net.c

   1 /*
   2  * QEMU System Emulator
   3  *
   4  * Copyright (c) 2003-2008 Fabrice Bellard
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24 #include "qemu-common.h"
  25 #include "net.h"
  26 #include "console.h"
  27 #include "sysemu.h"
  28 #include "qemu-timer.h"
  29 #include "qemu-char.h"
  30 #include "audio/audio.h"
  31
  32 #include <unistd.h>
  33 #include <fcntl.h>
  34 #include <signal.h>
  35 #include <time.h>
  36 #include <errno.h>
  37 #include <sys/time.h>
  38 #include <zlib.h>
  39
  40 #ifndef _WIN32
  41 #include <sys/times.h>
  42 #include <sys/wait.h>
  43 #include <termios.h>
  44 #include <sys/mman.h>
  45 #include <sys/ioctl.h>
  46 #include <sys/resource.h>
  47 #include <sys/socket.h>
  48 #include <netinet/in.h>
  49 #include <net/if.h>
  50 #ifdef __NetBSD__
  51 #include <net/if_tap.h>
  52 #endif
  53 #ifdef __linux__
  54 #include <linux/if_tun.h>
  55 #endif
  56 #include <arpa/inet.h>
  57 #include <dirent.h>
  58 #include <netdb.h>
  59 #include <sys/select.h>
  60 #ifdef _BSD
  61 #include <sys/stat.h>
  62 #ifdef __FreeBSD__
  63 #include <libutil.h>
  64 #else
  65 #include <util.h>
  66 #endif
  67 #elif defined (__GLIBC__) && defined (__FreeBSD_kernel__)
  68 #include <freebsd/stdlib.h>
  69 #else
  70 #ifdef __linux__
  71 #include <pty.h>
  72 #include <malloc.h>
  73 #include <linux/rtc.h>
  74
  75 /* For the benefit of older linux systems which don't supply it,
  76    we use a local copy of hpet.h. */
  77 /* #include <linux/hpet.h> */
  78 #include "hpet.h"
  79
  80 #include <linux/ppdev.h>
  81 #include <linux/parport.h>
  82 #endif
  83 #ifdef __sun__
  84 #include <sys/stat.h>
  85 #include <sys/ethernet.h>
  86 #include <sys/sockio.h>
  87 #include <netinet/arp.h>
  88 #include <netinet/in.h>
  89 #include <netinet/in_systm.h>
  90 #include <netinet/ip.h>
  91 #include <netinet/ip_icmp.h> // must come after ip.h
  92 #include <netinet/udp.h>
  93 #include <netinet/tcp.h>
  94 #include <net/if.h>
  95 #include <syslog.h>
  96 #include <stropts.h>
  97 #endif
  98 #endif
  99 #endif
 100
 101 #include "qemu_socket.h"
 102
 103 #if defined(CONFIG_SLIRP)
 104 #include "libslirp.h"
 105 #endif
 106
 107 #if defined(__OpenBSD__)
 108 #include <util.h>
 109 #endif
 110
 111 #if defined(CONFIG_VDE)
 112 #include <libvdeplug.h>
 113 #endif
 114
 115 #ifdef _WIN32
 116 #include <malloc.h>
 117 #include <sys/timeb.h>
 118 #include <mmsystem.h>
 119 #define getopt_long_only getopt_long
 120 #define memalign(align, size) malloc(size)
 121 #endif
 122
 123 // FIXME: #include "qemu-kvm.h"
 124
 125 static VLANState *first_vlan;
 126
 127 /***********************************************************/
 128 /* network device redirectors */
 129
 130 #if defined(DEBUG_NET) || defined(DEBUG_SLIRP)
 131 static void hex_dump(FILE *f, const uint8_t *buf, int size)
 132 {
 133     int len, i, j, c;
 134
 135     for(i=0;i<size;i+=16) {
 136         len = size - i;
 137         if (len > 16)
 138             len = 16;
 139         fprintf(f, "%08x ", i);
 140         for(j=0;j<16;j++) {
 141             if (j < len)
 142                 fprintf(f, " %02x", buf[i+j]);
 143             else
 144                 fprintf(f, "   ");
 145         }
 146         fprintf(f, " ");
 147         for(j=0;j<len;j++) {
 148             c = buf[i+j];
 149             if (c < ' ' || c > '~')
 150                 c = '.';
 151             fprintf(f, "%c", c);
 152         }
 153         fprintf(f, "\n");
 154     }
 155 }
 156 #endif
 157
 158 static int parse_macaddr(uint8_t *macaddr, const char *p)
 159 {
 160     int i;
 161     char *last_char;
 162     long int offset;
 163
 164     errno = 0;
 165     offset = strtol(p, &last_char, 0);
 166     if (0 == errno && '\0' == *last_char &&
 167             offset >= 0 && offset <= 0xFFFFFF) {
 168         macaddr[3] = (offset & 0xFF0000) >> 16;
 169         macaddr[4] = (offset & 0xFF00) >> 8;
 170         macaddr[5] = offset & 0xFF;
 171         return 0;
 172     } else {
 173         for(i = 0; i < 6; i++) {
 174             macaddr[i] = strtol(p, (char **)&p, 16);
 175             if (i == 5) {
 176                 if (*p != '\0')
 177                     return -1;
 178             } else {
 179                 if (*p != ':' && *p != '-')
 180                     return -1;
 181                 p++;
 182             }
 183         }
 184         return 0;
 185     }
 186
 187     return -1;
 188 }
 189
 190 static int get_str_sep(char *buf, int buf_size, const char **pp, int sep)
 191 {
 192     const char *p, *p1;
 193     int len;
 194     p = *pp;
 195     p1 = strchr(p, sep);
 196     if (!p1)
 197         return -1;
 198     len = p1 - p;
 199     p1++;
 200     if (buf_size > 0) {
 201         if (len > buf_size - 1)
 202             len = buf_size - 1;
 203         memcpy(buf, p, len);
 204         buf[len] = '\0';
 205     }
 206     *pp = p1;
 207     return 0;
 208 }
 209
 210 int parse_host_src_port(struct sockaddr_in *haddr,
 211                         struct sockaddr_in *saddr,
 212                         const char *input_str)
 213 {
 214     char *str = strdup(input_str);
 215     char *host_str = str;
 216     char *src_str;
 217     const char *src_str2;
 218     char *ptr;
 219
 220     /*
 221      * Chop off any extra arguments at the end of the string which
 222      * would start with a comma, then fill in the src port information
 223      * if it was provided else use the "any address" and "any port".
 224      */
 225     if ((ptr = strchr(str,',')))
 226         *ptr = '\0';
 227
 228     if ((src_str = strchr(input_str,'@'))) {
 229         *src_str = '\0';
 230         src_str++;
 231     }
 232
 233     if (parse_host_port(haddr, host_str) < 0)
 234         goto fail;
 235
 236     src_str2 = src_str;
 237     if (!src_str || *src_str == '\0')
 238         src_str2 = ":0";
 239
 240     if (parse_host_port(saddr, src_str2) < 0)
 241         goto fail;
 242
 243     free(str);
 244     return(0);
 245
 246 fail:
 247     free(str);
 248     return -1;
 249 }
 250
 251 int parse_host_port(struct sockaddr_in *saddr, const char *str)
 252 {
 253     char buf[512];
 254     struct hostent *he;
 255     const char *p, *r;
 256     int port;
 257
 258     p = str;
 259     if (get_str_sep(buf, sizeof(buf), &p, ':') < 0)
 260         return -1;
 261     saddr->sin_family = AF_INET;
 262     if (buf[0] == '\0') {
 263         saddr->sin_addr.s_addr = 0;
 264     } else {
 265         if (qemu_isdigit(buf[0])) {
 266             if (!inet_aton(buf, &saddr->sin_addr))
 267                 return -1;
 268         } else {
 269             if ((he = gethostbyname(buf)) == NULL)
 270                 return - 1;
 271             saddr->sin_addr = *(struct in_addr *)he->h_addr;
 272         }
 273     }
 274     port = strtol(p, (char **)&r, 0);
 275     if (r == p)
 276         return -1;
 277     saddr->sin_port = htons(port);
 278     return 0;
 279 }
 280
 281 #if !defined(_WIN32) && 0
 282 static int parse_unix_path(struct sockaddr_un *uaddr, const char *str)
 283 {
 284     const char *p;
 285     int len;
 286
 287     len = MIN(108, strlen(str));
 288     p = strchr(str, ',');
 289     if (p)
 290         len = MIN(len, p - str);
 291
 292     memset(uaddr, 0, sizeof(*uaddr));
 293
 294     uaddr->sun_family = AF_UNIX;
 295     memcpy(uaddr->sun_path, str, len);
 296
 297     return 0;
 298 }
 299 #endif
 300
 301 void qemu_format_nic_info_str(VLANClientState *vc, uint8_t macaddr[6])
 302 {
 303     snprintf(vc->info_str, sizeof(vc->info_str),
 304              "model=%s,macaddr=%02x:%02x:%02x:%02x:%02x:%02x",
 305              vc->model,
 306              macaddr[0], macaddr[1], macaddr[2],
 307              macaddr[3], macaddr[4], macaddr[5]);
 308 }
 309
 310 static char *assign_name(VLANClientState *vc1, const char *model)
 311 {
 312     VLANState *vlan;
 313     char buf[256];
 314     int id = 0;
 315
 316     for (vlan = first_vlan; vlan; vlan = vlan->next) {
 317         VLANClientState *vc;
 318
 319         for (vc = vlan->first_client; vc; vc = vc->next)
 320             if (vc != vc1 && strcmp(vc->model, model) == 0)
 321                 id++;
 322     }
 323
 324     snprintf(buf, sizeof(buf), "%s.%d", model, id);
 325
 326     return strdup(buf);
 327 }
 328
 329 VLANClientState *qemu_new_vlan_client(VLANState *vlan,
 330                                       const char *model,
 331                                       const char *name,
 332                                       IOReadHandler *fd_read,
 333                                       IOCanRWHandler *fd_can_read,
 334                                       void *opaque)
 335 {
 336     VLANClientState *vc, **pvc;
 337     vc = qemu_mallocz(sizeof(VLANClientState));
 338     vc->model = strdup(model);
 339     if (name)
 340         vc->name = strdup(name);
 341     else
 342         vc->name = assign_name(vc, model);
 343     vc->fd_read = fd_read;
 344     vc->fd_can_read = fd_can_read;
 345     vc->opaque = opaque;
 346     vc->vlan = vlan;
 347
 348     vc->next = NULL;
 349     pvc = &vlan->first_client;
 350     while (*pvc != NULL)
 351         pvc = &(*pvc)->next;
 352     *pvc = vc;
 353     return vc;
 354 }
 355
 356 void qemu_del_vlan_client(VLANClientState *vc)
 357 {
 358     VLANClientState **pvc = &vc->vlan->first_client;
 359
 360     while (*pvc != NULL)
 361         if (*pvc == vc) {
 362             *pvc = vc->next;
 363             free(vc->name);
 364             free(vc->model);
 365             free(vc);
 366             break;
 367         } else
 368             pvc = &(*pvc)->next;
 369 }
 370
 371 VLANClientState *qemu_find_vlan_client(VLANState *vlan, void *opaque)
 372 {
 373     VLANClientState **pvc = &vlan->first_client;
 374
 375     while (*pvc != NULL)
 376         if ((*pvc)->opaque == opaque)
 377             return *pvc;
 378         else
 379             pvc = &(*pvc)->next;
 380
 381     return NULL;
 382 }
 383
 384 int qemu_can_send_packet(VLANClientState *vc1)
 385 {
 386     VLANState *vlan = vc1->vlan;
 387     VLANClientState *vc;
 388
 389     for(vc = vlan->first_client; vc != NULL; vc = vc->next) {
 390         if (vc != vc1) {
 391             if (vc->fd_can_read && vc->fd_can_read(vc->opaque))
 392                 return 1;
 393         }
 394     }
 395     return 0;
 396 }
 397
 398 int qemu_send_packet(VLANClientState *vc1, const uint8_t *buf, int size)
 399 {
 400     VLANState *vlan = vc1->vlan;
 401     VLANClientState *vc;
 402     int ret = -EAGAIN;
 403
 404     if (vc1->link_down)
 405         return 0;
 406
 407 #ifdef DEBUG_NET
 408     printf("vlan %d send:\n", vlan->id);
 409     hex_dump(stdout, buf, size);
 410 #endif
 411     for(vc = vlan->first_client; vc != NULL; vc = vc->next) {
 412         if (vc != vc1 && !vc->link_down) {
 413             if (!vc->fd_can_read || vc->fd_can_read(vc->opaque)) {
 414                 vc->fd_read(vc->opaque, buf, size);
 415                 ret = 0;
 416             }
 417         }
 418     }
 419     return ret;
 420 }
 421
 422 static ssize_t vc_sendv_compat(VLANClientState *vc, const struct iovec *iov,
 423                                int iovcnt)
 424 {
 425     uint8_t buffer[4096];
 426     size_t offset = 0;
 427     int i;
 428
 429     for (i = 0; i < iovcnt; i++) {
 430         size_t len;
 431
 432         len = MIN(sizeof(buffer) - offset, iov[i].iov_len);
 433         memcpy(buffer + offset, iov[i].iov_base, len);
 434         offset += len;
 435     }
 436
 437     vc->fd_read(vc->opaque, buffer, offset);
 438
 439     return offset;
 440 }
 441
 442 static ssize_t calc_iov_length(const struct iovec *iov, int iovcnt)
 443 {
 444     size_t offset = 0;
 445     int i;
 446
 447     for (i = 0; i < iovcnt; i++)
 448         offset += iov[i].iov_len;
 449     return offset;
 450 }
 451
 452 ssize_t qemu_sendv_packet(VLANClientState *vc1, const struct iovec *iov,
 453                           int iovcnt)
 454 {
 455     VLANState *vlan = vc1->vlan;
 456     VLANClientState *vc;
 457     ssize_t max_len = 0;
 458
 459     if (vc1->link_down)
 460         return calc_iov_length(iov, iovcnt);
 461
 462     for (vc = vlan->first_client; vc != NULL; vc = vc->next) {
 463         ssize_t len = 0;
 464
 465         if (vc == vc1)
 466             continue;
 467
 468         if (vc->link_down)
 469             len = calc_iov_length(iov, iovcnt);
 470         if (vc->fd_readv)
 471             len = vc->fd_readv(vc->opaque, iov, iovcnt);
 472         else if (vc->fd_read)
 473             len = vc_sendv_compat(vc, iov, iovcnt);
 474
 475         max_len = MAX(max_len, len);
 476     }
 477
 478     return max_len;
 479 }
 480
 481 #if defined(CONFIG_SLIRP)
 482
 483 /* slirp network adapter */
 484
 485 static int slirp_inited;
 486 static int slirp_restrict;
 487 static char *slirp_ip;
 488 static VLANClientState *slirp_vc;
 489
 490 int slirp_can_output(void)
 491 {
 492     return !slirp_vc || qemu_can_send_packet(slirp_vc);
 493 }
 494
 495 void slirp_output(const uint8_t *pkt, int pkt_len)
 496 {
 497 #ifdef DEBUG_SLIRP
 498     printf("slirp output:\n");
 499     hex_dump(stdout, pkt, pkt_len);
 500 #endif
 501     if (!slirp_vc)
 502         return;
 503     qemu_send_packet(slirp_vc, pkt, pkt_len);
 504 }
 505
 506 int slirp_is_inited(void)
 507 {
 508     return slirp_inited;
 509 }
 510
 511 static void slirp_receive(void *opaque, const uint8_t *buf, int size)
 512 {
 513 #ifdef DEBUG_SLIRP
 514     printf("slirp input:\n");
 515     hex_dump(stdout, buf, size);
 516 #endif
 517     slirp_input(buf, size);
 518 }
 519
 520 static int net_slirp_init(VLANState *vlan, const char *model, const char *name)
 521 {
 522     if (!slirp_inited) {
 523         slirp_inited = 1;
 524         slirp_init(slirp_restrict, slirp_ip);
 525     }
 526     slirp_vc = qemu_new_vlan_client(vlan, model, name,
 527                                     slirp_receive, NULL, NULL);
 528     slirp_vc->info_str[0] = '\0';
 529     return 0;
 530 }
 531
 532 void net_slirp_redir(const char *redir_str)
 533 {
 534     int is_udp;
 535     char buf[256], *r;
 536     const char *p;
 537     struct in_addr guest_addr;
 538     int host_port, guest_port;
 539
 540     if (!slirp_inited) {
 541         slirp_inited = 1;
 542         slirp_init(slirp_restrict, slirp_ip);
 543     }
 544
 545     p = redir_str;
 546     if (get_str_sep(buf, sizeof(buf), &p, ':') < 0)
 547         goto fail;
 548     if (!strcmp(buf, "tcp")) {
 549         is_udp = 0;
 550     } else if (!strcmp(buf, "udp")) {
 551         is_udp = 1;
 552     } else {
 553         goto fail;
 554     }
 555
 556     if (get_str_sep(buf, sizeof(buf), &p, ':') < 0)
 557         goto fail;
 558     host_port = strtol(buf, &r, 0);
 559     if (r == buf)
 560         goto fail;
 561
 562     if (get_str_sep(buf, sizeof(buf), &p, ':') < 0)
 563         goto fail;
 564     if (buf[0] == '\0') {
 565         pstrcpy(buf, sizeof(buf), "10.0.2.15");
 566     }
 567     if (!inet_aton(buf, &guest_addr))
 568         goto fail;
 569
 570     guest_port = strtol(p, &r, 0);
 571     if (r == p)
 572         goto fail;
 573
 574     if (slirp_redir(is_udp, host_port, guest_addr, guest_port) < 0) {
 575         fprintf(stderr, "qemu: could not set up redirection\n");
 576         exit(1);
 577     }
 578     return;
 579  fail:
 580     fprintf(stderr, "qemu: syntax: -redir [tcp|udp]:host-port:[guest-host]:guest-port\n");
 581     exit(1);
 582 }
 583
 584 #ifndef _WIN32
 585
 586 static char smb_dir[1024];
 587
 588 static void erase_dir(char *dir_name)
 589 {
 590     DIR *d;
 591     struct dirent *de;
 592     char filename[1024];
 593
 594     /* erase all the files in the directory */
 595     if ((d = opendir(dir_name)) != 0) {
 596         for(;;) {
 597             de = readdir(d);
 598             if (!de)
 599                 break;
 600             if (strcmp(de->d_name, ".") != 0 &&
 601                 strcmp(de->d_name, "..") != 0) {
 602                 snprintf(filename, sizeof(filename), "%s/%s",
 603                          smb_dir, de->d_name);
 604                 if (unlink(filename) != 0)  /* is it a directory? */
 605                     erase_dir(filename);
 606             }
 607         }
 608         closedir(d);
 609         rmdir(dir_name);
 610     }
 611 }
 612
 613 /* automatic user mode samba server configuration */
 614 static void smb_exit(void)
 615 {
 616     erase_dir(smb_dir);
 617 }
 618
 619 /* automatic user mode samba server configuration */
 620 void net_slirp_smb(const char *exported_dir)
 621 {
 622     char smb_conf[1024];
 623     char smb_cmdline[1024];
 624     FILE *f;
 625
 626     if (!slirp_inited) {
 627         slirp_inited = 1;
 628         slirp_init(slirp_restrict, slirp_ip);
 629     }
 630
 631     /* XXX: better tmp dir construction */
 632     snprintf(smb_dir, sizeof(smb_dir), "/tmp/qemu-smb.%d", getpid());
 633     if (mkdir(smb_dir, 0700) < 0) {
 634         fprintf(stderr, "qemu: could not create samba server dir '%s'\n", smb_dir);
 635         exit(1);
 636     }
 637     snprintf(smb_conf, sizeof(smb_conf), "%s/%s", smb_dir, "smb.conf");
 638
 639     f = fopen(smb_conf, "w");
 640     if (!f) {
 641         fprintf(stderr, "qemu: could not create samba server configuration file '%s'\n", smb_conf);
 642         exit(1);
 643     }
 644     fprintf(f,
 645             "[global]\n"
 646             "private dir=%s\n"
 647             "smb ports=0\n"
 648             "socket address=127.0.0.1\n"
 649             "pid directory=%s\n"
 650             "lock directory=%s\n"
 651             "log file=%s/log.smbd\n"
 652             "smb passwd file=%s/smbpasswd\n"
 653             "security = share\n"
 654             "[qemu]\n"
 655             "path=%s\n"
 656             "read only=no\n"
 657             "guest ok=yes\n",
 658             smb_dir,
 659             smb_dir,
 660             smb_dir,
 661             smb_dir,
 662             smb_dir,
 663             exported_dir
 664             );
 665     fclose(f);
 666     atexit(smb_exit);
 667
 668     snprintf(smb_cmdline, sizeof(smb_cmdline), "%s -s %s",
 669              SMBD_COMMAND, smb_conf);
 670
 671     slirp_add_exec(0, smb_cmdline, 4, 139);
 672 }
 673
 674 #endif /* !defined(_WIN32) */
 675 void do_info_slirp(void)
 676 {
 677     slirp_stats();
 678 }
 679
 680 struct VMChannel {
 681     CharDriverState *hd;
 682     int port;
 683 } *vmchannels;
 684
 685 static int vmchannel_can_read(void *opaque)
 686 {
 687     struct VMChannel *vmc = (struct VMChannel*)opaque;
 688     return slirp_socket_can_recv(4, vmc->port);
 689 }
 690
 691 static void vmchannel_read(void *opaque, const uint8_t *buf, int size)
 692 {
 693     struct VMChannel *vmc = (struct VMChannel*)opaque;
 694     slirp_socket_recv(4, vmc->port, buf, size);
 695 }
 696
 697 #endif /* CONFIG_SLIRP */
 698
 699 #ifdef _WIN32
 700
 701 int tap_has_vnet_hdr(void *opaque)
 702 {
 703     return 0;
 704 }
 705
 706 void tap_using_vnet_hdr(void *opaque, int using_vnet_hdr)
 707 {
 708 }
 709
 710 #else /* !defined(_WIN32) */
 711
 712 /* Maximum GSO packet size (64k) plus plenty of room for
 713  * the ethernet and virtio_net headers
 714  */
 715 #define TAP_BUFSIZE (4096 + 65536)
 716
 717 #ifdef IFF_VNET_HDR
 718 #include <linux/virtio_net.h>
 719 #endif
 720
 721 typedef struct TAPState {
 722     VLANClientState *vc;
 723     int fd;
 724     char down_script[1024];
 725     char down_script_arg[128];
 726     char buf[TAP_BUFSIZE];
 727     int size;
 728     unsigned int has_vnet_hdr : 1;
 729     unsigned int using_vnet_hdr : 1;
 730 } TAPState;
 731
 732 #ifdef HAVE_IOVEC
 733 static ssize_t tap_receive_iov(void *opaque, const struct iovec *iov,
 734                                int iovcnt)
 735 {
 736     TAPState *s = opaque;
 737     ssize_t len;
 738
 739     do {
 740         len = writev(s->fd, iov, iovcnt);
 741     } while (len == -1 && (errno == EINTR || errno == EAGAIN));
 742
 743     return len;
 744 }
 745 #endif
 746
 747 static void tap_receive(void *opaque, const uint8_t *buf, int size)
 748 {
 749     struct iovec iov[2];
 750     int i = 0;
 751
 752 #ifdef IFF_VNET_HDR
 753     TAPState *s = opaque;
 754     struct virtio_net_hdr hdr = { 0, };
 755
 756     if (s->has_vnet_hdr && !s->using_vnet_hdr) {
 757         iov[i].iov_base = &hdr;
 758         iov[i].iov_len  = sizeof(hdr);
 759         i++;
 760     }
 761 #endif
 762
 763     iov[i].iov_base = (char *) buf;
 764     iov[i].iov_len  = size;
 765     i++;
 766
 767     tap_receive_iov(opaque, iov, i);
 768 }
 769
 770 static int tap_can_send(void *opaque)
 771 {
 772     TAPState *s = opaque;
 773     VLANClientState *vc;
 774     int can_receive = 0;
 775
 776     /* Check to see if any of our clients can receive a packet */
 777     for (vc = s->vc->vlan->first_client; vc; vc = vc->next) {
 778         /* Skip ourselves */
 779         if (vc == s->vc)
 780             continue;
 781
 782         if (!vc->fd_can_read) {
 783             /* no fd_can_read handler, they always can receive */
 784             can_receive = 1;
 785         } else
 786             can_receive = vc->fd_can_read(vc->opaque);
 787
 788         /* Once someone can receive, we try to send a packet */
 789         if (can_receive)
 790             break;
 791     }
 792
 793     return can_receive;
 794 }
 795
 796 static int tap_send_packet(TAPState *s)
 797 {
 798     uint8_t *buf = (uint8_t *)s->buf;
 799     int size = s->size;
 800
 801 #ifdef IFF_VNET_HDR
 802     if (s->has_vnet_hdr && !s->using_vnet_hdr) {
 803         buf += sizeof(struct virtio_net_hdr);
 804         size -= sizeof(struct virtio_net_hdr);
 805     }
 806 #endif
 807
 808     return qemu_send_packet(s->vc, buf, size);
 809 }
 810
 811 static void tap_send(void *opaque)
 812 {
 813     TAPState *s = opaque;
 814
 815     /* First try to send any buffered packet */
 816     if (s->size > 0) {
 817         int err;
 818
 819         /* If noone can receive the packet, buffer it */
 820         err = tap_send_packet(s);
 821         if (err == -EAGAIN)
 822             return;
 823     }
 824
 825     /* Read packets until we hit EAGAIN */
 826     do {
 827 #ifdef __sun__
 828         struct strbuf sbuf;
 829         int f = 0;
 830         sbuf.maxlen = sizeof(s->buf);
 831         sbuf.buf = s->buf;
 832         s->size = getmsg(s->fd, NULL, &sbuf, &f) >=0 ? sbuf.len : -1;
 833 #else
 834         // FIXME: kvm_sleep_begin();
 835         s->size = read(s->fd, s->buf, sizeof(s->buf));
 836         // FIXME: kvm_sleep_end();
 837 #endif
 838
 839         if (s->size == -1 && errno == EINTR)
 840             continue;
 841
 842         if (s->size > 0) {
 843             int err;
 844
 845             /* If noone can receive the packet, buffer it */
 846             err = tap_send_packet(s);
 847             if (err == -EAGAIN)
 848                 break;
 849         }
 850     } while (s->size > 0);
 851 }
 852
 853 int tap_has_vnet_hdr(void *opaque)
 854 {
 855     VLANClientState *vc = opaque;
 856     TAPState *s = vc->opaque;
 857
 858     return s ? s->has_vnet_hdr : 0;
 859 }
 860
 861 void tap_using_vnet_hdr(void *opaque, int using_vnet_hdr)
 862 {
 863     VLANClientState *vc = opaque;
 864     TAPState *s = vc->opaque;
 865
 866     if (!s || !s->has_vnet_hdr)
 867         return;
 868
 869     s->using_vnet_hdr = using_vnet_hdr != 0;
 870 }
 871
 872 static int tap_probe_vnet_hdr(int fd)
 873 {
 874 #if defined(TUNGETIFF) && defined(IFF_VNET_HDR)
 875     struct ifreq ifr;
 876
 877     if (ioctl(fd, TUNGETIFF, &ifr) != 0) {
 878         fprintf(stderr, "TUNGETIFF ioctl() failed: %s\n", strerror(errno));
 879         return 0;
 880     }
 881
 882     return ifr.ifr_flags & IFF_VNET_HDR;
 883 #else
 884     return 0;
 885 #endif
 886 }
 887
 888 #ifdef TUNSETOFFLOAD
 889 static void tap_set_offload(VLANClientState *vc, int csum, int tso4, int tso6,
 890                             int ecn)
 891 {
 892     TAPState *s = vc->opaque;
 893     unsigned int offload = 0;
 894
 895     if (csum) {
 896         offload |= TUN_F_CSUM;
 897         if (tso4)
 898             offload |= TUN_F_TSO4;
 899         if (tso6)
 900             offload |= TUN_F_TSO6;
 901         if ((tso4 || tso6) && ecn)
 902             offload |= TUN_F_TSO_ECN;
 903     }
 904
 905     if (ioctl(s->fd, TUNSETOFFLOAD, offload) != 0)
 906         fprintf(stderr, "TUNSETOFFLOAD ioctl() failed: %s\n",
 907                 strerror(errno));
 908 }
 909 #endif /* TUNSETOFFLOAD */
 910
 911 /* fd support */
 912
 913 static TAPState *net_tap_fd_init(VLANState *vlan,
 914                                  const char *model,
 915                                  const char *name,
 916                                  int fd,
 917                                  int vnet_hdr)
 918 {
 919     TAPState *s;
 920
 921     s = qemu_mallocz(sizeof(TAPState));
 922     s->fd = fd;
 923     s->has_vnet_hdr = vnet_hdr != 0;
 924     s->vc = qemu_new_vlan_client(vlan, model, name, tap_receive, NULL, s);
 925 #ifdef HAVE_IOVEC
 926     s->vc->fd_readv = tap_receive_iov;
 927 #endif
 928 #ifdef TUNSETOFFLOAD
 929     s->vc->set_offload = tap_set_offload;
 930 #endif
 931     qemu_set_fd_handler2(s->fd, tap_can_send, tap_send, NULL, s);
 932     snprintf(s->vc->info_str, sizeof(s->vc->info_str), "fd=%d", fd);
 933     return s;
 934 }
 935
 936 #if defined (_BSD) || defined (__FreeBSD_kernel__)
 937 static int tap_open(char *ifname, int ifname_size)
 938 {
 939     int fd;
 940     char *dev;
 941     struct stat s;
 942
 943     TFR(fd = open("/dev/tap", O_RDWR));
 944     if (fd < 0) {
 945         fprintf(stderr, "warning: could not open /dev/tap: no virtual network emulation\n");
 946         return -1;
 947     }
 948
 949     fstat(fd, &s);
 950     dev = devname(s.st_rdev, S_IFCHR);
 951     pstrcpy(ifname, ifname_size, dev);
 952
 953     fcntl(fd, F_SETFL, O_NONBLOCK);
 954     return fd;
 955 }
 956 #elif defined(__sun__)
 957 #define TUNNEWPPA       (('T'<<16) | 0x0001)
 958 /*
 959  * Allocate TAP device, returns opened fd.
 960  * Stores dev name in the first arg(must be large enough).
 961  */
 962 int tap_alloc(char *dev, size_t dev_size)
 963 {
 964     int tap_fd, if_fd, ppa = -1;
 965     static int ip_fd = 0;
 966     char *ptr;
 967
 968     static int arp_fd = 0;
 969     int ip_muxid, arp_muxid;
 970     struct strioctl  strioc_if, strioc_ppa;
 971     int link_type = I_PLINK;;
 972     struct lifreq ifr;
 973     char actual_name[32] = "";
 974
 975     memset(&ifr, 0x0, sizeof(ifr));
 976
 977     if( *dev ){
 978        ptr = dev;
 979        while( *ptr && !qemu_isdigit((int)*ptr) ) ptr++;
 980        ppa = atoi(ptr);
 981     }
 982
 983     /* Check if IP device was opened */
 984     if( ip_fd )
 985        close(ip_fd);
 986
 987     TFR(ip_fd = open("/dev/udp", O_RDWR, 0));
 988     if (ip_fd < 0) {
 989        syslog(LOG_ERR, "Can't open /dev/ip (actually /dev/udp)");
 990        return -1;
 991     }
 992
 993     TFR(tap_fd = open("/dev/tap", O_RDWR, 0));
 994     if (tap_fd < 0) {
 995        syslog(LOG_ERR, "Can't open /dev/tap");
 996        return -1;
 997     }
 998
 999     /* Assign a new PPA and get its unit number. */
1000     strioc_ppa.ic_cmd = TUNNEWPPA;
1001     strioc_ppa.ic_timout = 0;
1002     strioc_ppa.ic_len = sizeof(ppa);
1003     strioc_ppa.ic_dp = (char *)&ppa;
1004     if ((ppa = ioctl (tap_fd, I_STR, &strioc_ppa)) < 0)
1005        syslog (LOG_ERR, "Can't assign new interface");
1006
1007     TFR(if_fd = open("/dev/tap", O_RDWR, 0));
1008     if (if_fd < 0) {
1009        syslog(LOG_ERR, "Can't open /dev/tap (2)");
1010        return -1;
1011     }
1012     if(ioctl(if_fd, I_PUSH, "ip") < 0){
1013        syslog(LOG_ERR, "Can't push IP module");
1014        return -1;
1015     }
1016
1017     if (ioctl(if_fd, SIOCGLIFFLAGS, &ifr) < 0)
1018         syslog(LOG_ERR, "Can't get flags\n");
1019
1020     snprintf (actual_name, 32, "tap%d", ppa);
1021     pstrcpy(ifr.lifr_name, sizeof(ifr.lifr_name), actual_name);
1022
1023     ifr.lifr_ppa = ppa;
1024     /* Assign ppa according to the unit number returned by tun device */
1025
1026     if (ioctl (if_fd, SIOCSLIFNAME, &ifr) < 0)
1027         syslog (LOG_ERR, "Can't set PPA %d", ppa);
1028     if (ioctl(if_fd, SIOCGLIFFLAGS, &ifr) <0)
1029         syslog (LOG_ERR, "Can't get flags\n");
1030     /* Push arp module to if_fd */
1031     if (ioctl (if_fd, I_PUSH, "arp") < 0)
1032         syslog (LOG_ERR, "Can't push ARP module (2)");
1033
1034     /* Push arp module to ip_fd */
1035     if (ioctl (ip_fd, I_POP, NULL) < 0)
1036         syslog (LOG_ERR, "I_POP failed\n");
1037     if (ioctl (ip_fd, I_PUSH, "arp") < 0)
1038         syslog (LOG_ERR, "Can't push ARP module (3)\n");
1039     /* Open arp_fd */
1040     TFR(arp_fd = open ("/dev/tap", O_RDWR, 0));
1041     if (arp_fd < 0)
1042        syslog (LOG_ERR, "Can't open %s\n", "/dev/tap");
1043
1044     /* Set ifname to arp */
1045     strioc_if.ic_cmd = SIOCSLIFNAME;
1046     strioc_if.ic_timout = 0;
1047     strioc_if.ic_len = sizeof(ifr);
1048     strioc_if.ic_dp = (char *)&ifr;
1049     if (ioctl(arp_fd, I_STR, &strioc_if) < 0){
1050         syslog (LOG_ERR, "Can't set ifname to arp\n");
1051     }
1052
1053     if((ip_muxid = ioctl(ip_fd, I_LINK, if_fd)) < 0){
1054        syslog(LOG_ERR, "Can't link TAP device to IP");
1055        return -1;
1056     }
1057
1058     if ((arp_muxid = ioctl (ip_fd, link_type, arp_fd)) < 0)
1059         syslog (LOG_ERR, "Can't link TAP device to ARP");
1060
1061     close (if_fd);
1062
1063     memset(&ifr, 0x0, sizeof(ifr));
1064     pstrcpy(ifr.lifr_name, sizeof(ifr.lifr_name), actual_name);
1065     ifr.lifr_ip_muxid  = ip_muxid;
1066     ifr.lifr_arp_muxid = arp_muxid;
1067
1068     if (ioctl (ip_fd, SIOCSLIFMUXID, &ifr) < 0)
1069     {
1070       ioctl (ip_fd, I_PUNLINK , arp_muxid);
1071       ioctl (ip_fd, I_PUNLINK, ip_muxid);
1072       syslog (LOG_ERR, "Can't set multiplexor id");
1073     }
1074
1075     snprintf(dev, dev_size, "tap%d", ppa);
1076     return tap_fd;
1077 }
1078
1079 static int tap_open(char *ifname, int ifname_size, int *vnet_hdr)
1080 {
1081     char  dev[10]="";
1082     int fd;
1083     if( (fd = tap_alloc(dev, sizeof(dev))) < 0 ){
1084        fprintf(stderr, "Cannot allocate TAP device\n");
1085        return -1;
1086     }
1087     pstrcpy(ifname, ifname_size, dev);
1088     fcntl(fd, F_SETFL, O_NONBLOCK);
1089     return fd;
1090 }
1091 #elif defined (_AIX)
1092 static int tap_open(char *ifname, int ifname_size)
1093 {
1094     fprintf (stderr, "no tap on AIX\n");
1095     return -1;
1096 }
1097 #else
1098 static int tap_open(char *ifname, int ifname_size, int *vnet_hdr)
1099 {
1100     struct ifreq ifr;
1101     int fd, ret;
1102
1103     TFR(fd = open("/dev/net/tun", O_RDWR));
1104     if (fd < 0) {
1105         fprintf(stderr, "warning: could not open /dev/net/tun: no virtual network emulation\n");
1106         return -1;
1107     }
1108     memset(&ifr, 0, sizeof(ifr));
1109     ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
1110
1111 #if defined(TUNGETFEATURES) && defined(IFF_VNET_HDR)
1112     {
1113         unsigned int features;
1114
1115         if (ioctl(fd, TUNGETFEATURES, &features) == 0 &&
1116             features & IFF_VNET_HDR) {
1117             *vnet_hdr = 1;
1118             ifr.ifr_flags |= IFF_VNET_HDR;
1119         }
1120     }
1121 #endif
1122
1123     if (ifname[0] != '\0')
1124         pstrcpy(ifr.ifr_name, IFNAMSIZ, ifname);
1125     else
1126         pstrcpy(ifr.ifr_name, IFNAMSIZ, "tap%d");
1127     ret = ioctl(fd, TUNSETIFF, (void *) &ifr);
1128     if (ret != 0) {
1129         fprintf(stderr, "warning: could not configure /dev/net/tun: no virtual network emulation\n");
1130         close(fd);
1131         return -1;
1132     }
1133     pstrcpy(ifname, ifname_size, ifr.ifr_name);
1134     fcntl(fd, F_SETFL, O_NONBLOCK);
1135     return fd;
1136 }
1137 #endif
1138
1139 static int launch_script(const char *setup_script, const char *ifname, int fd)
1140 {
1141     int pid, status;
1142     char *args[3];
1143     char **parg;
1144
1145         /* try to launch network script */
1146         pid = fork();
1147         if (pid >= 0) {
1148             if (pid == 0) {
1149                 int open_max = sysconf (_SC_OPEN_MAX), i;
1150                 for (i = 0; i < open_max; i++)
1151                     if (i != STDIN_FILENO &&
1152                         i != STDOUT_FILENO &&
1153                         i != STDERR_FILENO &&
1154                         i != fd)
1155                         close(i);
1156
1157                 parg = args;
1158                 *parg++ = (char *)setup_script;
1159                 *parg++ = (char *)ifname;
1160                 *parg++ = NULL;
1161                 execv(setup_script, args);
1162                 _exit(1);
1163             }
1164             while (waitpid(pid, &status, 0) != pid);
1165             if (!WIFEXITED(status) ||
1166                 WEXITSTATUS(status) != 0) {
1167                 fprintf(stderr, "%s: could not launch network script\n",
1168                         setup_script);
1169                 return -1;
1170             }
1171         }
1172     return 0;
1173 }
1174
1175 static int net_tap_init(VLANState *vlan, const char *model,
1176                         const char *name, const char *ifname1,
1177                         const char *setup_script, const char *down_script)
1178 {
1179     TAPState *s;
1180     int fd;
1181     int vnet_hdr;
1182     char ifname[128];
1183
1184     if (ifname1 != NULL)
1185         pstrcpy(ifname, sizeof(ifname), ifname1);
1186     else
1187         ifname[0] = '\0';
1188     vnet_hdr = 0;
1189     TFR(fd = tap_open(ifname, sizeof(ifname), &vnet_hdr));
1190     if (fd < 0)
1191         return -1;
1192
1193     if (!setup_script || !strcmp(setup_script, "no"))
1194         setup_script = "";
1195     if (setup_script[0] != '\0') {
1196         if (launch_script(setup_script, ifname, fd))
1197             return -1;
1198     }
1199     s = net_tap_fd_init(vlan, model, name, fd, vnet_hdr);
1200     if (!s)
1201         return -1;
1202
1203     snprintf(s->vc->info_str, sizeof(s->vc->info_str),
1204              "ifname=%s,script=%s,downscript=%s",
1205              ifname, setup_script, down_script);
1206     if (down_script && strcmp(down_script, "no")) {
1207         snprintf(s->down_script, sizeof(s->down_script), "%s", down_script);
1208         snprintf(s->down_script_arg, sizeof(s->down_script_arg), "%s", ifname);
1209     }
1210     return 0;
1211 }
1212
1213 #endif /* !_WIN32 */
1214
1215 #if defined(CONFIG_VDE)
1216 typedef struct VDEState {
1217     VLANClientState *vc;
1218     VDECONN *vde;
1219 } VDEState;
1220
1221 static void vde_to_qemu(void *opaque)
1222 {
1223     VDEState *s = opaque;
1224     uint8_t buf[4096];
1225     int size;
1226
1227     size = vde_recv(s->vde, buf, sizeof(buf), 0);
1228     if (size > 0) {
1229         qemu_send_packet(s->vc, buf, size);
1230     }
1231 }
1232
1233 static void vde_from_qemu(void *opaque, const uint8_t *buf, int size)
1234 {
1235     VDEState *s = opaque;
1236     int ret;
1237     for(;;) {
1238         ret = vde_send(s->vde, buf, size, 0);
1239         if (ret < 0 && errno == EINTR) {
1240         } else {
1241             break;
1242         }
1243     }
1244 }
1245
1246 static int net_vde_init(VLANState *vlan, const char *model,
1247                         const char *name, const char *sock,
1248                         int port, const char *group, int mode)
1249 {
1250     VDEState *s;
1251     char *init_group = strlen(group) ? (char *)group : NULL;
1252     char *init_sock = strlen(sock) ? (char *)sock : NULL;
1253
1254     struct vde_open_args args = {
1255         .port = port,
1256         .group = init_group,
1257         .mode = mode,
1258     };
1259
1260     s = qemu_mallocz(sizeof(VDEState));
1261     s->vde = vde_open(init_sock, "QEMU", &args);
1262     if (!s->vde){
1263         free(s);
1264         return -1;
1265     }
1266     s->vc = qemu_new_vlan_client(vlan, model, name, vde_from_qemu, NULL, s);
1267     qemu_set_fd_handler(vde_datafd(s->vde), vde_to_qemu, NULL, s);
1268     snprintf(s->vc->info_str, sizeof(s->vc->info_str), "sock=%s,fd=%d",
1269              sock, vde_datafd(s->vde));
1270     return 0;
1271 }
1272 #endif
1273
1274 /* network connection */
1275 typedef struct NetSocketState {
1276     VLANClientState *vc;
1277     int fd;
1278     int state; /* 0 = getting length, 1 = getting data */
1279     int index;
1280     int packet_len;
1281     uint8_t buf[4096];
1282     struct sockaddr_in dgram_dst; /* contains inet host and port destination iff connectionless (SOCK_DGRAM) */
1283 } NetSocketState;
1284
1285 typedef struct NetSocketListenState {
1286     VLANState *vlan;
1287     char *model;
1288     char *name;
1289     int fd;
1290 } NetSocketListenState;
1291
1292 /* XXX: we consider we can send the whole packet without blocking */
1293 static void net_socket_receive(void *opaque, const uint8_t *buf, int size)
1294 {
1295     NetSocketState *s = opaque;
1296     uint32_t len;
1297     len = htonl(size);
1298
1299     send_all(s->fd, (const uint8_t *)&len, sizeof(len));
1300     send_all(s->fd, buf, size);
1301 }
1302
1303 static void net_socket_receive_dgram(void *opaque, const uint8_t *buf, int size)
1304 {
1305     NetSocketState *s = opaque;
1306     sendto(s->fd, buf, size, 0,
1307            (struct sockaddr *)&s->dgram_dst, sizeof(s->dgram_dst));
1308 }
1309
1310 static void net_socket_send(void *opaque)
1311 {
1312     NetSocketState *s = opaque;
1313     int l, size, err;
1314     uint8_t buf1[4096];
1315     const uint8_t *buf;
1316
1317     size = recv(s->fd, buf1, sizeof(buf1), 0);
1318     if (size < 0) {
1319         err = socket_error();
1320         if (err != EWOULDBLOCK)
1321             goto eoc;
1322     } else if (size == 0) {
1323         /* end of connection */
1324     eoc:
1325         qemu_set_fd_handler(s->fd, NULL, NULL, NULL);
1326         closesocket(s->fd);
1327         return;
1328     }
1329     buf = buf1;
1330     while (size > 0) {
1331         /* reassemble a packet from the network */
1332         switch(s->state) {
1333         case 0:
1334             l = 4 - s->index;
1335             if (l > size)
1336                 l = size;
1337             memcpy(s->buf + s->index, buf, l);
1338             buf += l;
1339             size -= l;
1340             s->index += l;
1341             if (s->index == 4) {
1342                 /* got length */
1343                 s->packet_len = ntohl(*(uint32_t *)s->buf);
1344                 s->index = 0;
1345                 s->state = 1;
1346             }
1347             break;
1348         case 1:
1349             l = s->packet_len - s->index;
1350             if (l > size)
1351                 l = size;
1352             memcpy(s->buf + s->index, buf, l);
1353             s->index += l;
1354             buf += l;
1355             size -= l;
1356             if (s->index >= s->packet_len) {
1357                 qemu_send_packet(s->vc, s->buf, s->packet_len);
1358                 s->index = 0;
1359                 s->state = 0;
1360             }
1361             break;
1362         }
1363     }
1364 }
1365
1366 static void net_socket_send_dgram(void *opaque)
1367 {
1368     NetSocketState *s = opaque;
1369     int size;
1370
1371     size = recv(s->fd, s->buf, sizeof(s->buf), 0);
1372     if (size < 0)
1373         return;
1374     if (size == 0) {
1375         /* end of connection */
1376         qemu_set_fd_handler(s->fd, NULL, NULL, NULL);
1377         return;
1378     }
1379     qemu_send_packet(s->vc, s->buf, size);
1380 }
1381
1382 static int net_socket_mcast_create(struct sockaddr_in *mcastaddr)
1383 {
1384     struct ip_mreq imr;
1385     int fd;
1386     int val, ret;
1387     if (!IN_MULTICAST(ntohl(mcastaddr->sin_addr.s_addr))) {
1388         fprintf(stderr, "qemu: error: specified mcastaddr \"%s\" (0x%08x) does not contain a multicast address\n",
1389                 inet_ntoa(mcastaddr->sin_addr),
1390                 (int)ntohl(mcastaddr->sin_addr.s_addr));
1391         return -1;
1392
1393     }
1394     fd = socket(PF_INET, SOCK_DGRAM, 0);
1395     if (fd < 0) {
1396         perror("socket(PF_INET, SOCK_DGRAM)");
1397         return -1;
1398     }
1399
1400     val = 1;
1401     ret=setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
1402                    (const char *)&val, sizeof(val));
1403     if (ret < 0) {
1404         perror("setsockopt(SOL_SOCKET, SO_REUSEADDR)");
1405         goto fail;
1406     }
1407
1408     ret = bind(fd, (struct sockaddr *)mcastaddr, sizeof(*mcastaddr));
1409     if (ret < 0) {
1410         perror("bind");
1411         goto fail;
1412     }
1413
1414     /* Add host to multicast group */
1415     imr.imr_multiaddr = mcastaddr->sin_addr;
1416     imr.imr_interface.s_addr = htonl(INADDR_ANY);
1417
1418     ret = setsockopt(fd, IPPROTO_IP, IP_ADD_MEMBERSHIP,
1419                      (const char *)&imr, sizeof(struct ip_mreq));
1420     if (ret < 0) {
1421         perror("setsockopt(IP_ADD_MEMBERSHIP)");
1422         goto fail;
1423     }
1424
1425     /* Force mcast msgs to loopback (eg. several QEMUs in same host */
1426     val = 1;
1427     ret=setsockopt(fd, IPPROTO_IP, IP_MULTICAST_LOOP,
1428                    (const char *)&val, sizeof(val));
1429     if (ret < 0) {
1430         perror("setsockopt(SOL_IP, IP_MULTICAST_LOOP)");
1431         goto fail;
1432     }
1433
1434     socket_set_nonblock(fd);
1435     return fd;
1436 fail:
1437     if (fd >= 0)
1438         closesocket(fd);
1439     return -1;
1440 }
1441
1442 static NetSocketState *net_socket_fd_init_dgram(VLANState *vlan,
1443                                                 const char *model,
1444                                                 const char *name,
1445                                                 int fd, int is_connected)
1446 {
1447     struct sockaddr_in saddr;
1448     int newfd;
1449     socklen_t saddr_len;
1450     NetSocketState *s;
1451
1452     /* fd passed: multicast: "learn" dgram_dst address from bound address and save it
1453      * Because this may be "shared" socket from a "master" process, datagrams would be recv()
1454      * by ONLY ONE process: we must "clone" this dgram socket --jjo
1455      */
1456
1457     if (is_connected) {
1458         if (getsockname(fd, (struct sockaddr *) &saddr, &saddr_len) == 0) {
1459             /* must be bound */
1460             if (saddr.sin_addr.s_addr==0) {
1461                 fprintf(stderr, "qemu: error: init_dgram: fd=%d unbound, cannot setup multicast dst addr\n",
1462                         fd);
1463                 return NULL;
1464             }
1465             /* clone dgram socket */
1466             newfd = net_socket_mcast_create(&saddr);
1467             if (newfd < 0) {
1468                 /* error already reported by net_socket_mcast_create() */
1469                 close(fd);
1470                 return NULL;
1471             }
1472             /* clone newfd to fd, close newfd */
1473             dup2(newfd, fd);
1474             close(newfd);
1475
1476         } else {
1477             fprintf(stderr, "qemu: error: init_dgram: fd=%d failed getsockname(): %s\n",
1478                     fd, strerror(errno));
1479             return NULL;
1480         }
1481     }
1482
1483     s = qemu_mallocz(sizeof(NetSocketState));
1484     s->fd = fd;
1485
1486     s->vc = qemu_new_vlan_client(vlan, model, name, net_socket_receive_dgram, NULL, s);
1487     qemu_set_fd_handler(s->fd, net_socket_send_dgram, NULL, s);
1488
1489     /* mcast: save bound address as dst */
1490     if (is_connected) s->dgram_dst=saddr;
1491
1492     snprintf(s->vc->info_str, sizeof(s->vc->info_str),
1493             "socket: fd=%d (%s mcast=%s:%d)",
1494             fd, is_connected? "cloned" : "",
1495             inet_ntoa(saddr.sin_addr), ntohs(saddr.sin_port));
1496     return s;
1497 }
1498
1499 static void net_socket_connect(void *opaque)
1500 {
1501     NetSocketState *s = opaque;
1502     qemu_set_fd_handler(s->fd, net_socket_send, NULL, s);
1503 }
1504
1505 static NetSocketState *net_socket_fd_init_stream(VLANState *vlan,
1506                                                  const char *model,
1507                                                  const char *name,
1508                                                  int fd, int is_connected)
1509 {
1510     NetSocketState *s;
1511     s = qemu_mallocz(sizeof(NetSocketState));
1512     s->fd = fd;
1513     s->vc = qemu_new_vlan_client(vlan, model, name,
1514                                  net_socket_receive, NULL, s);
1515     snprintf(s->vc->info_str, sizeof(s->vc->info_str),
1516              "socket: fd=%d", fd);
1517     if (is_connected) {
1518         net_socket_connect(s);
1519     } else {
1520         qemu_set_fd_handler(s->fd, NULL, net_socket_connect, s);
1521     }
1522     return s;
1523 }
1524
1525 static NetSocketState *net_socket_fd_init(VLANState *vlan,
1526                                           const char *model, const char *name,
1527                                           int fd, int is_connected)
1528 {
1529     int so_type=-1, optlen=sizeof(so_type);
1530
1531     if(getsockopt(fd, SOL_SOCKET, SO_TYPE, (char *)&so_type,
1532         (socklen_t *)&optlen)< 0) {
1533         fprintf(stderr, "qemu: error: getsockopt(SO_TYPE) for fd=%d failed\n", fd);
1534         return NULL;
1535     }
1536     switch(so_type) {
1537     case SOCK_DGRAM:
1538         return net_socket_fd_init_dgram(vlan, model, name, fd, is_connected);
1539     case SOCK_STREAM:
1540         return net_socket_fd_init_stream(vlan, model, name, fd, is_connected);
1541     default:
1542         /* who knows ... this could be a eg. a pty, do warn and continue as stream */
1543         fprintf(stderr, "qemu: warning: socket type=%d for fd=%d is not SOCK_DGRAM or SOCK_STREAM\n", so_type, fd);
1544         return net_socket_fd_init_stream(vlan, model, name, fd, is_connected);
1545     }
1546     return NULL;
1547 }
1548
1549 static void net_socket_accept(void *opaque)
1550 {
1551     NetSocketListenState *s = opaque;
1552     NetSocketState *s1;
1553     struct sockaddr_in saddr;
1554     socklen_t len;
1555     int fd;
1556
1557     for(;;) {
1558         len = sizeof(saddr);
1559         fd = accept(s->fd, (struct sockaddr *)&saddr, &len);
1560         if (fd < 0 && errno != EINTR) {
1561             return;
1562         } else if (fd >= 0) {
1563             break;
1564         }
1565     }
1566     s1 = net_socket_fd_init(s->vlan, s->model, s->name, fd, 1);
1567     if (!s1) {
1568         closesocket(fd);
1569     } else {
1570         snprintf(s1->vc->info_str, sizeof(s1->vc->info_str),
1571                  "socket: connection from %s:%d",
1572                  inet_ntoa(saddr.sin_addr), ntohs(saddr.sin_port));
1573     }
1574 }
1575
1576 static int net_socket_listen_init(VLANState *vlan,
1577                                   const char *model,
1578                                   const char *name,
1579                                   const char *host_str)
1580 {
1581     NetSocketListenState *s;
1582     int fd, val, ret;
1583     struct sockaddr_in saddr;
1584
1585     if (parse_host_port(&saddr, host_str) < 0)
1586         return -1;
1587
1588     s = qemu_mallocz(sizeof(NetSocketListenState));
1589
1590     fd = socket(PF_INET, SOCK_STREAM, 0);
1591     if (fd < 0) {
1592         perror("socket");
1593         return -1;
1594     }
1595     socket_set_nonblock(fd);
1596
1597     /* allow fast reuse */
1598     val = 1;
1599     setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, (const char *)&val, sizeof(val));
1600
1601     ret = bind(fd, (struct sockaddr *)&saddr, sizeof(saddr));
1602     if (ret < 0) {
1603         perror("bind");
1604         return -1;
1605     }
1606     ret = listen(fd, 0);
1607     if (ret < 0) {
1608         perror("listen");
1609         return -1;
1610     }
1611     s->vlan = vlan;
1612     s->model = strdup(model);
1613     s->name = strdup(name);
1614     s->fd = fd;
1615     qemu_set_fd_handler(fd, net_socket_accept, NULL, s);
1616     return 0;
1617 }
1618
1619 static int net_socket_connect_init(VLANState *vlan,
1620                                    const char *model,
1621                                    const char *name,
1622                                    const char *host_str)
1623 {
1624     NetSocketState *s;
1625     int fd, connected, ret, err;
1626     struct sockaddr_in saddr;
1627
1628     if (parse_host_port(&saddr, host_str) < 0)
1629         return -1;
1630
1631     fd = socket(PF_INET, SOCK_STREAM, 0);
1632     if (fd < 0) {
1633         perror("socket");
1634         return -1;
1635     }
1636     socket_set_nonblock(fd);
1637
1638     connected = 0;
1639     for(;;) {
1640         ret = connect(fd, (struct sockaddr *)&saddr, sizeof(saddr));
1641         if (ret < 0) {
1642             err = socket_error();
1643             if (err == EINTR || err == EWOULDBLOCK) {
1644             } else if (err == EINPROGRESS) {
1645                 break;
1646 #ifdef _WIN32
1647             } else if (err == WSAEALREADY) {
1648                 break;
1649 #endif
1650             } else {
1651                 perror("connect");
1652                 closesocket(fd);
1653                 return -1;
1654             }
1655         } else {
1656             connected = 1;
1657             break;
1658         }
1659     }
1660     s = net_socket_fd_init(vlan, model, name, fd, connected);
1661     if (!s)
1662         return -1;
1663     snprintf(s->vc->info_str, sizeof(s->vc->info_str),
1664              "socket: connect to %s:%d",
1665              inet_ntoa(saddr.sin_addr), ntohs(saddr.sin_port));
1666     return 0;
1667 }
1668
1669 static int net_socket_mcast_init(VLANState *vlan,
1670                                  const char *model,
1671                                  const char *name,
1672                                  const char *host_str)
1673 {
1674     NetSocketState *s;
1675     int fd;
1676     struct sockaddr_in saddr;
1677
1678     if (parse_host_port(&saddr, host_str) < 0)
1679         return -1;
1680
1681
1682     fd = net_socket_mcast_create(&saddr);
1683     if (fd < 0)
1684         return -1;
1685
1686     s = net_socket_fd_init(vlan, model, name, fd, 0);
1687     if (!s)
1688         return -1;
1689
1690     s->dgram_dst = saddr;
1691
1692     snprintf(s->vc->info_str, sizeof(s->vc->info_str),
1693              "socket: mcast=%s:%d",
1694              inet_ntoa(saddr.sin_addr), ntohs(saddr.sin_port));
1695     return 0;
1696
1697 }
1698
1699 /* find or alloc a new VLAN */
1700 VLANState *qemu_find_vlan(int id)
1701 {
1702     VLANState **pvlan, *vlan;
1703     for(vlan = first_vlan; vlan != NULL; vlan = vlan->next) {
1704         if (vlan->id == id)
1705             return vlan;
1706     }
1707     vlan = qemu_mallocz(sizeof(VLANState));
1708     vlan->id = id;
1709     vlan->next = NULL;
1710     pvlan = &first_vlan;
1711     while (*pvlan != NULL)
1712         pvlan = &(*pvlan)->next;
1713     *pvlan = vlan;
1714     return vlan;
1715 }
1716
1717 static int nic_get_free_idx(void)
1718 {
1719     int index;
1720
1721     for (index = 0; index < MAX_NICS; index++)
1722         if (!nd_table[index].used)
1723             return index;
1724     return -1;
1725 }
1726
1727 void qemu_check_nic_model(NICInfo *nd, const char *model)
1728 {
1729     const char *models[2];
1730
1731     models[0] = model;
1732     models[1] = NULL;
1733
1734     qemu_check_nic_model_list(nd, models, model);
1735 }
1736
1737 void qemu_check_nic_model_list(NICInfo *nd, const char * const *models,
1738                                const char *default_model)
1739 {
1740     int i, exit_status = 0;
1741
1742     if (!nd->model)
1743         nd->model = strdup(default_model);
1744
1745     if (strcmp(nd->model, "?") != 0) {
1746         for (i = 0 ; models[i]; i++)
1747             if (strcmp(nd->model, models[i]) == 0)
1748                 return;
1749
1750         fprintf(stderr, "qemu: Unsupported NIC model: %s\n", nd->model);
1751         exit_status = 1;
1752     }
1753
1754     fprintf(stderr, "qemu: Supported NIC models: ");
1755     for (i = 0 ; models[i]; i++)
1756         fprintf(stderr, "%s%c", models[i], models[i+1] ? ',' : '\n');
1757
1758     exit(exit_status);
1759 }
1760
1761 int net_client_init(const char *device, const char *p)
1762 {
1763     char buf[1024];
1764     int vlan_id, ret;
1765     VLANState *vlan;
1766     char *name = NULL;
1767
1768     vlan_id = 0;
1769     if (get_param_value(buf, sizeof(buf), "vlan", p)) {
1770         vlan_id = strtol(buf, NULL, 0);
1771     }
1772     vlan = qemu_find_vlan(vlan_id);
1773     if (!vlan) {
1774         fprintf(stderr, "Could not create vlan %d\n", vlan_id);
1775         return -1;
1776     }
1777     if (get_param_value(buf, sizeof(buf), "name", p)) {
1778         name = strdup(buf);
1779     }
1780     if (!strcmp(device, "nic")) {
1781         NICInfo *nd;
1782         uint8_t *macaddr;
1783         int idx = nic_get_free_idx();
1784
1785         if (idx == -1 || nb_nics >= MAX_NICS) {
1786             fprintf(stderr, "Too Many NICs\n");
1787             return -1;
1788         }
1789         nd = &nd_table[idx];
1790         macaddr = nd->macaddr;
1791         macaddr[0] = 0x52;
1792         macaddr[1] = 0x54;
1793         macaddr[2] = 0x00;
1794         macaddr[3] = 0x12;
1795         macaddr[4] = 0x34;
1796         macaddr[5] = 0x56 + idx;
1797
1798         if (get_param_value(buf, sizeof(buf), "macaddr", p)) {
1799             if (parse_macaddr(macaddr, buf) < 0) {
1800                 fprintf(stderr, "invalid syntax for ethernet address\n");
1801                 return -1;
1802             }
1803         }
1804         if (get_param_value(buf, sizeof(buf), "model", p)) {
1805             nd->model = strdup(buf);
1806         }
1807         nd->vlan = vlan;
1808         nd->name = name;
1809         nd->used = 1;
1810         name = NULL;
1811         nb_nics++;
1812         vlan->nb_guest_devs++;
1813         ret = idx;
1814     } else
1815     if (!strcmp(device, "none")) {
1816         /* does nothing. It is needed to signal that no network cards
1817            are wanted */
1818         ret = 0;
1819     } else
1820 #ifdef CONFIG_SLIRP
1821     if (!strcmp(device, "user")) {
1822         if (get_param_value(buf, sizeof(buf), "hostname", p)) {
1823             pstrcpy(slirp_hostname, sizeof(slirp_hostname), buf);
1824         }
1825         if (get_param_value(buf, sizeof(buf), "restrict", p)) {
1826             slirp_restrict = (buf[0] == 'y') ? 1 : 0;
1827         }
1828         if (get_param_value(buf, sizeof(buf), "ip", p)) {
1829             slirp_ip = strdup(buf);
1830         }
1831         vlan->nb_host_devs++;
1832         ret = net_slirp_init(vlan, device, name);
1833     } else if (!strcmp(device, "channel")) {
1834         long port;
1835         char name[20], *devname;
1836         struct VMChannel *vmc;
1837
1838         port = strtol(p, &devname, 10);
1839         devname++;
1840         if (port < 1 || port > 65535) {
1841             fprintf(stderr, "vmchannel wrong port number\n");
1842             return -1;
1843         }
1844         vmc = malloc(sizeof(struct VMChannel));
1845         snprintf(name, 20, "vmchannel%ld", port);
1846         vmc->hd = qemu_chr_open(name, devname, NULL);
1847         if (!vmc->hd) {
1848             fprintf(stderr, "qemu: could not open vmchannel device"
1849                     "'%s'\n", devname);
1850             return -1;
1851         }
1852         vmc->port = port;
1853         slirp_add_exec(3, vmc->hd, 4, port);
1854         qemu_chr_add_handlers(vmc->hd, vmchannel_can_read, vmchannel_read,
1855                 NULL, vmc);
1856         ret = 0;
1857     } else
1858 #endif
1859 #ifdef _WIN32
1860     if (!strcmp(device, "tap")) {
1861         char ifname[64];
1862         if (get_param_value(ifname, sizeof(ifname), "ifname", p) <= 0) {
1863             fprintf(stderr, "tap: no interface name\n");
1864             return -1;
1865         }
1866         vlan->nb_host_devs++;
1867         ret = tap_win32_init(vlan, device, name, ifname);
1868     } else
1869 #elif defined (_AIX)
1870 #else
1871     if (!strcmp(device, "tap")) {
1872         char ifname[64];
1873         char setup_script[1024], down_script[1024];
1874         int fd;
1875         vlan->nb_host_devs++;
1876         if (get_param_value(buf, sizeof(buf), "fd", p) > 0) {
1877             fd = strtol(buf, NULL, 0);
1878             fcntl(fd, F_SETFL, O_NONBLOCK);
1879             ret = -1;
1880                 if (net_tap_fd_init(vlan, device, name, fd,
1881                                     tap_probe_vnet_hdr(fd)))
1882                 ret = 0;
1883         } else {
1884             if (get_param_value(ifname, sizeof(ifname), "ifname", p) <= 0) {
1885                 ifname[0] = '\0';
1886             }
1887             if (get_param_value(setup_script, sizeof(setup_script), "script", p) == 0) {
1888                 pstrcpy(setup_script, sizeof(setup_script), DEFAULT_NETWORK_SCRIPT);
1889             }
1890             if (get_param_value(down_script, sizeof(down_script), "downscript", p) == 0) {
1891                 pstrcpy(down_script, sizeof(down_script), DEFAULT_NETWORK_DOWN_SCRIPT);
1892             }
1893             ret = net_tap_init(vlan, device, name, ifname, setup_script, down_script);
1894         }
1895     } else
1896 #endif
1897     if (!strcmp(device, "socket")) {
1898         if (get_param_value(buf, sizeof(buf), "fd", p) > 0) {
1899             int fd;
1900             fd = strtol(buf, NULL, 0);
1901             ret = -1;
1902             if (net_socket_fd_init(vlan, device, name, fd, 1))
1903                 ret = 0;
1904         } else if (get_param_value(buf, sizeof(buf), "listen", p) > 0) {
1905             ret = net_socket_listen_init(vlan, device, name, buf);
1906         } else if (get_param_value(buf, sizeof(buf), "connect", p) > 0) {
1907             ret = net_socket_connect_init(vlan, device, name, buf);
1908         } else if (get_param_value(buf, sizeof(buf), "mcast", p) > 0) {
1909             ret = net_socket_mcast_init(vlan, device, name, buf);
1910         } else {
1911             fprintf(stderr, "Unknown socket options: %s\n", p);
1912             return -1;
1913         }
1914         vlan->nb_host_devs++;
1915     } else
1916 #ifdef CONFIG_VDE
1917     if (!strcmp(device, "vde")) {
1918         char vde_sock[1024], vde_group[512];
1919         int vde_port, vde_mode;
1920         vlan->nb_host_devs++;
1921         if (get_param_value(vde_sock, sizeof(vde_sock), "sock", p) <= 0) {
1922             vde_sock[0] = '\0';
1923         }
1924         if (get_param_value(buf, sizeof(buf), "port", p) > 0) {
1925             vde_port = strtol(buf, NULL, 10);
1926         } else {
1927             vde_port = 0;
1928         }
1929         if (get_param_value(vde_group, sizeof(vde_group), "group", p) <= 0) {
1930             vde_group[0] = '\0';
1931         }
1932         if (get_param_value(buf, sizeof(buf), "mode", p) > 0) {
1933             vde_mode = strtol(buf, NULL, 8);
1934         } else {
1935             vde_mode = 0700;
1936         }
1937         ret = net_vde_init(vlan, device, name, vde_sock, vde_port, vde_group, vde_mode);
1938     } else
1939 #endif
1940     {
1941         fprintf(stderr, "Unknown network device: %s\n", device);
1942         if (name)
1943             free(name);
1944         return -1;
1945     }
1946     if (ret < 0) {
1947         fprintf(stderr, "Could not initialize device '%s'\n", device);
1948     }
1949     if (name)
1950         free(name);
1951     return ret;
1952 }
1953
1954 void net_client_uninit(NICInfo *nd)
1955 {
1956     nd->vlan->nb_guest_devs--;
1957     nb_nics--;
1958     nd->used = 0;
1959     free((void *)nd->model);
1960 }
1961
1962 static int net_host_check_device(const char *device)
1963 {
1964     int i;
1965     const char *valid_param_list[] = { "tap", "socket"
1966 #ifdef CONFIG_SLIRP
1967                                        ,"user"
1968 #endif
1969 #ifdef CONFIG_VDE
1970                                        ,"vde"
1971 #endif
1972     };
1973     for (i = 0; i < sizeof(valid_param_list) / sizeof(char *); i++) {
1974         if (!strncmp(valid_param_list[i], device,
1975                      strlen(valid_param_list[i])))
1976             return 1;
1977     }
1978
1979     return 0;
1980 }
1981
1982 void net_host_device_add(const char *device, const char *opts)
1983 {
1984     if (!net_host_check_device(device)) {
1985         term_printf("invalid host network device %s\n", device);
1986         return;
1987     }
1988     net_client_init(device, opts);
1989 }
1990
1991 void net_host_device_remove(int vlan_id, const char *device)
1992 {
1993     VLANState *vlan;
1994     VLANClientState *vc;
1995
1996     if (!net_host_check_device(device)) {
1997         term_printf("invalid host network device %s\n", device);
1998         return;
1999     }
2000
2001     vlan = qemu_find_vlan(vlan_id);
2002     if (!vlan) {
2003         term_printf("can't find vlan %d\n", vlan_id);
2004         return;
2005     }
2006
2007    for(vc = vlan->first_client; vc != NULL; vc = vc->next)
2008         if (!strcmp(vc->name, device))
2009             break;
2010
2011     if (!vc) {
2012         term_printf("can't find device %s\n", device);
2013         return;
2014     }
2015     qemu_del_vlan_client(vc);
2016 }
2017
2018 int net_client_parse(const char *str)
2019 {
2020     const char *p;
2021     char *q;
2022     char device[64];
2023
2024     p = str;
2025     q = device;
2026     while (*p != '\0' && *p != ',') {
2027         if ((q - device) < sizeof(device) - 1)
2028             *q++ = *p;
2029         p++;
2030     }
2031     *q = '\0';
2032     if (*p == ',')
2033         p++;
2034
2035     return net_client_init(device, p);
2036 }
2037
2038 void do_info_network(void)
2039 {
2040     VLANState *vlan;
2041     VLANClientState *vc;
2042
2043     for(vlan = first_vlan; vlan != NULL; vlan = vlan->next) {
2044         term_printf("VLAN %d devices:\n", vlan->id);
2045         for(vc = vlan->first_client; vc != NULL; vc = vc->next)
2046             term_printf("  %s: %s\n", vc->name, vc->info_str);
2047     }
2048 }
2049
2050 int do_set_link(const char *name, const char *up_or_down)
2051 {
2052     VLANState *vlan;
2053     VLANClientState *vc = NULL;
2054
2055     for (vlan = first_vlan; vlan != NULL; vlan = vlan->next)
2056         for (vc = vlan->first_client; vc != NULL; vc = vc->next)
2057             if (strcmp(vc->name, name) == 0)
2058                 goto done;
2059 done:
2060
2061     if (!vc) {
2062         term_printf("could not find network device '%s'", name);
2063         return 0;
2064     }
2065
2066     if (strcmp(up_or_down, "up") == 0)
2067         vc->link_down = 0;
2068     else if (strcmp(up_or_down, "down") == 0)
2069         vc->link_down = 1;
2070     else
2071         term_printf("invalid link status '%s'; only 'up' or 'down' valid\n",
2072                     up_or_down);
2073
2074     if (vc->link_status_changed)
2075         vc->link_status_changed(vc);
2076
2077     return 1;
2078 }
2079
2080 void net_cleanup(void)
2081 {
2082     VLANState *vlan;
2083
2084 #if !defined(_WIN32)
2085     /* close network clients */
2086     for(vlan = first_vlan; vlan != NULL; vlan = vlan->next) {
2087         VLANClientState *vc;
2088
2089         for(vc = vlan->first_client; vc != NULL; vc = vc->next) {
2090             if (vc->fd_read == tap_receive) {
2091                 TAPState *s = vc->opaque;
2092
2093                 if (s->down_script[0])
2094                     launch_script(s->down_script, s->down_script_arg, s->fd);
2095             }
2096 #if defined(CONFIG_VDE)
2097             if (vc->fd_read == vde_from_qemu) {
2098                 VDEState *s = vc->opaque;
2099                 vde_close(s->vde);
2100             }
2101 #endif
2102         }
2103     }
2104 #endif
2105 }
2106
2107 void net_client_check(void)
2108 {
2109     VLANState *vlan;
2110
2111     for(vlan = first_vlan; vlan != NULL; vlan = vlan->next) {
2112         if (vlan->nb_guest_devs == 0 && vlan->nb_host_devs == 0)
2113             continue;
2114         if (vlan->nb_guest_devs == 0)
2115             fprintf(stderr, "Warning: vlan %d with no nics\n", vlan->id);
2116         if (vlan->nb_host_devs == 0)
2117             fprintf(stderr,
2118                     "Warning: vlan %d is not connected to host network\n",
2119                     vlan->id);
2120     }
2121 }