xref: /qemu/net/tap.c (revision 2e753bcc7db1a7d2af0d47c0ae7e82a04ce69d63)
15281d757SMark McLoughlin /*
25281d757SMark McLoughlin  * QEMU System Emulator
35281d757SMark McLoughlin  *
45281d757SMark McLoughlin  * Copyright (c) 2003-2008 Fabrice Bellard
55281d757SMark McLoughlin  * Copyright (c) 2009 Red Hat, Inc.
65281d757SMark McLoughlin  *
75281d757SMark McLoughlin  * Permission is hereby granted, free of charge, to any person obtaining a copy
85281d757SMark McLoughlin  * of this software and associated documentation files (the "Software"), to deal
95281d757SMark McLoughlin  * in the Software without restriction, including without limitation the rights
105281d757SMark McLoughlin  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
115281d757SMark McLoughlin  * copies of the Software, and to permit persons to whom the Software is
125281d757SMark McLoughlin  * furnished to do so, subject to the following conditions:
135281d757SMark McLoughlin  *
145281d757SMark McLoughlin  * The above copyright notice and this permission notice shall be included in
155281d757SMark McLoughlin  * all copies or substantial portions of the Software.
165281d757SMark McLoughlin  *
175281d757SMark McLoughlin  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
185281d757SMark McLoughlin  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
195281d757SMark McLoughlin  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
205281d757SMark McLoughlin  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
215281d757SMark McLoughlin  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
225281d757SMark McLoughlin  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
235281d757SMark McLoughlin  * THE SOFTWARE.
245281d757SMark McLoughlin  */
255281d757SMark McLoughlin 
261422e32dSPaolo Bonzini #include "tap_int.h"
275281d757SMark McLoughlin 
285281d757SMark McLoughlin #include "config-host.h"
295281d757SMark McLoughlin 
305281d757SMark McLoughlin #include <sys/ioctl.h>
315281d757SMark McLoughlin #include <sys/stat.h>
325281d757SMark McLoughlin #include <sys/wait.h>
3371f4effcSAlexander Graf #include <sys/socket.h>
345281d757SMark McLoughlin #include <net/if.h>
355281d757SMark McLoughlin 
361422e32dSPaolo Bonzini #include "net/net.h"
37a245fc18SPaolo Bonzini #include "clients.h"
3883c9089eSPaolo Bonzini #include "monitor/monitor.h"
399c17d615SPaolo Bonzini #include "sysemu/sysemu.h"
405281d757SMark McLoughlin #include "qemu-common.h"
411de7afc9SPaolo Bonzini #include "qemu/error-report.h"
425281d757SMark McLoughlin 
431422e32dSPaolo Bonzini #include "net/tap.h"
445281d757SMark McLoughlin 
450d09e41aSPaolo Bonzini #include "net/vhost_net.h"
4682b0d80eSMichael S. Tsirkin 
475281d757SMark McLoughlin typedef struct TAPState {
484e68f7a0SStefan Hajnoczi     NetClientState nc;
495281d757SMark McLoughlin     int fd;
505281d757SMark McLoughlin     char down_script[1024];
515281d757SMark McLoughlin     char down_script_arg[128];
52d32fcad3SScott Feldman     uint8_t buf[NET_BUFSIZE];
53ec45f083SJason Wang     bool read_poll;
54ec45f083SJason Wang     bool write_poll;
55ec45f083SJason Wang     bool using_vnet_hdr;
56ec45f083SJason Wang     bool has_ufo;
5716dbaf90SJason Wang     bool enabled;
5882b0d80eSMichael S. Tsirkin     VHostNetState *vhost_net;
59ef4252b1SMichael S. Tsirkin     unsigned host_vnet_hdr_len;
605281d757SMark McLoughlin } TAPState;
615281d757SMark McLoughlin 
625281d757SMark McLoughlin static int launch_script(const char *setup_script, const char *ifname, int fd);
635281d757SMark McLoughlin 
645281d757SMark McLoughlin static int tap_can_send(void *opaque);
655281d757SMark McLoughlin static void tap_send(void *opaque);
665281d757SMark McLoughlin static void tap_writable(void *opaque);
675281d757SMark McLoughlin 
685281d757SMark McLoughlin static void tap_update_fd_handler(TAPState *s)
695281d757SMark McLoughlin {
705281d757SMark McLoughlin     qemu_set_fd_handler2(s->fd,
7116dbaf90SJason Wang                          s->read_poll && s->enabled ? tap_can_send : NULL,
7216dbaf90SJason Wang                          s->read_poll && s->enabled ? tap_send     : NULL,
7316dbaf90SJason Wang                          s->write_poll && s->enabled ? tap_writable : NULL,
745281d757SMark McLoughlin                          s);
755281d757SMark McLoughlin }
765281d757SMark McLoughlin 
77ec45f083SJason Wang static void tap_read_poll(TAPState *s, bool enable)
785281d757SMark McLoughlin {
79ec45f083SJason Wang     s->read_poll = enable;
805281d757SMark McLoughlin     tap_update_fd_handler(s);
815281d757SMark McLoughlin }
825281d757SMark McLoughlin 
83ec45f083SJason Wang static void tap_write_poll(TAPState *s, bool enable)
845281d757SMark McLoughlin {
85ec45f083SJason Wang     s->write_poll = enable;
865281d757SMark McLoughlin     tap_update_fd_handler(s);
875281d757SMark McLoughlin }
885281d757SMark McLoughlin 
895281d757SMark McLoughlin static void tap_writable(void *opaque)
905281d757SMark McLoughlin {
915281d757SMark McLoughlin     TAPState *s = opaque;
925281d757SMark McLoughlin 
93ec45f083SJason Wang     tap_write_poll(s, false);
945281d757SMark McLoughlin 
953e35ba93SMark McLoughlin     qemu_flush_queued_packets(&s->nc);
965281d757SMark McLoughlin }
975281d757SMark McLoughlin 
985281d757SMark McLoughlin static ssize_t tap_write_packet(TAPState *s, const struct iovec *iov, int iovcnt)
995281d757SMark McLoughlin {
1005281d757SMark McLoughlin     ssize_t len;
1015281d757SMark McLoughlin 
1025281d757SMark McLoughlin     do {
1035281d757SMark McLoughlin         len = writev(s->fd, iov, iovcnt);
1045281d757SMark McLoughlin     } while (len == -1 && errno == EINTR);
1055281d757SMark McLoughlin 
1065281d757SMark McLoughlin     if (len == -1 && errno == EAGAIN) {
107ec45f083SJason Wang         tap_write_poll(s, true);
1085281d757SMark McLoughlin         return 0;
1095281d757SMark McLoughlin     }
1105281d757SMark McLoughlin 
1115281d757SMark McLoughlin     return len;
1125281d757SMark McLoughlin }
1135281d757SMark McLoughlin 
1144e68f7a0SStefan Hajnoczi static ssize_t tap_receive_iov(NetClientState *nc, const struct iovec *iov,
1155281d757SMark McLoughlin                                int iovcnt)
1165281d757SMark McLoughlin {
1173e35ba93SMark McLoughlin     TAPState *s = DO_UPCAST(TAPState, nc, nc);
1185281d757SMark McLoughlin     const struct iovec *iovp = iov;
1195281d757SMark McLoughlin     struct iovec iov_copy[iovcnt + 1];
120ef4252b1SMichael S. Tsirkin     struct virtio_net_hdr_mrg_rxbuf hdr = { };
1215281d757SMark McLoughlin 
122ef4252b1SMichael S. Tsirkin     if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
1235281d757SMark McLoughlin         iov_copy[0].iov_base = &hdr;
124ef4252b1SMichael S. Tsirkin         iov_copy[0].iov_len =  s->host_vnet_hdr_len;
1255281d757SMark McLoughlin         memcpy(&iov_copy[1], iov, iovcnt * sizeof(*iov));
1265281d757SMark McLoughlin         iovp = iov_copy;
1275281d757SMark McLoughlin         iovcnt++;
1285281d757SMark McLoughlin     }
1295281d757SMark McLoughlin 
1305281d757SMark McLoughlin     return tap_write_packet(s, iovp, iovcnt);
1315281d757SMark McLoughlin }
1325281d757SMark McLoughlin 
1334e68f7a0SStefan Hajnoczi static ssize_t tap_receive_raw(NetClientState *nc, const uint8_t *buf, size_t size)
1345281d757SMark McLoughlin {
1353e35ba93SMark McLoughlin     TAPState *s = DO_UPCAST(TAPState, nc, nc);
1365281d757SMark McLoughlin     struct iovec iov[2];
1375281d757SMark McLoughlin     int iovcnt = 0;
138ef4252b1SMichael S. Tsirkin     struct virtio_net_hdr_mrg_rxbuf hdr = { };
1395281d757SMark McLoughlin 
140ef4252b1SMichael S. Tsirkin     if (s->host_vnet_hdr_len) {
1415281d757SMark McLoughlin         iov[iovcnt].iov_base = &hdr;
142ef4252b1SMichael S. Tsirkin         iov[iovcnt].iov_len  = s->host_vnet_hdr_len;
1435281d757SMark McLoughlin         iovcnt++;
1445281d757SMark McLoughlin     }
1455281d757SMark McLoughlin 
1465281d757SMark McLoughlin     iov[iovcnt].iov_base = (char *)buf;
1475281d757SMark McLoughlin     iov[iovcnt].iov_len  = size;
1485281d757SMark McLoughlin     iovcnt++;
1495281d757SMark McLoughlin 
1505281d757SMark McLoughlin     return tap_write_packet(s, iov, iovcnt);
1515281d757SMark McLoughlin }
1525281d757SMark McLoughlin 
1534e68f7a0SStefan Hajnoczi static ssize_t tap_receive(NetClientState *nc, const uint8_t *buf, size_t size)
1545281d757SMark McLoughlin {
1553e35ba93SMark McLoughlin     TAPState *s = DO_UPCAST(TAPState, nc, nc);
1565281d757SMark McLoughlin     struct iovec iov[1];
1575281d757SMark McLoughlin 
158ef4252b1SMichael S. Tsirkin     if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
1593e35ba93SMark McLoughlin         return tap_receive_raw(nc, buf, size);
1605281d757SMark McLoughlin     }
1615281d757SMark McLoughlin 
1625281d757SMark McLoughlin     iov[0].iov_base = (char *)buf;
1635281d757SMark McLoughlin     iov[0].iov_len  = size;
1645281d757SMark McLoughlin 
1655281d757SMark McLoughlin     return tap_write_packet(s, iov, 1);
1665281d757SMark McLoughlin }
1675281d757SMark McLoughlin 
1685281d757SMark McLoughlin static int tap_can_send(void *opaque)
1695281d757SMark McLoughlin {
1705281d757SMark McLoughlin     TAPState *s = opaque;
1715281d757SMark McLoughlin 
1723e35ba93SMark McLoughlin     return qemu_can_send_packet(&s->nc);
1735281d757SMark McLoughlin }
1745281d757SMark McLoughlin 
175966ea5ecSMark McLoughlin #ifndef __sun__
176966ea5ecSMark McLoughlin ssize_t tap_read_packet(int tapfd, uint8_t *buf, int maxlen)
1775281d757SMark McLoughlin {
1785281d757SMark McLoughlin     return read(tapfd, buf, maxlen);
1795281d757SMark McLoughlin }
1805281d757SMark McLoughlin #endif
1815281d757SMark McLoughlin 
1824e68f7a0SStefan Hajnoczi static void tap_send_completed(NetClientState *nc, ssize_t len)
1835281d757SMark McLoughlin {
1843e35ba93SMark McLoughlin     TAPState *s = DO_UPCAST(TAPState, nc, nc);
185ec45f083SJason Wang     tap_read_poll(s, true);
1865281d757SMark McLoughlin }
1875281d757SMark McLoughlin 
1885281d757SMark McLoughlin static void tap_send(void *opaque)
1895281d757SMark McLoughlin {
1905281d757SMark McLoughlin     TAPState *s = opaque;
191be1636b3SMark McLoughlin     int size;
1925281d757SMark McLoughlin 
1935819c918SMark McLoughlin     do {
1945819c918SMark McLoughlin         uint8_t *buf = s->buf;
1955819c918SMark McLoughlin 
1965281d757SMark McLoughlin         size = tap_read_packet(s->fd, s->buf, sizeof(s->buf));
1975281d757SMark McLoughlin         if (size <= 0) {
1985819c918SMark McLoughlin             break;
1995281d757SMark McLoughlin         }
2005281d757SMark McLoughlin 
201ef4252b1SMichael S. Tsirkin         if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
202ef4252b1SMichael S. Tsirkin             buf  += s->host_vnet_hdr_len;
203ef4252b1SMichael S. Tsirkin             size -= s->host_vnet_hdr_len;
2045281d757SMark McLoughlin         }
2055281d757SMark McLoughlin 
2063e35ba93SMark McLoughlin         size = qemu_send_packet_async(&s->nc, buf, size, tap_send_completed);
2075281d757SMark McLoughlin         if (size == 0) {
208ec45f083SJason Wang             tap_read_poll(s, false);
2095281d757SMark McLoughlin         }
2103e35ba93SMark McLoughlin     } while (size > 0 && qemu_can_send_packet(&s->nc));
2115281d757SMark McLoughlin }
2125281d757SMark McLoughlin 
213ec45f083SJason Wang bool tap_has_ufo(NetClientState *nc)
2145281d757SMark McLoughlin {
2153e35ba93SMark McLoughlin     TAPState *s = DO_UPCAST(TAPState, nc, nc);
2165281d757SMark McLoughlin 
2172be64a68SLaszlo Ersek     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
2185281d757SMark McLoughlin 
2195281d757SMark McLoughlin     return s->has_ufo;
2205281d757SMark McLoughlin }
2215281d757SMark McLoughlin 
222e96dfd11SVincenzo Maffione bool tap_has_vnet_hdr(NetClientState *nc)
2235281d757SMark McLoughlin {
2243e35ba93SMark McLoughlin     TAPState *s = DO_UPCAST(TAPState, nc, nc);
2255281d757SMark McLoughlin 
2262be64a68SLaszlo Ersek     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
2275281d757SMark McLoughlin 
228ef4252b1SMichael S. Tsirkin     return !!s->host_vnet_hdr_len;
2295281d757SMark McLoughlin }
2305281d757SMark McLoughlin 
231e96dfd11SVincenzo Maffione bool tap_has_vnet_hdr_len(NetClientState *nc, int len)
232445d892fSMichael S. Tsirkin {
233445d892fSMichael S. Tsirkin     TAPState *s = DO_UPCAST(TAPState, nc, nc);
234445d892fSMichael S. Tsirkin 
2352be64a68SLaszlo Ersek     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
236445d892fSMichael S. Tsirkin 
237e96dfd11SVincenzo Maffione     return !!tap_probe_vnet_hdr_len(s->fd, len);
238445d892fSMichael S. Tsirkin }
239445d892fSMichael S. Tsirkin 
2404e68f7a0SStefan Hajnoczi void tap_set_vnet_hdr_len(NetClientState *nc, int len)
241445d892fSMichael S. Tsirkin {
242445d892fSMichael S. Tsirkin     TAPState *s = DO_UPCAST(TAPState, nc, nc);
243445d892fSMichael S. Tsirkin 
2442be64a68SLaszlo Ersek     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
245445d892fSMichael S. Tsirkin     assert(len == sizeof(struct virtio_net_hdr_mrg_rxbuf) ||
246445d892fSMichael S. Tsirkin            len == sizeof(struct virtio_net_hdr));
247445d892fSMichael S. Tsirkin 
248445d892fSMichael S. Tsirkin     tap_fd_set_vnet_hdr_len(s->fd, len);
249445d892fSMichael S. Tsirkin     s->host_vnet_hdr_len = len;
250445d892fSMichael S. Tsirkin }
251445d892fSMichael S. Tsirkin 
252ec45f083SJason Wang void tap_using_vnet_hdr(NetClientState *nc, bool using_vnet_hdr)
2535281d757SMark McLoughlin {
2543e35ba93SMark McLoughlin     TAPState *s = DO_UPCAST(TAPState, nc, nc);
2555281d757SMark McLoughlin 
2562be64a68SLaszlo Ersek     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
257ef4252b1SMichael S. Tsirkin     assert(!!s->host_vnet_hdr_len == using_vnet_hdr);
2585281d757SMark McLoughlin 
2595281d757SMark McLoughlin     s->using_vnet_hdr = using_vnet_hdr;
2605281d757SMark McLoughlin }
2615281d757SMark McLoughlin 
2624e68f7a0SStefan Hajnoczi void tap_set_offload(NetClientState *nc, int csum, int tso4,
2635281d757SMark McLoughlin                      int tso6, int ecn, int ufo)
2645281d757SMark McLoughlin {
2653e35ba93SMark McLoughlin     TAPState *s = DO_UPCAST(TAPState, nc, nc);
26627a6375dSMichael S. Tsirkin     if (s->fd < 0) {
26727a6375dSMichael S. Tsirkin         return;
26827a6375dSMichael S. Tsirkin     }
2695281d757SMark McLoughlin 
27027a6375dSMichael S. Tsirkin     tap_fd_set_offload(s->fd, csum, tso4, tso6, ecn, ufo);
2715281d757SMark McLoughlin }
2725281d757SMark McLoughlin 
2734e68f7a0SStefan Hajnoczi static void tap_cleanup(NetClientState *nc)
2745281d757SMark McLoughlin {
2753e35ba93SMark McLoughlin     TAPState *s = DO_UPCAST(TAPState, nc, nc);
2765281d757SMark McLoughlin 
27782b0d80eSMichael S. Tsirkin     if (s->vhost_net) {
27882b0d80eSMichael S. Tsirkin         vhost_net_cleanup(s->vhost_net);
27943849424SMichael S. Tsirkin         s->vhost_net = NULL;
28082b0d80eSMichael S. Tsirkin     }
28182b0d80eSMichael S. Tsirkin 
2823e35ba93SMark McLoughlin     qemu_purge_queued_packets(nc);
2835281d757SMark McLoughlin 
2845281d757SMark McLoughlin     if (s->down_script[0])
2855281d757SMark McLoughlin         launch_script(s->down_script, s->down_script_arg, s->fd);
2865281d757SMark McLoughlin 
287ec45f083SJason Wang     tap_read_poll(s, false);
288ec45f083SJason Wang     tap_write_poll(s, false);
2895281d757SMark McLoughlin     close(s->fd);
29027a6375dSMichael S. Tsirkin     s->fd = -1;
2915281d757SMark McLoughlin }
2925281d757SMark McLoughlin 
2934e68f7a0SStefan Hajnoczi static void tap_poll(NetClientState *nc, bool enable)
294ceb69615SMichael S. Tsirkin {
295ceb69615SMichael S. Tsirkin     TAPState *s = DO_UPCAST(TAPState, nc, nc);
296ceb69615SMichael S. Tsirkin     tap_read_poll(s, enable);
297ceb69615SMichael S. Tsirkin     tap_write_poll(s, enable);
298ceb69615SMichael S. Tsirkin }
299ceb69615SMichael S. Tsirkin 
3004e68f7a0SStefan Hajnoczi int tap_get_fd(NetClientState *nc)
30195d528a2SMichael S. Tsirkin {
30295d528a2SMichael S. Tsirkin     TAPState *s = DO_UPCAST(TAPState, nc, nc);
3032be64a68SLaszlo Ersek     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
30495d528a2SMichael S. Tsirkin     return s->fd;
30595d528a2SMichael S. Tsirkin }
30695d528a2SMichael S. Tsirkin 
3075281d757SMark McLoughlin /* fd support */
3085281d757SMark McLoughlin 
3093e35ba93SMark McLoughlin static NetClientInfo net_tap_info = {
3102be64a68SLaszlo Ersek     .type = NET_CLIENT_OPTIONS_KIND_TAP,
3113e35ba93SMark McLoughlin     .size = sizeof(TAPState),
3123e35ba93SMark McLoughlin     .receive = tap_receive,
3133e35ba93SMark McLoughlin     .receive_raw = tap_receive_raw,
3143e35ba93SMark McLoughlin     .receive_iov = tap_receive_iov,
315ceb69615SMichael S. Tsirkin     .poll = tap_poll,
3163e35ba93SMark McLoughlin     .cleanup = tap_cleanup,
317*2e753bccSVincenzo Maffione     .has_ufo = tap_has_ufo,
318*2e753bccSVincenzo Maffione     .has_vnet_hdr = tap_has_vnet_hdr,
319*2e753bccSVincenzo Maffione     .has_vnet_hdr_len = tap_has_vnet_hdr_len,
320*2e753bccSVincenzo Maffione     .using_vnet_hdr = tap_using_vnet_hdr,
321*2e753bccSVincenzo Maffione     .set_offload = tap_set_offload,
322*2e753bccSVincenzo Maffione     .set_vnet_hdr_len = tap_set_vnet_hdr_len,
3233e35ba93SMark McLoughlin };
3243e35ba93SMark McLoughlin 
3254e68f7a0SStefan Hajnoczi static TAPState *net_tap_fd_init(NetClientState *peer,
3265281d757SMark McLoughlin                                  const char *model,
3275281d757SMark McLoughlin                                  const char *name,
3285281d757SMark McLoughlin                                  int fd,
3295281d757SMark McLoughlin                                  int vnet_hdr)
3305281d757SMark McLoughlin {
3314e68f7a0SStefan Hajnoczi     NetClientState *nc;
3325281d757SMark McLoughlin     TAPState *s;
3335281d757SMark McLoughlin 
334ab5f3f84SStefan Hajnoczi     nc = qemu_new_net_client(&net_tap_info, peer, model, name);
3353e35ba93SMark McLoughlin 
3363e35ba93SMark McLoughlin     s = DO_UPCAST(TAPState, nc, nc);
3373e35ba93SMark McLoughlin 
3385281d757SMark McLoughlin     s->fd = fd;
339ef4252b1SMichael S. Tsirkin     s->host_vnet_hdr_len = vnet_hdr ? sizeof(struct virtio_net_hdr) : 0;
340ec45f083SJason Wang     s->using_vnet_hdr = false;
3419c282718SMark McLoughlin     s->has_ufo = tap_probe_has_ufo(s->fd);
34216dbaf90SJason Wang     s->enabled = true;
3433e35ba93SMark McLoughlin     tap_set_offload(&s->nc, 0, 0, 0, 0, 0);
34458ddcd50SMichael S. Tsirkin     /*
34558ddcd50SMichael S. Tsirkin      * Make sure host header length is set correctly in tap:
34658ddcd50SMichael S. Tsirkin      * it might have been modified by another instance of qemu.
34758ddcd50SMichael S. Tsirkin      */
34858ddcd50SMichael S. Tsirkin     if (tap_probe_vnet_hdr_len(s->fd, s->host_vnet_hdr_len)) {
34958ddcd50SMichael S. Tsirkin         tap_fd_set_vnet_hdr_len(s->fd, s->host_vnet_hdr_len);
35058ddcd50SMichael S. Tsirkin     }
351ec45f083SJason Wang     tap_read_poll(s, true);
35282b0d80eSMichael S. Tsirkin     s->vhost_net = NULL;
3535281d757SMark McLoughlin     return s;
3545281d757SMark McLoughlin }
3555281d757SMark McLoughlin 
3565281d757SMark McLoughlin static int launch_script(const char *setup_script, const char *ifname, int fd)
3575281d757SMark McLoughlin {
3585281d757SMark McLoughlin     int pid, status;
3595281d757SMark McLoughlin     char *args[3];
3605281d757SMark McLoughlin     char **parg;
3615281d757SMark McLoughlin 
3625281d757SMark McLoughlin     /* try to launch network script */
3635281d757SMark McLoughlin     pid = fork();
3645281d757SMark McLoughlin     if (pid == 0) {
3655281d757SMark McLoughlin         int open_max = sysconf(_SC_OPEN_MAX), i;
3665281d757SMark McLoughlin 
3675281d757SMark McLoughlin         for (i = 0; i < open_max; i++) {
3685281d757SMark McLoughlin             if (i != STDIN_FILENO &&
3695281d757SMark McLoughlin                 i != STDOUT_FILENO &&
3705281d757SMark McLoughlin                 i != STDERR_FILENO &&
3715281d757SMark McLoughlin                 i != fd) {
3725281d757SMark McLoughlin                 close(i);
3735281d757SMark McLoughlin             }
3745281d757SMark McLoughlin         }
3755281d757SMark McLoughlin         parg = args;
3765281d757SMark McLoughlin         *parg++ = (char *)setup_script;
3775281d757SMark McLoughlin         *parg++ = (char *)ifname;
3789678d950SBlue Swirl         *parg = NULL;
3795281d757SMark McLoughlin         execv(setup_script, args);
3805281d757SMark McLoughlin         _exit(1);
3815281d757SMark McLoughlin     } else if (pid > 0) {
3825281d757SMark McLoughlin         while (waitpid(pid, &status, 0) != pid) {
3835281d757SMark McLoughlin             /* loop */
3845281d757SMark McLoughlin         }
3855281d757SMark McLoughlin 
3865281d757SMark McLoughlin         if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
3875281d757SMark McLoughlin             return 0;
3885281d757SMark McLoughlin         }
3895281d757SMark McLoughlin     }
3905281d757SMark McLoughlin     fprintf(stderr, "%s: could not launch network script\n", setup_script);
3915281d757SMark McLoughlin     return -1;
3925281d757SMark McLoughlin }
3935281d757SMark McLoughlin 
394a7c36ee4SCorey Bryant static int recv_fd(int c)
395a7c36ee4SCorey Bryant {
396a7c36ee4SCorey Bryant     int fd;
397a7c36ee4SCorey Bryant     uint8_t msgbuf[CMSG_SPACE(sizeof(fd))];
398a7c36ee4SCorey Bryant     struct msghdr msg = {
399a7c36ee4SCorey Bryant         .msg_control = msgbuf,
400a7c36ee4SCorey Bryant         .msg_controllen = sizeof(msgbuf),
401a7c36ee4SCorey Bryant     };
402a7c36ee4SCorey Bryant     struct cmsghdr *cmsg;
403a7c36ee4SCorey Bryant     struct iovec iov;
404a7c36ee4SCorey Bryant     uint8_t req[1];
405a7c36ee4SCorey Bryant     ssize_t len;
406a7c36ee4SCorey Bryant 
407a7c36ee4SCorey Bryant     cmsg = CMSG_FIRSTHDR(&msg);
408a7c36ee4SCorey Bryant     cmsg->cmsg_level = SOL_SOCKET;
409a7c36ee4SCorey Bryant     cmsg->cmsg_type = SCM_RIGHTS;
410a7c36ee4SCorey Bryant     cmsg->cmsg_len = CMSG_LEN(sizeof(fd));
411a7c36ee4SCorey Bryant     msg.msg_controllen = cmsg->cmsg_len;
412a7c36ee4SCorey Bryant 
413a7c36ee4SCorey Bryant     iov.iov_base = req;
414a7c36ee4SCorey Bryant     iov.iov_len = sizeof(req);
415a7c36ee4SCorey Bryant 
416a7c36ee4SCorey Bryant     msg.msg_iov = &iov;
417a7c36ee4SCorey Bryant     msg.msg_iovlen = 1;
418a7c36ee4SCorey Bryant 
419a7c36ee4SCorey Bryant     len = recvmsg(c, &msg, 0);
420a7c36ee4SCorey Bryant     if (len > 0) {
421a7c36ee4SCorey Bryant         memcpy(&fd, CMSG_DATA(cmsg), sizeof(fd));
422a7c36ee4SCorey Bryant         return fd;
423a7c36ee4SCorey Bryant     }
424a7c36ee4SCorey Bryant 
425a7c36ee4SCorey Bryant     return len;
426a7c36ee4SCorey Bryant }
427a7c36ee4SCorey Bryant 
428a7c36ee4SCorey Bryant static int net_bridge_run_helper(const char *helper, const char *bridge)
429a7c36ee4SCorey Bryant {
430a7c36ee4SCorey Bryant     sigset_t oldmask, mask;
431a7c36ee4SCorey Bryant     int pid, status;
432a7c36ee4SCorey Bryant     char *args[5];
433a7c36ee4SCorey Bryant     char **parg;
434a7c36ee4SCorey Bryant     int sv[2];
435a7c36ee4SCorey Bryant 
436a7c36ee4SCorey Bryant     sigemptyset(&mask);
437a7c36ee4SCorey Bryant     sigaddset(&mask, SIGCHLD);
438a7c36ee4SCorey Bryant     sigprocmask(SIG_BLOCK, &mask, &oldmask);
439a7c36ee4SCorey Bryant 
440a7c36ee4SCorey Bryant     if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) {
441a7c36ee4SCorey Bryant         return -1;
442a7c36ee4SCorey Bryant     }
443a7c36ee4SCorey Bryant 
444a7c36ee4SCorey Bryant     /* try to launch bridge helper */
445a7c36ee4SCorey Bryant     pid = fork();
446a7c36ee4SCorey Bryant     if (pid == 0) {
447a7c36ee4SCorey Bryant         int open_max = sysconf(_SC_OPEN_MAX), i;
448a7c36ee4SCorey Bryant         char fd_buf[6+10];
449a7c36ee4SCorey Bryant         char br_buf[6+IFNAMSIZ] = {0};
450a7c36ee4SCorey Bryant         char helper_cmd[PATH_MAX + sizeof(fd_buf) + sizeof(br_buf) + 15];
451a7c36ee4SCorey Bryant 
452a7c36ee4SCorey Bryant         for (i = 0; i < open_max; i++) {
453a7c36ee4SCorey Bryant             if (i != STDIN_FILENO &&
454a7c36ee4SCorey Bryant                 i != STDOUT_FILENO &&
455a7c36ee4SCorey Bryant                 i != STDERR_FILENO &&
456a7c36ee4SCorey Bryant                 i != sv[1]) {
457a7c36ee4SCorey Bryant                 close(i);
458a7c36ee4SCorey Bryant             }
459a7c36ee4SCorey Bryant         }
460a7c36ee4SCorey Bryant 
461a7c36ee4SCorey Bryant         snprintf(fd_buf, sizeof(fd_buf), "%s%d", "--fd=", sv[1]);
462a7c36ee4SCorey Bryant 
463a7c36ee4SCorey Bryant         if (strrchr(helper, ' ') || strrchr(helper, '\t')) {
464a7c36ee4SCorey Bryant             /* assume helper is a command */
465a7c36ee4SCorey Bryant 
466a7c36ee4SCorey Bryant             if (strstr(helper, "--br=") == NULL) {
467a7c36ee4SCorey Bryant                 snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge);
468a7c36ee4SCorey Bryant             }
469a7c36ee4SCorey Bryant 
470a7c36ee4SCorey Bryant             snprintf(helper_cmd, sizeof(helper_cmd), "%s %s %s %s",
471a7c36ee4SCorey Bryant                      helper, "--use-vnet", fd_buf, br_buf);
472a7c36ee4SCorey Bryant 
473a7c36ee4SCorey Bryant             parg = args;
474a7c36ee4SCorey Bryant             *parg++ = (char *)"sh";
475a7c36ee4SCorey Bryant             *parg++ = (char *)"-c";
476a7c36ee4SCorey Bryant             *parg++ = helper_cmd;
477a7c36ee4SCorey Bryant             *parg++ = NULL;
478a7c36ee4SCorey Bryant 
479a7c36ee4SCorey Bryant             execv("/bin/sh", args);
480a7c36ee4SCorey Bryant         } else {
481a7c36ee4SCorey Bryant             /* assume helper is just the executable path name */
482a7c36ee4SCorey Bryant 
483a7c36ee4SCorey Bryant             snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge);
484a7c36ee4SCorey Bryant 
485a7c36ee4SCorey Bryant             parg = args;
486a7c36ee4SCorey Bryant             *parg++ = (char *)helper;
487a7c36ee4SCorey Bryant             *parg++ = (char *)"--use-vnet";
488a7c36ee4SCorey Bryant             *parg++ = fd_buf;
489a7c36ee4SCorey Bryant             *parg++ = br_buf;
490a7c36ee4SCorey Bryant             *parg++ = NULL;
491a7c36ee4SCorey Bryant 
492a7c36ee4SCorey Bryant             execv(helper, args);
493a7c36ee4SCorey Bryant         }
494a7c36ee4SCorey Bryant         _exit(1);
495a7c36ee4SCorey Bryant 
496a7c36ee4SCorey Bryant     } else if (pid > 0) {
497a7c36ee4SCorey Bryant         int fd;
498a7c36ee4SCorey Bryant 
499a7c36ee4SCorey Bryant         close(sv[1]);
500a7c36ee4SCorey Bryant 
501a7c36ee4SCorey Bryant         do {
502a7c36ee4SCorey Bryant             fd = recv_fd(sv[0]);
503a7c36ee4SCorey Bryant         } while (fd == -1 && errno == EINTR);
504a7c36ee4SCorey Bryant 
505a7c36ee4SCorey Bryant         close(sv[0]);
506a7c36ee4SCorey Bryant 
507a7c36ee4SCorey Bryant         while (waitpid(pid, &status, 0) != pid) {
508a7c36ee4SCorey Bryant             /* loop */
509a7c36ee4SCorey Bryant         }
510a7c36ee4SCorey Bryant         sigprocmask(SIG_SETMASK, &oldmask, NULL);
511a7c36ee4SCorey Bryant         if (fd < 0) {
512a7c36ee4SCorey Bryant             fprintf(stderr, "failed to recv file descriptor\n");
513a7c36ee4SCorey Bryant             return -1;
514a7c36ee4SCorey Bryant         }
515a7c36ee4SCorey Bryant 
516a7c36ee4SCorey Bryant         if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
517a7c36ee4SCorey Bryant             return fd;
518a7c36ee4SCorey Bryant         }
519a7c36ee4SCorey Bryant     }
520a7c36ee4SCorey Bryant     fprintf(stderr, "failed to launch bridge helper\n");
521a7c36ee4SCorey Bryant     return -1;
522a7c36ee4SCorey Bryant }
523a7c36ee4SCorey Bryant 
5241a0c0958SLaszlo Ersek int net_init_bridge(const NetClientOptions *opts, const char *name,
5254e68f7a0SStefan Hajnoczi                     NetClientState *peer)
526a7c36ee4SCorey Bryant {
527f79b51b0SLaszlo Ersek     const NetdevBridgeOptions *bridge;
528f79b51b0SLaszlo Ersek     const char *helper, *br;
529f79b51b0SLaszlo Ersek 
530a7c36ee4SCorey Bryant     TAPState *s;
531a7c36ee4SCorey Bryant     int fd, vnet_hdr;
532a7c36ee4SCorey Bryant 
533f79b51b0SLaszlo Ersek     assert(opts->kind == NET_CLIENT_OPTIONS_KIND_BRIDGE);
534f79b51b0SLaszlo Ersek     bridge = opts->bridge;
535a7c36ee4SCorey Bryant 
536f79b51b0SLaszlo Ersek     helper = bridge->has_helper ? bridge->helper : DEFAULT_BRIDGE_HELPER;
537f79b51b0SLaszlo Ersek     br     = bridge->has_br     ? bridge->br     : DEFAULT_BRIDGE_INTERFACE;
538f79b51b0SLaszlo Ersek 
539f79b51b0SLaszlo Ersek     fd = net_bridge_run_helper(helper, br);
540a7c36ee4SCorey Bryant     if (fd == -1) {
541a7c36ee4SCorey Bryant         return -1;
542a7c36ee4SCorey Bryant     }
543a7c36ee4SCorey Bryant 
544a7c36ee4SCorey Bryant     fcntl(fd, F_SETFL, O_NONBLOCK);
545a7c36ee4SCorey Bryant 
546a7c36ee4SCorey Bryant     vnet_hdr = tap_probe_vnet_hdr(fd);
547a7c36ee4SCorey Bryant 
548d33d93b2SStefan Hajnoczi     s = net_tap_fd_init(peer, "bridge", name, fd, vnet_hdr);
549a7c36ee4SCorey Bryant     if (!s) {
550a7c36ee4SCorey Bryant         close(fd);
551a7c36ee4SCorey Bryant         return -1;
552a7c36ee4SCorey Bryant     }
553a7c36ee4SCorey Bryant 
554f79b51b0SLaszlo Ersek     snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s,br=%s", helper,
555f79b51b0SLaszlo Ersek              br);
556a7c36ee4SCorey Bryant 
557a7c36ee4SCorey Bryant     return 0;
558a7c36ee4SCorey Bryant }
559a7c36ee4SCorey Bryant 
56008c573a8SLaszlo Ersek static int net_tap_init(const NetdevTapOptions *tap, int *vnet_hdr,
56108c573a8SLaszlo Ersek                         const char *setup_script, char *ifname,
562264986e2SJason Wang                         size_t ifname_sz, int mq_required)
5635281d757SMark McLoughlin {
5645281d757SMark McLoughlin     int fd, vnet_hdr_required;
5655281d757SMark McLoughlin 
56608c573a8SLaszlo Ersek     if (tap->has_vnet_hdr) {
56708c573a8SLaszlo Ersek         *vnet_hdr = tap->vnet_hdr;
5685281d757SMark McLoughlin         vnet_hdr_required = *vnet_hdr;
5695281d757SMark McLoughlin     } else {
57008c573a8SLaszlo Ersek         *vnet_hdr = 1;
5715281d757SMark McLoughlin         vnet_hdr_required = 0;
5725281d757SMark McLoughlin     }
5735281d757SMark McLoughlin 
574264986e2SJason Wang     TFR(fd = tap_open(ifname, ifname_sz, vnet_hdr, vnet_hdr_required,
575264986e2SJason Wang                       mq_required));
5765281d757SMark McLoughlin     if (fd < 0) {
5775281d757SMark McLoughlin         return -1;
5785281d757SMark McLoughlin     }
5795281d757SMark McLoughlin 
5805281d757SMark McLoughlin     if (setup_script &&
5815281d757SMark McLoughlin         setup_script[0] != '\0' &&
5825281d757SMark McLoughlin         strcmp(setup_script, "no") != 0 &&
5835281d757SMark McLoughlin         launch_script(setup_script, ifname, fd)) {
5845281d757SMark McLoughlin         close(fd);
5855281d757SMark McLoughlin         return -1;
5865281d757SMark McLoughlin     }
5875281d757SMark McLoughlin 
5885281d757SMark McLoughlin     return fd;
5895281d757SMark McLoughlin }
5905281d757SMark McLoughlin 
591264986e2SJason Wang #define MAX_TAP_QUEUES 1024
592264986e2SJason Wang 
5935193e5fbSJason Wang static int net_init_tap_one(const NetdevTapOptions *tap, NetClientState *peer,
5945193e5fbSJason Wang                             const char *model, const char *name,
5955193e5fbSJason Wang                             const char *ifname, const char *script,
5965193e5fbSJason Wang                             const char *downscript, const char *vhostfdname,
5975193e5fbSJason Wang                             int vnet_hdr, int fd)
5985193e5fbSJason Wang {
5995193e5fbSJason Wang     TAPState *s;
6005193e5fbSJason Wang 
6015193e5fbSJason Wang     s = net_tap_fd_init(peer, model, name, fd, vnet_hdr);
6025193e5fbSJason Wang     if (!s) {
6035193e5fbSJason Wang         close(fd);
6045193e5fbSJason Wang         return -1;
6055193e5fbSJason Wang     }
6065193e5fbSJason Wang 
6075193e5fbSJason Wang     if (tap_set_sndbuf(s->fd, tap) < 0) {
6085193e5fbSJason Wang         return -1;
6095193e5fbSJason Wang     }
6105193e5fbSJason Wang 
611264986e2SJason Wang     if (tap->has_fd || tap->has_fds) {
6125193e5fbSJason Wang         snprintf(s->nc.info_str, sizeof(s->nc.info_str), "fd=%d", fd);
6135193e5fbSJason Wang     } else if (tap->has_helper) {
6145193e5fbSJason Wang         snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s",
6155193e5fbSJason Wang                  tap->helper);
6165193e5fbSJason Wang     } else {
6175193e5fbSJason Wang         snprintf(s->nc.info_str, sizeof(s->nc.info_str),
6185193e5fbSJason Wang                  "ifname=%s,script=%s,downscript=%s", ifname, script,
6195193e5fbSJason Wang                  downscript);
6205193e5fbSJason Wang 
6215193e5fbSJason Wang         if (strcmp(downscript, "no") != 0) {
6225193e5fbSJason Wang             snprintf(s->down_script, sizeof(s->down_script), "%s", downscript);
6235193e5fbSJason Wang             snprintf(s->down_script_arg, sizeof(s->down_script_arg),
6245193e5fbSJason Wang                      "%s", ifname);
6255193e5fbSJason Wang         }
6265193e5fbSJason Wang     }
6275193e5fbSJason Wang 
6285193e5fbSJason Wang     if (tap->has_vhost ? tap->vhost :
6295193e5fbSJason Wang         vhostfdname || (tap->has_vhostforce && tap->vhostforce)) {
6305193e5fbSJason Wang         int vhostfd;
6315193e5fbSJason Wang 
6327873df40SJason Wang         if (tap->has_vhostfd || tap->has_vhostfds) {
6335193e5fbSJason Wang             vhostfd = monitor_handle_fd_param(cur_mon, vhostfdname);
6345193e5fbSJason Wang             if (vhostfd == -1) {
6355193e5fbSJason Wang                 return -1;
6365193e5fbSJason Wang             }
6375193e5fbSJason Wang         } else {
6385193e5fbSJason Wang             vhostfd = -1;
6395193e5fbSJason Wang         }
6405193e5fbSJason Wang 
6415193e5fbSJason Wang         s->vhost_net = vhost_net_init(&s->nc, vhostfd,
6425193e5fbSJason Wang                                       tap->has_vhostforce && tap->vhostforce);
6435193e5fbSJason Wang         if (!s->vhost_net) {
6445193e5fbSJason Wang             error_report("vhost-net requested but could not be initialized");
6455193e5fbSJason Wang             return -1;
6465193e5fbSJason Wang         }
647264986e2SJason Wang     } else if (tap->has_vhostfd || tap->has_vhostfds) {
6485193e5fbSJason Wang         error_report("vhostfd= is not valid without vhost");
6495193e5fbSJason Wang         return -1;
6505193e5fbSJason Wang     }
6515193e5fbSJason Wang 
6525193e5fbSJason Wang     return 0;
6535193e5fbSJason Wang }
6545193e5fbSJason Wang 
655264986e2SJason Wang static int get_fds(char *str, char *fds[], int max)
656264986e2SJason Wang {
657264986e2SJason Wang     char *ptr = str, *this;
658264986e2SJason Wang     size_t len = strlen(str);
659264986e2SJason Wang     int i = 0;
660264986e2SJason Wang 
661264986e2SJason Wang     while (i < max && ptr < str + len) {
662264986e2SJason Wang         this = strchr(ptr, ':');
663264986e2SJason Wang 
664264986e2SJason Wang         if (this == NULL) {
665264986e2SJason Wang             fds[i] = g_strdup(ptr);
666264986e2SJason Wang         } else {
667264986e2SJason Wang             fds[i] = g_strndup(ptr, this - ptr);
668264986e2SJason Wang         }
669264986e2SJason Wang 
670264986e2SJason Wang         i++;
671264986e2SJason Wang         if (this == NULL) {
672264986e2SJason Wang             break;
673264986e2SJason Wang         } else {
674264986e2SJason Wang             ptr = this + 1;
675264986e2SJason Wang         }
676264986e2SJason Wang     }
677264986e2SJason Wang 
678264986e2SJason Wang     return i;
679264986e2SJason Wang }
680264986e2SJason Wang 
6811a0c0958SLaszlo Ersek int net_init_tap(const NetClientOptions *opts, const char *name,
6824e68f7a0SStefan Hajnoczi                  NetClientState *peer)
6835281d757SMark McLoughlin {
68408c573a8SLaszlo Ersek     const NetdevTapOptions *tap;
685264986e2SJason Wang     int fd, vnet_hdr = 0, i = 0, queues;
68608c573a8SLaszlo Ersek     /* for the no-fd, no-helper case */
68708c573a8SLaszlo Ersek     const char *script = NULL; /* suppress wrong "uninit'd use" gcc warning */
6885193e5fbSJason Wang     const char *downscript = NULL;
689264986e2SJason Wang     const char *vhostfdname;
69008c573a8SLaszlo Ersek     char ifname[128];
69108c573a8SLaszlo Ersek 
69208c573a8SLaszlo Ersek     assert(opts->kind == NET_CLIENT_OPTIONS_KIND_TAP);
69308c573a8SLaszlo Ersek     tap = opts->tap;
694264986e2SJason Wang     queues = tap->has_queues ? tap->queues : 1;
695264986e2SJason Wang     vhostfdname = tap->has_vhostfd ? tap->vhostfd : NULL;
69608c573a8SLaszlo Ersek 
697ce675a75SJason Wang     /* QEMU vlans does not support multiqueue tap, in this case peer is set.
698ce675a75SJason Wang      * For -netdev, peer is always NULL. */
699ce675a75SJason Wang     if (peer && (tap->has_queues || tap->has_fds || tap->has_vhostfds)) {
700085d8134SPeter Maydell         error_report("Multiqueue tap cannot be used with QEMU vlans");
701ce675a75SJason Wang         return -1;
702ce675a75SJason Wang     }
703ce675a75SJason Wang 
70408c573a8SLaszlo Ersek     if (tap->has_fd) {
70508c573a8SLaszlo Ersek         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
706264986e2SJason Wang             tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
707c87826a8SJason Wang             tap->has_fds || tap->has_vhostfds) {
708a7c36ee4SCorey Bryant             error_report("ifname=, script=, downscript=, vnet_hdr=, "
709c87826a8SJason Wang                          "helper=, queues=, fds=, and vhostfds= "
710c87826a8SJason Wang                          "are invalid with fd=");
7115281d757SMark McLoughlin             return -1;
7125281d757SMark McLoughlin         }
7135281d757SMark McLoughlin 
714a96ed02fSNicholas Bellinger         fd = monitor_handle_fd_param(cur_mon, tap->fd);
7155281d757SMark McLoughlin         if (fd == -1) {
7165281d757SMark McLoughlin             return -1;
7175281d757SMark McLoughlin         }
7185281d757SMark McLoughlin 
7195281d757SMark McLoughlin         fcntl(fd, F_SETFL, O_NONBLOCK);
7205281d757SMark McLoughlin 
7215281d757SMark McLoughlin         vnet_hdr = tap_probe_vnet_hdr(fd);
722a7c36ee4SCorey Bryant 
72302cd8090SAnthony Liguori         if (net_init_tap_one(tap, peer, "tap", name, NULL,
724264986e2SJason Wang                              script, downscript,
725264986e2SJason Wang                              vhostfdname, vnet_hdr, fd)) {
726264986e2SJason Wang             return -1;
727264986e2SJason Wang         }
728264986e2SJason Wang     } else if (tap->has_fds) {
729264986e2SJason Wang         char *fds[MAX_TAP_QUEUES];
730264986e2SJason Wang         char *vhost_fds[MAX_TAP_QUEUES];
731264986e2SJason Wang         int nfds, nvhosts;
732a7c36ee4SCorey Bryant 
733264986e2SJason Wang         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
734264986e2SJason Wang             tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
735c87826a8SJason Wang             tap->has_vhostfd) {
736264986e2SJason Wang             error_report("ifname=, script=, downscript=, vnet_hdr=, "
737c87826a8SJason Wang                          "helper=, queues=, and vhostfd= "
738c87826a8SJason Wang                          "are invalid with fds=");
739264986e2SJason Wang             return -1;
740264986e2SJason Wang         }
741264986e2SJason Wang 
742264986e2SJason Wang         nfds = get_fds(tap->fds, fds, MAX_TAP_QUEUES);
743264986e2SJason Wang         if (tap->has_vhostfds) {
744264986e2SJason Wang             nvhosts = get_fds(tap->vhostfds, vhost_fds, MAX_TAP_QUEUES);
745264986e2SJason Wang             if (nfds != nvhosts) {
746264986e2SJason Wang                 error_report("The number of fds passed does not match the "
747264986e2SJason Wang                              "number of vhostfds passed");
748264986e2SJason Wang                 return -1;
749264986e2SJason Wang             }
750264986e2SJason Wang         }
751264986e2SJason Wang 
752264986e2SJason Wang         for (i = 0; i < nfds; i++) {
753264986e2SJason Wang             fd = monitor_handle_fd_param(cur_mon, fds[i]);
754264986e2SJason Wang             if (fd == -1) {
755264986e2SJason Wang                 return -1;
756264986e2SJason Wang             }
757264986e2SJason Wang 
758264986e2SJason Wang             fcntl(fd, F_SETFL, O_NONBLOCK);
759264986e2SJason Wang 
760264986e2SJason Wang             if (i == 0) {
761264986e2SJason Wang                 vnet_hdr = tap_probe_vnet_hdr(fd);
762264986e2SJason Wang             } else if (vnet_hdr != tap_probe_vnet_hdr(fd)) {
763264986e2SJason Wang                 error_report("vnet_hdr not consistent across given tap fds");
764264986e2SJason Wang                 return -1;
765264986e2SJason Wang             }
766264986e2SJason Wang 
767264986e2SJason Wang             if (net_init_tap_one(tap, peer, "tap", name, ifname,
768264986e2SJason Wang                                  script, downscript,
769264986e2SJason Wang                                  tap->has_vhostfds ? vhost_fds[i] : NULL,
770264986e2SJason Wang                                  vnet_hdr, fd)) {
771264986e2SJason Wang                 return -1;
772264986e2SJason Wang             }
773264986e2SJason Wang         }
77408c573a8SLaszlo Ersek     } else if (tap->has_helper) {
77508c573a8SLaszlo Ersek         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
776c87826a8SJason Wang             tap->has_vnet_hdr || tap->has_queues || tap->has_vhostfds) {
777a7c36ee4SCorey Bryant             error_report("ifname=, script=, downscript=, and vnet_hdr= "
778c87826a8SJason Wang                          "queues=, and vhostfds= are invalid with helper=");
779a7c36ee4SCorey Bryant             return -1;
780a7c36ee4SCorey Bryant         }
781a7c36ee4SCorey Bryant 
78208c573a8SLaszlo Ersek         fd = net_bridge_run_helper(tap->helper, DEFAULT_BRIDGE_INTERFACE);
783a7c36ee4SCorey Bryant         if (fd == -1) {
784a7c36ee4SCorey Bryant             return -1;
785a7c36ee4SCorey Bryant         }
786a7c36ee4SCorey Bryant 
787a7c36ee4SCorey Bryant         fcntl(fd, F_SETFL, O_NONBLOCK);
788a7c36ee4SCorey Bryant         vnet_hdr = tap_probe_vnet_hdr(fd);
789a7c36ee4SCorey Bryant 
790264986e2SJason Wang         if (net_init_tap_one(tap, peer, "bridge", name, ifname,
791264986e2SJason Wang                              script, downscript, vhostfdname,
792264986e2SJason Wang                              vnet_hdr, fd)) {
793264986e2SJason Wang             return -1;
794264986e2SJason Wang         }
7955281d757SMark McLoughlin     } else {
796c87826a8SJason Wang         if (tap->has_vhostfds) {
797c87826a8SJason Wang             error_report("vhostfds= is invalid if fds= wasn't specified");
798c87826a8SJason Wang             return -1;
799c87826a8SJason Wang         }
80008c573a8SLaszlo Ersek         script = tap->has_script ? tap->script : DEFAULT_NETWORK_SCRIPT;
8015193e5fbSJason Wang         downscript = tap->has_downscript ? tap->downscript :
8025193e5fbSJason Wang             DEFAULT_NETWORK_DOWN_SCRIPT;
803264986e2SJason Wang 
804264986e2SJason Wang         if (tap->has_ifname) {
805264986e2SJason Wang             pstrcpy(ifname, sizeof ifname, tap->ifname);
806264986e2SJason Wang         } else {
807264986e2SJason Wang             ifname[0] = '\0';
808264986e2SJason Wang         }
809264986e2SJason Wang 
810264986e2SJason Wang         for (i = 0; i < queues; i++) {
811264986e2SJason Wang             fd = net_tap_init(tap, &vnet_hdr, i >= 1 ? "no" : script,
812264986e2SJason Wang                               ifname, sizeof ifname, queues > 1);
813929fe497SJuergen Lock             if (fd == -1) {
814929fe497SJuergen Lock                 return -1;
815929fe497SJuergen Lock             }
816a7c36ee4SCorey Bryant 
817264986e2SJason Wang             if (queues > 1 && i == 0 && !tap->has_ifname) {
818264986e2SJason Wang                 if (tap_fd_get_ifname(fd, ifname)) {
819264986e2SJason Wang                     error_report("Fail to get ifname");
820264986e2SJason Wang                     return -1;
821264986e2SJason Wang                 }
8225281d757SMark McLoughlin             }
8235281d757SMark McLoughlin 
824264986e2SJason Wang             if (net_init_tap_one(tap, peer, "tap", name, ifname,
825264986e2SJason Wang                                  i >= 1 ? "no" : script,
826264986e2SJason Wang                                  i >= 1 ? "no" : downscript,
827264986e2SJason Wang                                  vhostfdname, vnet_hdr, fd)) {
828264986e2SJason Wang                 return -1;
829264986e2SJason Wang             }
830264986e2SJason Wang         }
831264986e2SJason Wang     }
832264986e2SJason Wang 
833264986e2SJason Wang     return 0;
8345281d757SMark McLoughlin }
835b202554cSMichael S. Tsirkin 
8364e68f7a0SStefan Hajnoczi VHostNetState *tap_get_vhost_net(NetClientState *nc)
837b202554cSMichael S. Tsirkin {
838b202554cSMichael S. Tsirkin     TAPState *s = DO_UPCAST(TAPState, nc, nc);
8392be64a68SLaszlo Ersek     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
840b202554cSMichael S. Tsirkin     return s->vhost_net;
841b202554cSMichael S. Tsirkin }
84216dbaf90SJason Wang 
84316dbaf90SJason Wang int tap_enable(NetClientState *nc)
84416dbaf90SJason Wang {
84516dbaf90SJason Wang     TAPState *s = DO_UPCAST(TAPState, nc, nc);
84616dbaf90SJason Wang     int ret;
84716dbaf90SJason Wang 
84816dbaf90SJason Wang     if (s->enabled) {
84916dbaf90SJason Wang         return 0;
85016dbaf90SJason Wang     } else {
85116dbaf90SJason Wang         ret = tap_fd_enable(s->fd);
85216dbaf90SJason Wang         if (ret == 0) {
85316dbaf90SJason Wang             s->enabled = true;
85416dbaf90SJason Wang             tap_update_fd_handler(s);
85516dbaf90SJason Wang         }
85616dbaf90SJason Wang         return ret;
85716dbaf90SJason Wang     }
85816dbaf90SJason Wang }
85916dbaf90SJason Wang 
86016dbaf90SJason Wang int tap_disable(NetClientState *nc)
86116dbaf90SJason Wang {
86216dbaf90SJason Wang     TAPState *s = DO_UPCAST(TAPState, nc, nc);
86316dbaf90SJason Wang     int ret;
86416dbaf90SJason Wang 
86516dbaf90SJason Wang     if (s->enabled == 0) {
86616dbaf90SJason Wang         return 0;
86716dbaf90SJason Wang     } else {
86816dbaf90SJason Wang         ret = tap_fd_disable(s->fd);
86916dbaf90SJason Wang         if (ret == 0) {
87016dbaf90SJason Wang             qemu_purge_queued_packets(nc);
87116dbaf90SJason Wang             s->enabled = false;
87216dbaf90SJason Wang             tap_update_fd_handler(s);
87316dbaf90SJason Wang         }
87416dbaf90SJason Wang         return ret;
87516dbaf90SJason Wang     }
87616dbaf90SJason Wang }
877