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; 192*756ae78bSWangkai (Kevin,C) int packets = 0; 1935281d757SMark McLoughlin 19468e5ec64SStefan Hajnoczi while (qemu_can_send_packet(&s->nc)) { 1955819c918SMark McLoughlin uint8_t *buf = s->buf; 1965819c918SMark McLoughlin 1975281d757SMark McLoughlin size = tap_read_packet(s->fd, s->buf, sizeof(s->buf)); 1985281d757SMark McLoughlin if (size <= 0) { 1995819c918SMark McLoughlin break; 2005281d757SMark McLoughlin } 2015281d757SMark McLoughlin 202ef4252b1SMichael S. Tsirkin if (s->host_vnet_hdr_len && !s->using_vnet_hdr) { 203ef4252b1SMichael S. Tsirkin buf += s->host_vnet_hdr_len; 204ef4252b1SMichael S. Tsirkin size -= s->host_vnet_hdr_len; 2055281d757SMark McLoughlin } 2065281d757SMark McLoughlin 2073e35ba93SMark McLoughlin size = qemu_send_packet_async(&s->nc, buf, size, tap_send_completed); 2085281d757SMark McLoughlin if (size == 0) { 209ec45f083SJason Wang tap_read_poll(s, false); 21068e5ec64SStefan Hajnoczi break; 21168e5ec64SStefan Hajnoczi } else if (size < 0) { 21268e5ec64SStefan Hajnoczi break; 2135281d757SMark McLoughlin } 214*756ae78bSWangkai (Kevin,C) 215*756ae78bSWangkai (Kevin,C) /* 216*756ae78bSWangkai (Kevin,C) * When the host keeps receiving more packets while tap_send() is 217*756ae78bSWangkai (Kevin,C) * running we can hog the QEMU global mutex. Limit the number of 218*756ae78bSWangkai (Kevin,C) * packets that are processed per tap_send() callback to prevent 219*756ae78bSWangkai (Kevin,C) * stalling the guest. 220*756ae78bSWangkai (Kevin,C) */ 221*756ae78bSWangkai (Kevin,C) packets++; 222*756ae78bSWangkai (Kevin,C) if (packets >= 50) { 223*756ae78bSWangkai (Kevin,C) break; 224*756ae78bSWangkai (Kevin,C) } 22568e5ec64SStefan Hajnoczi } 2265281d757SMark McLoughlin } 2275281d757SMark McLoughlin 2283bac80d3SVincenzo Maffione static bool tap_has_ufo(NetClientState *nc) 2295281d757SMark McLoughlin { 2303e35ba93SMark McLoughlin TAPState *s = DO_UPCAST(TAPState, nc, nc); 2315281d757SMark McLoughlin 2322be64a68SLaszlo Ersek assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP); 2335281d757SMark McLoughlin 2345281d757SMark McLoughlin return s->has_ufo; 2355281d757SMark McLoughlin } 2365281d757SMark McLoughlin 2373bac80d3SVincenzo Maffione static bool tap_has_vnet_hdr(NetClientState *nc) 2385281d757SMark McLoughlin { 2393e35ba93SMark McLoughlin TAPState *s = DO_UPCAST(TAPState, nc, nc); 2405281d757SMark McLoughlin 2412be64a68SLaszlo Ersek assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP); 2425281d757SMark McLoughlin 243ef4252b1SMichael S. Tsirkin return !!s->host_vnet_hdr_len; 2445281d757SMark McLoughlin } 2455281d757SMark McLoughlin 2463bac80d3SVincenzo Maffione static bool tap_has_vnet_hdr_len(NetClientState *nc, int len) 247445d892fSMichael S. Tsirkin { 248445d892fSMichael S. Tsirkin TAPState *s = DO_UPCAST(TAPState, nc, nc); 249445d892fSMichael S. Tsirkin 2502be64a68SLaszlo Ersek assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP); 251445d892fSMichael S. Tsirkin 252e96dfd11SVincenzo Maffione return !!tap_probe_vnet_hdr_len(s->fd, len); 253445d892fSMichael S. Tsirkin } 254445d892fSMichael S. Tsirkin 2553bac80d3SVincenzo Maffione static void tap_set_vnet_hdr_len(NetClientState *nc, int len) 256445d892fSMichael S. Tsirkin { 257445d892fSMichael S. Tsirkin TAPState *s = DO_UPCAST(TAPState, nc, nc); 258445d892fSMichael S. Tsirkin 2592be64a68SLaszlo Ersek assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP); 260445d892fSMichael S. Tsirkin assert(len == sizeof(struct virtio_net_hdr_mrg_rxbuf) || 261445d892fSMichael S. Tsirkin len == sizeof(struct virtio_net_hdr)); 262445d892fSMichael S. Tsirkin 263445d892fSMichael S. Tsirkin tap_fd_set_vnet_hdr_len(s->fd, len); 264445d892fSMichael S. Tsirkin s->host_vnet_hdr_len = len; 265445d892fSMichael S. Tsirkin } 266445d892fSMichael S. Tsirkin 2673bac80d3SVincenzo Maffione static void tap_using_vnet_hdr(NetClientState *nc, bool using_vnet_hdr) 2685281d757SMark McLoughlin { 2693e35ba93SMark McLoughlin TAPState *s = DO_UPCAST(TAPState, nc, nc); 2705281d757SMark McLoughlin 2712be64a68SLaszlo Ersek assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP); 272ef4252b1SMichael S. Tsirkin assert(!!s->host_vnet_hdr_len == using_vnet_hdr); 2735281d757SMark McLoughlin 2745281d757SMark McLoughlin s->using_vnet_hdr = using_vnet_hdr; 2755281d757SMark McLoughlin } 2765281d757SMark McLoughlin 2773bac80d3SVincenzo Maffione static void tap_set_offload(NetClientState *nc, int csum, int tso4, 2785281d757SMark McLoughlin int tso6, int ecn, int ufo) 2795281d757SMark McLoughlin { 2803e35ba93SMark McLoughlin TAPState *s = DO_UPCAST(TAPState, nc, nc); 28127a6375dSMichael S. Tsirkin if (s->fd < 0) { 28227a6375dSMichael S. Tsirkin return; 28327a6375dSMichael S. Tsirkin } 2845281d757SMark McLoughlin 28527a6375dSMichael S. Tsirkin tap_fd_set_offload(s->fd, csum, tso4, tso6, ecn, ufo); 2865281d757SMark McLoughlin } 2875281d757SMark McLoughlin 2884e68f7a0SStefan Hajnoczi static void tap_cleanup(NetClientState *nc) 2895281d757SMark McLoughlin { 2903e35ba93SMark McLoughlin TAPState *s = DO_UPCAST(TAPState, nc, nc); 2915281d757SMark McLoughlin 29282b0d80eSMichael S. Tsirkin if (s->vhost_net) { 29382b0d80eSMichael S. Tsirkin vhost_net_cleanup(s->vhost_net); 29443849424SMichael S. Tsirkin s->vhost_net = NULL; 29582b0d80eSMichael S. Tsirkin } 29682b0d80eSMichael S. Tsirkin 2973e35ba93SMark McLoughlin qemu_purge_queued_packets(nc); 2985281d757SMark McLoughlin 2995281d757SMark McLoughlin if (s->down_script[0]) 3005281d757SMark McLoughlin launch_script(s->down_script, s->down_script_arg, s->fd); 3015281d757SMark McLoughlin 302ec45f083SJason Wang tap_read_poll(s, false); 303ec45f083SJason Wang tap_write_poll(s, false); 3045281d757SMark McLoughlin close(s->fd); 30527a6375dSMichael S. Tsirkin s->fd = -1; 3065281d757SMark McLoughlin } 3075281d757SMark McLoughlin 3084e68f7a0SStefan Hajnoczi static void tap_poll(NetClientState *nc, bool enable) 309ceb69615SMichael S. Tsirkin { 310ceb69615SMichael S. Tsirkin TAPState *s = DO_UPCAST(TAPState, nc, nc); 311ceb69615SMichael S. Tsirkin tap_read_poll(s, enable); 312ceb69615SMichael S. Tsirkin tap_write_poll(s, enable); 313ceb69615SMichael S. Tsirkin } 314ceb69615SMichael S. Tsirkin 3154e68f7a0SStefan Hajnoczi int tap_get_fd(NetClientState *nc) 31695d528a2SMichael S. Tsirkin { 31795d528a2SMichael S. Tsirkin TAPState *s = DO_UPCAST(TAPState, nc, nc); 3182be64a68SLaszlo Ersek assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP); 31995d528a2SMichael S. Tsirkin return s->fd; 32095d528a2SMichael S. Tsirkin } 32195d528a2SMichael S. Tsirkin 3225281d757SMark McLoughlin /* fd support */ 3235281d757SMark McLoughlin 3243e35ba93SMark McLoughlin static NetClientInfo net_tap_info = { 3252be64a68SLaszlo Ersek .type = NET_CLIENT_OPTIONS_KIND_TAP, 3263e35ba93SMark McLoughlin .size = sizeof(TAPState), 3273e35ba93SMark McLoughlin .receive = tap_receive, 3283e35ba93SMark McLoughlin .receive_raw = tap_receive_raw, 3293e35ba93SMark McLoughlin .receive_iov = tap_receive_iov, 330ceb69615SMichael S. Tsirkin .poll = tap_poll, 3313e35ba93SMark McLoughlin .cleanup = tap_cleanup, 3322e753bccSVincenzo Maffione .has_ufo = tap_has_ufo, 3332e753bccSVincenzo Maffione .has_vnet_hdr = tap_has_vnet_hdr, 3342e753bccSVincenzo Maffione .has_vnet_hdr_len = tap_has_vnet_hdr_len, 3352e753bccSVincenzo Maffione .using_vnet_hdr = tap_using_vnet_hdr, 3362e753bccSVincenzo Maffione .set_offload = tap_set_offload, 3372e753bccSVincenzo Maffione .set_vnet_hdr_len = tap_set_vnet_hdr_len, 3383e35ba93SMark McLoughlin }; 3393e35ba93SMark McLoughlin 3404e68f7a0SStefan Hajnoczi static TAPState *net_tap_fd_init(NetClientState *peer, 3415281d757SMark McLoughlin const char *model, 3425281d757SMark McLoughlin const char *name, 3435281d757SMark McLoughlin int fd, 3445281d757SMark McLoughlin int vnet_hdr) 3455281d757SMark McLoughlin { 3464e68f7a0SStefan Hajnoczi NetClientState *nc; 3475281d757SMark McLoughlin TAPState *s; 3485281d757SMark McLoughlin 349ab5f3f84SStefan Hajnoczi nc = qemu_new_net_client(&net_tap_info, peer, model, name); 3503e35ba93SMark McLoughlin 3513e35ba93SMark McLoughlin s = DO_UPCAST(TAPState, nc, nc); 3523e35ba93SMark McLoughlin 3535281d757SMark McLoughlin s->fd = fd; 354ef4252b1SMichael S. Tsirkin s->host_vnet_hdr_len = vnet_hdr ? sizeof(struct virtio_net_hdr) : 0; 355ec45f083SJason Wang s->using_vnet_hdr = false; 3569c282718SMark McLoughlin s->has_ufo = tap_probe_has_ufo(s->fd); 35716dbaf90SJason Wang s->enabled = true; 3583e35ba93SMark McLoughlin tap_set_offload(&s->nc, 0, 0, 0, 0, 0); 35958ddcd50SMichael S. Tsirkin /* 36058ddcd50SMichael S. Tsirkin * Make sure host header length is set correctly in tap: 36158ddcd50SMichael S. Tsirkin * it might have been modified by another instance of qemu. 36258ddcd50SMichael S. Tsirkin */ 36358ddcd50SMichael S. Tsirkin if (tap_probe_vnet_hdr_len(s->fd, s->host_vnet_hdr_len)) { 36458ddcd50SMichael S. Tsirkin tap_fd_set_vnet_hdr_len(s->fd, s->host_vnet_hdr_len); 36558ddcd50SMichael S. Tsirkin } 366ec45f083SJason Wang tap_read_poll(s, true); 36782b0d80eSMichael S. Tsirkin s->vhost_net = NULL; 3685281d757SMark McLoughlin return s; 3695281d757SMark McLoughlin } 3705281d757SMark McLoughlin 3715281d757SMark McLoughlin static int launch_script(const char *setup_script, const char *ifname, int fd) 3725281d757SMark McLoughlin { 3735281d757SMark McLoughlin int pid, status; 3745281d757SMark McLoughlin char *args[3]; 3755281d757SMark McLoughlin char **parg; 3765281d757SMark McLoughlin 3775281d757SMark McLoughlin /* try to launch network script */ 3785281d757SMark McLoughlin pid = fork(); 3795281d757SMark McLoughlin if (pid == 0) { 3805281d757SMark McLoughlin int open_max = sysconf(_SC_OPEN_MAX), i; 3815281d757SMark McLoughlin 38213a12f86SPankaj Gupta for (i = 3; i < open_max; i++) { 38313a12f86SPankaj Gupta if (i != fd) { 3845281d757SMark McLoughlin close(i); 3855281d757SMark McLoughlin } 3865281d757SMark McLoughlin } 3875281d757SMark McLoughlin parg = args; 3885281d757SMark McLoughlin *parg++ = (char *)setup_script; 3895281d757SMark McLoughlin *parg++ = (char *)ifname; 3909678d950SBlue Swirl *parg = NULL; 3915281d757SMark McLoughlin execv(setup_script, args); 3925281d757SMark McLoughlin _exit(1); 3935281d757SMark McLoughlin } else if (pid > 0) { 3945281d757SMark McLoughlin while (waitpid(pid, &status, 0) != pid) { 3955281d757SMark McLoughlin /* loop */ 3965281d757SMark McLoughlin } 3975281d757SMark McLoughlin 3985281d757SMark McLoughlin if (WIFEXITED(status) && WEXITSTATUS(status) == 0) { 3995281d757SMark McLoughlin return 0; 4005281d757SMark McLoughlin } 4015281d757SMark McLoughlin } 4025281d757SMark McLoughlin fprintf(stderr, "%s: could not launch network script\n", setup_script); 4035281d757SMark McLoughlin return -1; 4045281d757SMark McLoughlin } 4055281d757SMark McLoughlin 406a7c36ee4SCorey Bryant static int recv_fd(int c) 407a7c36ee4SCorey Bryant { 408a7c36ee4SCorey Bryant int fd; 409a7c36ee4SCorey Bryant uint8_t msgbuf[CMSG_SPACE(sizeof(fd))]; 410a7c36ee4SCorey Bryant struct msghdr msg = { 411a7c36ee4SCorey Bryant .msg_control = msgbuf, 412a7c36ee4SCorey Bryant .msg_controllen = sizeof(msgbuf), 413a7c36ee4SCorey Bryant }; 414a7c36ee4SCorey Bryant struct cmsghdr *cmsg; 415a7c36ee4SCorey Bryant struct iovec iov; 416a7c36ee4SCorey Bryant uint8_t req[1]; 417a7c36ee4SCorey Bryant ssize_t len; 418a7c36ee4SCorey Bryant 419a7c36ee4SCorey Bryant cmsg = CMSG_FIRSTHDR(&msg); 420a7c36ee4SCorey Bryant cmsg->cmsg_level = SOL_SOCKET; 421a7c36ee4SCorey Bryant cmsg->cmsg_type = SCM_RIGHTS; 422a7c36ee4SCorey Bryant cmsg->cmsg_len = CMSG_LEN(sizeof(fd)); 423a7c36ee4SCorey Bryant msg.msg_controllen = cmsg->cmsg_len; 424a7c36ee4SCorey Bryant 425a7c36ee4SCorey Bryant iov.iov_base = req; 426a7c36ee4SCorey Bryant iov.iov_len = sizeof(req); 427a7c36ee4SCorey Bryant 428a7c36ee4SCorey Bryant msg.msg_iov = &iov; 429a7c36ee4SCorey Bryant msg.msg_iovlen = 1; 430a7c36ee4SCorey Bryant 431a7c36ee4SCorey Bryant len = recvmsg(c, &msg, 0); 432a7c36ee4SCorey Bryant if (len > 0) { 433a7c36ee4SCorey Bryant memcpy(&fd, CMSG_DATA(cmsg), sizeof(fd)); 434a7c36ee4SCorey Bryant return fd; 435a7c36ee4SCorey Bryant } 436a7c36ee4SCorey Bryant 437a7c36ee4SCorey Bryant return len; 438a7c36ee4SCorey Bryant } 439a7c36ee4SCorey Bryant 440a7c36ee4SCorey Bryant static int net_bridge_run_helper(const char *helper, const char *bridge) 441a7c36ee4SCorey Bryant { 442a7c36ee4SCorey Bryant sigset_t oldmask, mask; 443a7c36ee4SCorey Bryant int pid, status; 444a7c36ee4SCorey Bryant char *args[5]; 445a7c36ee4SCorey Bryant char **parg; 446a7c36ee4SCorey Bryant int sv[2]; 447a7c36ee4SCorey Bryant 448a7c36ee4SCorey Bryant sigemptyset(&mask); 449a7c36ee4SCorey Bryant sigaddset(&mask, SIGCHLD); 450a7c36ee4SCorey Bryant sigprocmask(SIG_BLOCK, &mask, &oldmask); 451a7c36ee4SCorey Bryant 452a7c36ee4SCorey Bryant if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) { 453a7c36ee4SCorey Bryant return -1; 454a7c36ee4SCorey Bryant } 455a7c36ee4SCorey Bryant 456a7c36ee4SCorey Bryant /* try to launch bridge helper */ 457a7c36ee4SCorey Bryant pid = fork(); 458a7c36ee4SCorey Bryant if (pid == 0) { 459a7c36ee4SCorey Bryant int open_max = sysconf(_SC_OPEN_MAX), i; 460a7c36ee4SCorey Bryant char fd_buf[6+10]; 461a7c36ee4SCorey Bryant char br_buf[6+IFNAMSIZ] = {0}; 462a7c36ee4SCorey Bryant char helper_cmd[PATH_MAX + sizeof(fd_buf) + sizeof(br_buf) + 15]; 463a7c36ee4SCorey Bryant 46413a12f86SPankaj Gupta for (i = 3; i < open_max; i++) { 46513a12f86SPankaj Gupta if (i != sv[1]) { 466a7c36ee4SCorey Bryant close(i); 467a7c36ee4SCorey Bryant } 468a7c36ee4SCorey Bryant } 469a7c36ee4SCorey Bryant 470a7c36ee4SCorey Bryant snprintf(fd_buf, sizeof(fd_buf), "%s%d", "--fd=", sv[1]); 471a7c36ee4SCorey Bryant 472a7c36ee4SCorey Bryant if (strrchr(helper, ' ') || strrchr(helper, '\t')) { 473a7c36ee4SCorey Bryant /* assume helper is a command */ 474a7c36ee4SCorey Bryant 475a7c36ee4SCorey Bryant if (strstr(helper, "--br=") == NULL) { 476a7c36ee4SCorey Bryant snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge); 477a7c36ee4SCorey Bryant } 478a7c36ee4SCorey Bryant 479a7c36ee4SCorey Bryant snprintf(helper_cmd, sizeof(helper_cmd), "%s %s %s %s", 480a7c36ee4SCorey Bryant helper, "--use-vnet", fd_buf, br_buf); 481a7c36ee4SCorey Bryant 482a7c36ee4SCorey Bryant parg = args; 483a7c36ee4SCorey Bryant *parg++ = (char *)"sh"; 484a7c36ee4SCorey Bryant *parg++ = (char *)"-c"; 485a7c36ee4SCorey Bryant *parg++ = helper_cmd; 486a7c36ee4SCorey Bryant *parg++ = NULL; 487a7c36ee4SCorey Bryant 488a7c36ee4SCorey Bryant execv("/bin/sh", args); 489a7c36ee4SCorey Bryant } else { 490a7c36ee4SCorey Bryant /* assume helper is just the executable path name */ 491a7c36ee4SCorey Bryant 492a7c36ee4SCorey Bryant snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge); 493a7c36ee4SCorey Bryant 494a7c36ee4SCorey Bryant parg = args; 495a7c36ee4SCorey Bryant *parg++ = (char *)helper; 496a7c36ee4SCorey Bryant *parg++ = (char *)"--use-vnet"; 497a7c36ee4SCorey Bryant *parg++ = fd_buf; 498a7c36ee4SCorey Bryant *parg++ = br_buf; 499a7c36ee4SCorey Bryant *parg++ = NULL; 500a7c36ee4SCorey Bryant 501a7c36ee4SCorey Bryant execv(helper, args); 502a7c36ee4SCorey Bryant } 503a7c36ee4SCorey Bryant _exit(1); 504a7c36ee4SCorey Bryant 505a7c36ee4SCorey Bryant } else if (pid > 0) { 506a7c36ee4SCorey Bryant int fd; 507a7c36ee4SCorey Bryant 508a7c36ee4SCorey Bryant close(sv[1]); 509a7c36ee4SCorey Bryant 510a7c36ee4SCorey Bryant do { 511a7c36ee4SCorey Bryant fd = recv_fd(sv[0]); 512a7c36ee4SCorey Bryant } while (fd == -1 && errno == EINTR); 513a7c36ee4SCorey Bryant 514a7c36ee4SCorey Bryant close(sv[0]); 515a7c36ee4SCorey Bryant 516a7c36ee4SCorey Bryant while (waitpid(pid, &status, 0) != pid) { 517a7c36ee4SCorey Bryant /* loop */ 518a7c36ee4SCorey Bryant } 519a7c36ee4SCorey Bryant sigprocmask(SIG_SETMASK, &oldmask, NULL); 520a7c36ee4SCorey Bryant if (fd < 0) { 521a7c36ee4SCorey Bryant fprintf(stderr, "failed to recv file descriptor\n"); 522a7c36ee4SCorey Bryant return -1; 523a7c36ee4SCorey Bryant } 524a7c36ee4SCorey Bryant 525a7c36ee4SCorey Bryant if (WIFEXITED(status) && WEXITSTATUS(status) == 0) { 526a7c36ee4SCorey Bryant return fd; 527a7c36ee4SCorey Bryant } 528a7c36ee4SCorey Bryant } 529a7c36ee4SCorey Bryant fprintf(stderr, "failed to launch bridge helper\n"); 530a7c36ee4SCorey Bryant return -1; 531a7c36ee4SCorey Bryant } 532a7c36ee4SCorey Bryant 5331a0c0958SLaszlo Ersek int net_init_bridge(const NetClientOptions *opts, const char *name, 5344e68f7a0SStefan Hajnoczi NetClientState *peer) 535a7c36ee4SCorey Bryant { 536f79b51b0SLaszlo Ersek const NetdevBridgeOptions *bridge; 537f79b51b0SLaszlo Ersek const char *helper, *br; 538f79b51b0SLaszlo Ersek 539a7c36ee4SCorey Bryant TAPState *s; 540a7c36ee4SCorey Bryant int fd, vnet_hdr; 541a7c36ee4SCorey Bryant 542f79b51b0SLaszlo Ersek assert(opts->kind == NET_CLIENT_OPTIONS_KIND_BRIDGE); 543f79b51b0SLaszlo Ersek bridge = opts->bridge; 544a7c36ee4SCorey Bryant 545f79b51b0SLaszlo Ersek helper = bridge->has_helper ? bridge->helper : DEFAULT_BRIDGE_HELPER; 546f79b51b0SLaszlo Ersek br = bridge->has_br ? bridge->br : DEFAULT_BRIDGE_INTERFACE; 547f79b51b0SLaszlo Ersek 548f79b51b0SLaszlo Ersek fd = net_bridge_run_helper(helper, br); 549a7c36ee4SCorey Bryant if (fd == -1) { 550a7c36ee4SCorey Bryant return -1; 551a7c36ee4SCorey Bryant } 552a7c36ee4SCorey Bryant 553a7c36ee4SCorey Bryant fcntl(fd, F_SETFL, O_NONBLOCK); 554a7c36ee4SCorey Bryant 555a7c36ee4SCorey Bryant vnet_hdr = tap_probe_vnet_hdr(fd); 556a7c36ee4SCorey Bryant 557d33d93b2SStefan Hajnoczi s = net_tap_fd_init(peer, "bridge", name, fd, vnet_hdr); 558a7c36ee4SCorey Bryant if (!s) { 559a7c36ee4SCorey Bryant close(fd); 560a7c36ee4SCorey Bryant return -1; 561a7c36ee4SCorey Bryant } 562a7c36ee4SCorey Bryant 563f79b51b0SLaszlo Ersek snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s,br=%s", helper, 564f79b51b0SLaszlo Ersek br); 565a7c36ee4SCorey Bryant 566a7c36ee4SCorey Bryant return 0; 567a7c36ee4SCorey Bryant } 568a7c36ee4SCorey Bryant 56908c573a8SLaszlo Ersek static int net_tap_init(const NetdevTapOptions *tap, int *vnet_hdr, 57008c573a8SLaszlo Ersek const char *setup_script, char *ifname, 571264986e2SJason Wang size_t ifname_sz, int mq_required) 5725281d757SMark McLoughlin { 5735281d757SMark McLoughlin int fd, vnet_hdr_required; 5745281d757SMark McLoughlin 57508c573a8SLaszlo Ersek if (tap->has_vnet_hdr) { 57608c573a8SLaszlo Ersek *vnet_hdr = tap->vnet_hdr; 5775281d757SMark McLoughlin vnet_hdr_required = *vnet_hdr; 5785281d757SMark McLoughlin } else { 57908c573a8SLaszlo Ersek *vnet_hdr = 1; 5805281d757SMark McLoughlin vnet_hdr_required = 0; 5815281d757SMark McLoughlin } 5825281d757SMark McLoughlin 583264986e2SJason Wang TFR(fd = tap_open(ifname, ifname_sz, vnet_hdr, vnet_hdr_required, 584264986e2SJason Wang mq_required)); 5855281d757SMark McLoughlin if (fd < 0) { 5865281d757SMark McLoughlin return -1; 5875281d757SMark McLoughlin } 5885281d757SMark McLoughlin 5895281d757SMark McLoughlin if (setup_script && 5905281d757SMark McLoughlin setup_script[0] != '\0' && 5915281d757SMark McLoughlin strcmp(setup_script, "no") != 0 && 5925281d757SMark McLoughlin launch_script(setup_script, ifname, fd)) { 5935281d757SMark McLoughlin close(fd); 5945281d757SMark McLoughlin return -1; 5955281d757SMark McLoughlin } 5965281d757SMark McLoughlin 5975281d757SMark McLoughlin return fd; 5985281d757SMark McLoughlin } 5995281d757SMark McLoughlin 600264986e2SJason Wang #define MAX_TAP_QUEUES 1024 601264986e2SJason Wang 6025193e5fbSJason Wang static int net_init_tap_one(const NetdevTapOptions *tap, NetClientState *peer, 6035193e5fbSJason Wang const char *model, const char *name, 6045193e5fbSJason Wang const char *ifname, const char *script, 6055193e5fbSJason Wang const char *downscript, const char *vhostfdname, 6065193e5fbSJason Wang int vnet_hdr, int fd) 6075193e5fbSJason Wang { 6085193e5fbSJason Wang TAPState *s; 60981647a65SNikolay Nikolaev int vhostfd; 6105193e5fbSJason Wang 6115193e5fbSJason Wang s = net_tap_fd_init(peer, model, name, fd, vnet_hdr); 6125193e5fbSJason Wang if (!s) { 6135193e5fbSJason Wang return -1; 6145193e5fbSJason Wang } 6155193e5fbSJason Wang 6165193e5fbSJason Wang if (tap_set_sndbuf(s->fd, tap) < 0) { 6175193e5fbSJason Wang return -1; 6185193e5fbSJason Wang } 6195193e5fbSJason Wang 620264986e2SJason Wang if (tap->has_fd || tap->has_fds) { 6215193e5fbSJason Wang snprintf(s->nc.info_str, sizeof(s->nc.info_str), "fd=%d", fd); 6225193e5fbSJason Wang } else if (tap->has_helper) { 6235193e5fbSJason Wang snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s", 6245193e5fbSJason Wang tap->helper); 6255193e5fbSJason Wang } else { 6265193e5fbSJason Wang snprintf(s->nc.info_str, sizeof(s->nc.info_str), 6275193e5fbSJason Wang "ifname=%s,script=%s,downscript=%s", ifname, script, 6285193e5fbSJason Wang downscript); 6295193e5fbSJason Wang 6305193e5fbSJason Wang if (strcmp(downscript, "no") != 0) { 6315193e5fbSJason Wang snprintf(s->down_script, sizeof(s->down_script), "%s", downscript); 6325193e5fbSJason Wang snprintf(s->down_script_arg, sizeof(s->down_script_arg), 6335193e5fbSJason Wang "%s", ifname); 6345193e5fbSJason Wang } 6355193e5fbSJason Wang } 6365193e5fbSJason Wang 6375193e5fbSJason Wang if (tap->has_vhost ? tap->vhost : 6385193e5fbSJason Wang vhostfdname || (tap->has_vhostforce && tap->vhostforce)) { 63981647a65SNikolay Nikolaev VhostNetOptions options; 64081647a65SNikolay Nikolaev 6411a1bfac9SNikolay Nikolaev options.backend_type = VHOST_BACKEND_TYPE_KERNEL; 64281647a65SNikolay Nikolaev options.net_backend = &s->nc; 64381647a65SNikolay Nikolaev options.force = tap->has_vhostforce && tap->vhostforce; 6445193e5fbSJason Wang 6457873df40SJason Wang if (tap->has_vhostfd || tap->has_vhostfds) { 6465193e5fbSJason Wang vhostfd = monitor_handle_fd_param(cur_mon, vhostfdname); 6475193e5fbSJason Wang if (vhostfd == -1) { 6485193e5fbSJason Wang return -1; 6495193e5fbSJason Wang } 6505193e5fbSJason Wang } else { 65181647a65SNikolay Nikolaev vhostfd = open("/dev/vhost-net", O_RDWR); 65281647a65SNikolay Nikolaev if (vhostfd < 0) { 65381647a65SNikolay Nikolaev error_report("tap: open vhost char device failed: %s", 65481647a65SNikolay Nikolaev strerror(errno)); 65581647a65SNikolay Nikolaev return -1; 6565193e5fbSJason Wang } 65781647a65SNikolay Nikolaev } 65881647a65SNikolay Nikolaev options.opaque = (void *)(uintptr_t)vhostfd; 6595193e5fbSJason Wang 66081647a65SNikolay Nikolaev s->vhost_net = vhost_net_init(&options); 6615193e5fbSJason Wang if (!s->vhost_net) { 6625193e5fbSJason Wang error_report("vhost-net requested but could not be initialized"); 6635193e5fbSJason Wang return -1; 6645193e5fbSJason Wang } 665264986e2SJason Wang } else if (tap->has_vhostfd || tap->has_vhostfds) { 6665193e5fbSJason Wang error_report("vhostfd= is not valid without vhost"); 6675193e5fbSJason Wang return -1; 6685193e5fbSJason Wang } 6695193e5fbSJason Wang 6705193e5fbSJason Wang return 0; 6715193e5fbSJason Wang } 6725193e5fbSJason Wang 673264986e2SJason Wang static int get_fds(char *str, char *fds[], int max) 674264986e2SJason Wang { 675264986e2SJason Wang char *ptr = str, *this; 676264986e2SJason Wang size_t len = strlen(str); 677264986e2SJason Wang int i = 0; 678264986e2SJason Wang 679264986e2SJason Wang while (i < max && ptr < str + len) { 680264986e2SJason Wang this = strchr(ptr, ':'); 681264986e2SJason Wang 682264986e2SJason Wang if (this == NULL) { 683264986e2SJason Wang fds[i] = g_strdup(ptr); 684264986e2SJason Wang } else { 685264986e2SJason Wang fds[i] = g_strndup(ptr, this - ptr); 686264986e2SJason Wang } 687264986e2SJason Wang 688264986e2SJason Wang i++; 689264986e2SJason Wang if (this == NULL) { 690264986e2SJason Wang break; 691264986e2SJason Wang } else { 692264986e2SJason Wang ptr = this + 1; 693264986e2SJason Wang } 694264986e2SJason Wang } 695264986e2SJason Wang 696264986e2SJason Wang return i; 697264986e2SJason Wang } 698264986e2SJason Wang 6991a0c0958SLaszlo Ersek int net_init_tap(const NetClientOptions *opts, const char *name, 7004e68f7a0SStefan Hajnoczi NetClientState *peer) 7015281d757SMark McLoughlin { 70208c573a8SLaszlo Ersek const NetdevTapOptions *tap; 703264986e2SJason Wang int fd, vnet_hdr = 0, i = 0, queues; 70408c573a8SLaszlo Ersek /* for the no-fd, no-helper case */ 70508c573a8SLaszlo Ersek const char *script = NULL; /* suppress wrong "uninit'd use" gcc warning */ 7065193e5fbSJason Wang const char *downscript = NULL; 707264986e2SJason Wang const char *vhostfdname; 70808c573a8SLaszlo Ersek char ifname[128]; 70908c573a8SLaszlo Ersek 71008c573a8SLaszlo Ersek assert(opts->kind == NET_CLIENT_OPTIONS_KIND_TAP); 71108c573a8SLaszlo Ersek tap = opts->tap; 712264986e2SJason Wang queues = tap->has_queues ? tap->queues : 1; 713264986e2SJason Wang vhostfdname = tap->has_vhostfd ? tap->vhostfd : NULL; 71408c573a8SLaszlo Ersek 715ce675a75SJason Wang /* QEMU vlans does not support multiqueue tap, in this case peer is set. 716ce675a75SJason Wang * For -netdev, peer is always NULL. */ 717ce675a75SJason Wang if (peer && (tap->has_queues || tap->has_fds || tap->has_vhostfds)) { 718085d8134SPeter Maydell error_report("Multiqueue tap cannot be used with QEMU vlans"); 719ce675a75SJason Wang return -1; 720ce675a75SJason Wang } 721ce675a75SJason Wang 72208c573a8SLaszlo Ersek if (tap->has_fd) { 72308c573a8SLaszlo Ersek if (tap->has_ifname || tap->has_script || tap->has_downscript || 724264986e2SJason Wang tap->has_vnet_hdr || tap->has_helper || tap->has_queues || 725c87826a8SJason Wang tap->has_fds || tap->has_vhostfds) { 726a7c36ee4SCorey Bryant error_report("ifname=, script=, downscript=, vnet_hdr=, " 727c87826a8SJason Wang "helper=, queues=, fds=, and vhostfds= " 728c87826a8SJason Wang "are invalid with fd="); 7295281d757SMark McLoughlin return -1; 7305281d757SMark McLoughlin } 7315281d757SMark McLoughlin 732a96ed02fSNicholas Bellinger fd = monitor_handle_fd_param(cur_mon, tap->fd); 7335281d757SMark McLoughlin if (fd == -1) { 7345281d757SMark McLoughlin return -1; 7355281d757SMark McLoughlin } 7365281d757SMark McLoughlin 7375281d757SMark McLoughlin fcntl(fd, F_SETFL, O_NONBLOCK); 7385281d757SMark McLoughlin 7395281d757SMark McLoughlin vnet_hdr = tap_probe_vnet_hdr(fd); 740a7c36ee4SCorey Bryant 74102cd8090SAnthony Liguori if (net_init_tap_one(tap, peer, "tap", name, NULL, 742264986e2SJason Wang script, downscript, 743264986e2SJason Wang vhostfdname, vnet_hdr, fd)) { 744264986e2SJason Wang return -1; 745264986e2SJason Wang } 746264986e2SJason Wang } else if (tap->has_fds) { 747264986e2SJason Wang char *fds[MAX_TAP_QUEUES]; 748264986e2SJason Wang char *vhost_fds[MAX_TAP_QUEUES]; 749264986e2SJason Wang int nfds, nvhosts; 750a7c36ee4SCorey Bryant 751264986e2SJason Wang if (tap->has_ifname || tap->has_script || tap->has_downscript || 752264986e2SJason Wang tap->has_vnet_hdr || tap->has_helper || tap->has_queues || 753c87826a8SJason Wang tap->has_vhostfd) { 754264986e2SJason Wang error_report("ifname=, script=, downscript=, vnet_hdr=, " 755c87826a8SJason Wang "helper=, queues=, and vhostfd= " 756c87826a8SJason Wang "are invalid with fds="); 757264986e2SJason Wang return -1; 758264986e2SJason Wang } 759264986e2SJason Wang 760264986e2SJason Wang nfds = get_fds(tap->fds, fds, MAX_TAP_QUEUES); 761264986e2SJason Wang if (tap->has_vhostfds) { 762264986e2SJason Wang nvhosts = get_fds(tap->vhostfds, vhost_fds, MAX_TAP_QUEUES); 763264986e2SJason Wang if (nfds != nvhosts) { 764264986e2SJason Wang error_report("The number of fds passed does not match the " 765264986e2SJason Wang "number of vhostfds passed"); 766264986e2SJason Wang return -1; 767264986e2SJason Wang } 768264986e2SJason Wang } 769264986e2SJason Wang 770264986e2SJason Wang for (i = 0; i < nfds; i++) { 771264986e2SJason Wang fd = monitor_handle_fd_param(cur_mon, fds[i]); 772264986e2SJason Wang if (fd == -1) { 773264986e2SJason Wang return -1; 774264986e2SJason Wang } 775264986e2SJason Wang 776264986e2SJason Wang fcntl(fd, F_SETFL, O_NONBLOCK); 777264986e2SJason Wang 778264986e2SJason Wang if (i == 0) { 779264986e2SJason Wang vnet_hdr = tap_probe_vnet_hdr(fd); 780264986e2SJason Wang } else if (vnet_hdr != tap_probe_vnet_hdr(fd)) { 781264986e2SJason Wang error_report("vnet_hdr not consistent across given tap fds"); 782264986e2SJason Wang return -1; 783264986e2SJason Wang } 784264986e2SJason Wang 785264986e2SJason Wang if (net_init_tap_one(tap, peer, "tap", name, ifname, 786264986e2SJason Wang script, downscript, 787264986e2SJason Wang tap->has_vhostfds ? vhost_fds[i] : NULL, 788264986e2SJason Wang vnet_hdr, fd)) { 789264986e2SJason Wang return -1; 790264986e2SJason Wang } 791264986e2SJason Wang } 79208c573a8SLaszlo Ersek } else if (tap->has_helper) { 79308c573a8SLaszlo Ersek if (tap->has_ifname || tap->has_script || tap->has_downscript || 794c87826a8SJason Wang tap->has_vnet_hdr || tap->has_queues || tap->has_vhostfds) { 795a7c36ee4SCorey Bryant error_report("ifname=, script=, downscript=, and vnet_hdr= " 796c87826a8SJason Wang "queues=, and vhostfds= are invalid with helper="); 797a7c36ee4SCorey Bryant return -1; 798a7c36ee4SCorey Bryant } 799a7c36ee4SCorey Bryant 80008c573a8SLaszlo Ersek fd = net_bridge_run_helper(tap->helper, DEFAULT_BRIDGE_INTERFACE); 801a7c36ee4SCorey Bryant if (fd == -1) { 802a7c36ee4SCorey Bryant return -1; 803a7c36ee4SCorey Bryant } 804a7c36ee4SCorey Bryant 805a7c36ee4SCorey Bryant fcntl(fd, F_SETFL, O_NONBLOCK); 806a7c36ee4SCorey Bryant vnet_hdr = tap_probe_vnet_hdr(fd); 807a7c36ee4SCorey Bryant 808264986e2SJason Wang if (net_init_tap_one(tap, peer, "bridge", name, ifname, 809264986e2SJason Wang script, downscript, vhostfdname, 810264986e2SJason Wang vnet_hdr, fd)) { 81184f8f3daSGonglei close(fd); 812264986e2SJason Wang return -1; 813264986e2SJason Wang } 8145281d757SMark McLoughlin } else { 815c87826a8SJason Wang if (tap->has_vhostfds) { 816c87826a8SJason Wang error_report("vhostfds= is invalid if fds= wasn't specified"); 817c87826a8SJason Wang return -1; 818c87826a8SJason Wang } 81908c573a8SLaszlo Ersek script = tap->has_script ? tap->script : DEFAULT_NETWORK_SCRIPT; 8205193e5fbSJason Wang downscript = tap->has_downscript ? tap->downscript : 8215193e5fbSJason Wang DEFAULT_NETWORK_DOWN_SCRIPT; 822264986e2SJason Wang 823264986e2SJason Wang if (tap->has_ifname) { 824264986e2SJason Wang pstrcpy(ifname, sizeof ifname, tap->ifname); 825264986e2SJason Wang } else { 826264986e2SJason Wang ifname[0] = '\0'; 827264986e2SJason Wang } 828264986e2SJason Wang 829264986e2SJason Wang for (i = 0; i < queues; i++) { 830264986e2SJason Wang fd = net_tap_init(tap, &vnet_hdr, i >= 1 ? "no" : script, 831264986e2SJason Wang ifname, sizeof ifname, queues > 1); 832929fe497SJuergen Lock if (fd == -1) { 833929fe497SJuergen Lock return -1; 834929fe497SJuergen Lock } 835a7c36ee4SCorey Bryant 836264986e2SJason Wang if (queues > 1 && i == 0 && !tap->has_ifname) { 837264986e2SJason Wang if (tap_fd_get_ifname(fd, ifname)) { 838264986e2SJason Wang error_report("Fail to get ifname"); 83984f8f3daSGonglei close(fd); 840264986e2SJason Wang return -1; 841264986e2SJason Wang } 8425281d757SMark McLoughlin } 8435281d757SMark McLoughlin 844264986e2SJason Wang if (net_init_tap_one(tap, peer, "tap", name, ifname, 845264986e2SJason Wang i >= 1 ? "no" : script, 846264986e2SJason Wang i >= 1 ? "no" : downscript, 847264986e2SJason Wang vhostfdname, vnet_hdr, fd)) { 84884f8f3daSGonglei close(fd); 849264986e2SJason Wang return -1; 850264986e2SJason Wang } 851264986e2SJason Wang } 852264986e2SJason Wang } 853264986e2SJason Wang 854264986e2SJason Wang return 0; 8555281d757SMark McLoughlin } 856b202554cSMichael S. Tsirkin 8574e68f7a0SStefan Hajnoczi VHostNetState *tap_get_vhost_net(NetClientState *nc) 858b202554cSMichael S. Tsirkin { 859b202554cSMichael S. Tsirkin TAPState *s = DO_UPCAST(TAPState, nc, nc); 8602be64a68SLaszlo Ersek assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP); 861b202554cSMichael S. Tsirkin return s->vhost_net; 862b202554cSMichael S. Tsirkin } 86316dbaf90SJason Wang 86416dbaf90SJason Wang int tap_enable(NetClientState *nc) 86516dbaf90SJason Wang { 86616dbaf90SJason Wang TAPState *s = DO_UPCAST(TAPState, nc, nc); 86716dbaf90SJason Wang int ret; 86816dbaf90SJason Wang 86916dbaf90SJason Wang if (s->enabled) { 87016dbaf90SJason Wang return 0; 87116dbaf90SJason Wang } else { 87216dbaf90SJason Wang ret = tap_fd_enable(s->fd); 87316dbaf90SJason Wang if (ret == 0) { 87416dbaf90SJason Wang s->enabled = true; 87516dbaf90SJason Wang tap_update_fd_handler(s); 87616dbaf90SJason Wang } 87716dbaf90SJason Wang return ret; 87816dbaf90SJason Wang } 87916dbaf90SJason Wang } 88016dbaf90SJason Wang 88116dbaf90SJason Wang int tap_disable(NetClientState *nc) 88216dbaf90SJason Wang { 88316dbaf90SJason Wang TAPState *s = DO_UPCAST(TAPState, nc, nc); 88416dbaf90SJason Wang int ret; 88516dbaf90SJason Wang 88616dbaf90SJason Wang if (s->enabled == 0) { 88716dbaf90SJason Wang return 0; 88816dbaf90SJason Wang } else { 88916dbaf90SJason Wang ret = tap_fd_disable(s->fd); 89016dbaf90SJason Wang if (ret == 0) { 89116dbaf90SJason Wang qemu_purge_queued_packets(nc); 89216dbaf90SJason Wang s->enabled = false; 89316dbaf90SJason Wang tap_update_fd_handler(s); 89416dbaf90SJason Wang } 89516dbaf90SJason Wang return ret; 89616dbaf90SJason Wang } 89716dbaf90SJason Wang } 898