= str
, *this;
657 size_t len
= strlen(str
);
660 while (i
< max
&& ptr
< str
+ len
) {
661 this = strchr(ptr
, ':');
664 fds
[i
] = g_strdup(ptr
);
666 fds
[i
] = g_strndup(ptr
, this - ptr
);
680 int net_init_tap(const NetClientOptions
*opts
, const char *name
,
681 NetClientState
*peer
)
683 const NetdevTapOptions
*tap
;
684 int fd
, vnet_hdr
= 0, i
= 0, queues
;
685 /* for the no-fd, no-helper case */
686 const char *script
= NULL
; /* suppress wrong "uninit'd use" gcc warning */
687 const char *downscript
= NULL
;
688 const char *vhostfdname
;
691 assert(opts
->kind
== NET_CLIENT_OPTIONS_KIND_TAP
);
693 queues
= tap
->has_queues
? tap
->queues
: 1;
694 vhostfdname
= tap
->has_vhostfd
? tap
->vhostfd
: NULL
;
696 /* QEMU vlans does not support multiqueue tap, in this case peer is set.
697 * For -netdev, peer is always NULL. */
698 if (peer
&& (tap
->has_queues
|| tap
->has_fds
|| tap
->has_vhostfds
)) {
699 error_report("Multiqueue tap cannnot be used with QEMU vlans");
704 if (tap
->has_ifname
|| tap
->has_script
|| tap
->has_downscript
||
705 tap
->has_vnet_hdr
|| tap
->has_helper
|| tap
->has_queues
||
707 error_report("ifname=, script=, downscript=, vnet_hdr=, "
708 "helper=, queues=, and fds= are invalid with fd=");
712 fd
= monitor_handle_fd_param(cur_mon
, tap
->fd
);
717 fcntl(fd
, F_SETFL
, O_NONBLOCK
);
719 vnet_hdr
= tap_probe_vnet_hdr(fd
);
721 if (net_init_tap_one(tap
, peer
, "tap", name
, NULL
,
723 vhostfdname
, vnet_hdr
, fd
)) {
726 } else if (tap
->has_fds
) {
727 char *fds
[MAX_TAP_QUEUES
];
728 char *vhost_fds
[MAX_TAP_QUEUES
];
731 if (tap
->has_ifname
|| tap
->has_script
|| tap
->has_downscript
||
732 tap
->has_vnet_hdr
|| tap
->has_helper
|| tap
->has_queues
||
734 error_report("ifname=, script=, downscript=, vnet_hdr=, "
735 "helper=, queues=, and fd= are invalid with fds=");
739 nfds
= get_fds(tap
->fds
, fds
, MAX_TAP_QUEUES
);
740 if (tap
->has_vhostfds
) {
741 nvhosts
= get_fds(tap
->vhostfds
, vhost_fds
, MAX_TAP_QUEUES
);
742 if (nfds
!= nvhosts
) {
743 error_report("The number of fds passed does not match the "
744 "number of vhostfds passed");
749 for (i
= 0; i
< nfds
; i
++) {
750 fd
= monitor_handle_fd_param(cur_mon
, fds
[i
]);
755 fcntl(fd
, F_SETFL
, O_NONBLOCK
);
758 vnet_hdr
= tap_probe_vnet_hdr(fd
);
759 } else if (vnet_hdr
!= tap_probe_vnet_hdr(fd
)) {
760 error_report("vnet_hdr not consistent across given tap fds");
764 if (net_init_tap_one(tap
, peer
, "tap", name
, ifname
,
766 tap
->has_vhostfds
? vhost_fds
[i
] : NULL
,
771 } else if (tap
->has_helper
) {
772 if (tap
->has_ifname
|| tap
->has_script
|| tap
->has_downscript
||
773 tap
->has_vnet_hdr
|| tap
->has_queues
|| tap
->has_fds
) {
774 error_report("ifname=, script=, downscript=, and vnet_hdr= "
775 "queues=, and fds= are invalid with helper=");
779 fd
= net_bridge_run_helper(tap
->helper
, DEFAULT_BRIDGE_INTERFACE
);
784 fcntl(fd
, F_SETFL
, O_NONBLOCK
);
785 vnet_hdr
= tap_probe_vnet_hdr(fd
);
787 if (net_init_tap_one(tap
, peer
, "bridge", name
, ifname
,
788 script
, downscript
, vhostfdname
,
793 script
= tap
->has_script
? tap
->script
: DEFAULT_NETWORK_SCRIPT
;
794 downscript
= tap
->has_downscript
? tap
->downscript
:
795 DEFAULT_NETWORK_DOWN_SCRIPT
;
797 if (tap
->has_ifname
) {
798 pstrcpy(ifname
, sizeof ifname
, tap
->ifname
);
803 for (i
= 0; i
< queues
; i
++) {
804 fd
= net_tap_init(tap
, &vnet_hdr
, i
>= 1 ? "no" : script
,
805 ifname
, sizeof ifname
, queues
> 1);
810 if (queues
> 1 && i
== 0 && !tap
->has_ifname
) {
811 if (tap_fd_get_ifname(fd
, ifname
)) {
812 error_report("Fail to get ifname");
817 if (net_init_tap_one(tap
, peer
, "tap", name
, ifname
,
818 i
>= 1 ? "no" : script
,
819 i
>= 1 ? "no" : downscript
,
820 vhostfdname
, vnet_hdr
, fd
)) {
829 VHostNetState
*tap_get_vhost_net(NetClientState
*nc
)
831 TAPState
*s
= DO_UPCAST(TAPState
, nc
, nc
);
832 assert(nc
->info
->type
== NET_CLIENT_OPTIONS_KIND_TAP
);
836 int tap_enable(NetClientState
*nc
)
838 TAPState
*s
= DO_UPCAST(TAPState
, nc
, nc
);
844 ret
= tap_fd_enable(s
->fd
);
847 tap_update_fd_handler(s
);
853 int tap_disable(NetClientState
*nc
)
855 TAPState
*s
= DO_UPCAST(TAPState
, nc
, nc
);
858 if (s
->enabled
== 0) {
861 ret
= tap_fd_disable(s
->fd
);
863 qemu_purge_queued_packets(nc
);
865 tap_update_fd_handler(s
);