sys/platform/vkernel64/platform/init.c

   1 /*
   2  * Copyright (c) 2006 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Matthew Dillon <dillon@backplane.com>
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  *
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in
  15  *    the documentation and/or other materials provided with the
  16  *    distribution.
  17  * 3. Neither the name of The DragonFly Project nor the names of its
  18  *    contributors may be used to endorse or promote products derived
  19  *    from this software without specific, prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  */
  34
  35 #include <sys/types.h>
  36 #include <sys/systm.h>
  37 #include <sys/kernel.h>
  38 #include <sys/stat.h>
  39 #include <sys/mman.h>
  40 #include <sys/cons.h>
  41 #include <sys/random.h>
  42 #include <sys/vkernel.h>
  43 #include <sys/tls.h>
  44 #include <sys/reboot.h>
  45 #include <sys/proc.h>
  46 #include <sys/msgbuf.h>
  47 #include <sys/vmspace.h>
  48 #include <sys/socket.h>
  49 #include <sys/sockio.h>
  50 #include <sys/sysctl.h>
  51 #include <sys/un.h>
  52 #include <vm/vm_page.h>
  53 #include <vm/vm_map.h>
  54 #include <sys/mplock2.h>
  55 #include <sys/wait.h>
  56 #include <sys/vmm.h>
  57
  58 #include <machine/cpu.h>
  59 #include <machine/globaldata.h>
  60 #include <machine/tls.h>
  61 #include <machine/md_var.h>
  62 #include <machine/vmparam.h>
  63 #include <cpu/specialreg.h>
  64
  65 #include <net/if.h>
  66 #include <net/if_arp.h>
  67 #include <net/ethernet.h>
  68 #include <net/bridge/if_bridgevar.h>
  69 #include <netinet/in.h>
  70 #include <arpa/inet.h>
  71 #include <net/if_var.h>
  72
  73 #include <stdio.h>
  74 #include <stdlib.h>
  75 #include <stdarg.h>
  76 #include <stdbool.h>
  77 #include <unistd.h>
  78 #include <fcntl.h>
  79 #include <string.h>
  80 #include <err.h>
  81 #include <errno.h>
  82 #include <assert.h>
  83 #include <sysexits.h>
  84
  85 #define EX_VKERNEL_REBOOT       32
  86
  87 vm_phystable_t phys_avail[16];
  88 vm_paddr_t Maxmem;
  89 vm_paddr_t Maxmem_bytes;
  90 long physmem;
  91 int MemImageFd = -1;
  92 struct vkdisk_info DiskInfo[VKDISK_MAX];
  93 int DiskNum;
  94 struct vknetif_info NetifInfo[VKNETIF_MAX];
  95 int NetifNum;
  96 char *pid_file;
  97 vm_offset_t KvaStart;
  98 vm_offset_t KvaEnd;
  99 vm_offset_t KvaSize;
 100 vm_offset_t virtual_start;
 101 vm_offset_t virtual_end;
 102 vm_offset_t virtual2_start;
 103 vm_offset_t virtual2_end;
 104 vm_offset_t kernel_vm_end;
 105 vm_offset_t crashdumpmap;
 106 vm_offset_t clean_sva;
 107 vm_offset_t clean_eva;
 108 struct msgbuf *msgbufp;
 109 caddr_t ptvmmap;
 110 vpte_t  *KernelPTD;
 111 vpte_t  *KernelPTA;     /* Warning: Offset for direct VA translation */
 112 void *dmap_min_address;
 113 void *vkernel_stack;
 114 u_int cpu_feature;      /* XXX */
 115 int tsc_present;
 116 int tsc_invariant;
 117 int tsc_mpsync;
 118 int optcpus;            /* number of cpus - see mp_start() */
 119 int cpu_bits;
 120 int lwp_cpu_lock;       /* if/how to lock virtual CPUs to real CPUs */
 121 int real_ncpus;         /* number of real CPUs */
 122 int next_cpu;           /* next real CPU to lock a virtual CPU to */
 123 int vkernel_b_arg;      /* no of logical CPU bits - only SMP */
 124 int vkernel_B_arg;      /* no of core bits - only SMP */
 125 int vmm_enabled;        /* VMM HW assisted enable */
 126 int use_precise_timer = 0;      /* use a precise timer (more expensive) */
 127 int allow_tsc_timer = 1;        /* use the TSC cpu timer if possible */
 128 struct privatespace *CPU_prvspace;
 129
 130 tsc_uclock_t tsc_frequency;
 131 tsc_uclock_t tsc_oneus_approx;
 132
 133 extern uint64_t KPML4phys;      /* phys addr of kernel level 4 */
 134
 135 static struct trapframe proc0_tf;
 136 static void *proc0paddr;
 137
 138 static void init_sys_memory(char *imageFile);
 139 static void init_kern_memory(void);
 140 static void init_kern_memory_vmm(void);
 141 static void init_globaldata(void);
 142 static void init_vkernel(void);
 143 static void init_disk(char **diskExp, int *diskFlags, int diskFileNum, enum vkdisk_type type);
 144 static void init_netif(char *netifExp[], int netifFileNum);
 145 static void writepid(void);
 146 static void cleanpid(void);
 147 static int unix_connect(const char *path);
 148 static void usage_err(const char *ctl, ...) __printflike(1, 2);
 149 static void usage_help(_Bool);
 150 static void init_locks(void);
 151 static void handle_term(int);
 152
 153 pid_t childpid;
 154
 155 static int save_ac;
 156 static int prezeromem;
 157 static char **save_av;
 158
 159 /*
 160  * Kernel startup for virtual kernels - standard main()
 161  */
 162 int
 163 main(int ac, char **av)
 164 {
 165         char *memImageFile = NULL;
 166         char *netifFile[VKNETIF_MAX];
 167         char *diskFile[VKDISK_MAX];
 168         char *cdFile[VKDISK_MAX];
 169         char *suffix;
 170         char *endp;
 171         char *tmp;
 172         char *tok;
 173         int diskFlags[VKDISK_MAX];
 174         int netifFileNum = 0;
 175         int diskFileNum = 0;
 176         int cdFileNum = 0;
 177         int bootOnDisk = -1;    /* set below to vcd (0) or vkd (1) */
 178         int c;
 179         int i;
 180         int j;
 181         int n;
 182         int isq;
 183         int pos;
 184         int eflag;
 185         int dflag = 0;          /* disable vmm */
 186         int real_vkernel_enable;
 187         int supports_sse;
 188         uint32_t mxcsr_mask;
 189         size_t vsize;
 190         size_t msize;
 191         size_t kenv_size;
 192         size_t kenv_size2;
 193         int status;
 194         struct sigaction sa;
 195
 196         /*
 197          * Currently a bad hack but rtld-elf needs LD_SHAREDLIB_BASE to
 198          * be set to force it to mmap() shared libraries into low memory,
 199          * so our module loader can link against the related symbols.
 200          */
 201         if (getenv("LD_SHAREDLIB_BASE") == NULL) {
 202                 setenv("LD_SHAREDLIB_BASE", "0x10000000", 1);
 203                 execv(av[0], av);
 204                 fprintf(stderr, "Must run %s with full path\n", av[0]);
 205                 exit(1);
 206         }
 207
 208         while ((childpid = fork()) != 0) {
 209                 /* Ignore signals */
 210                 bzero(&sa, sizeof(sa));
 211                 sigemptyset(&sa.sa_mask);
 212                 sa.sa_handler = SIG_IGN;
 213                 sigaction(SIGINT, &sa, NULL);
 214                 sigaction(SIGQUIT, &sa, NULL);
 215                 sigaction(SIGHUP, &sa, NULL);
 216
 217                 /*
 218                  * Forward SIGTERM to the child so that
 219                  * the shutdown process initiates correctly.
 220                  */
 221                 sa.sa_handler = handle_term;
 222                 sigaction(SIGTERM, &sa, NULL);
 223
 224                 /*
 225                  * Wait for child to terminate, exit if
 226                  * someone stole our child.
 227                  */
 228                 while (waitpid(childpid, &status, 0) != childpid) {
 229                         if (errno == ECHILD)
 230                                 exit(1);
 231                 }
 232                 if (WEXITSTATUS(status) != EX_VKERNEL_REBOOT)
 233                         return 0;
 234         }
 235
 236         /*
 237          * Starting for real
 238          */
 239         save_ac = ac;
 240         save_av = av;
 241         eflag = 0;
 242         pos = 0;
 243         kenv_size = 0;
 244
 245         /*
 246          * Process options
 247          */
 248         kernel_mem_readonly = 1;
 249         optcpus = 2;
 250         cpu_bits = 1;
 251         vkernel_b_arg = 0;
 252         vkernel_B_arg = 0;
 253         lwp_cpu_lock = LCL_NONE;
 254
 255         real_vkernel_enable = 0;
 256         vsize = sizeof(real_vkernel_enable);
 257         sysctlbyname("vm.vkernel_enable", &real_vkernel_enable, &vsize, NULL,0);
 258
 259         if (real_vkernel_enable == 0) {
 260                 errx(1, "vm.vkernel_enable is 0, must be set "
 261                         "to 1 to execute a vkernel!");
 262         }
 263
 264         real_ncpus = 1;
 265         vsize = sizeof(real_ncpus);
 266         sysctlbyname("hw.ncpu", &real_ncpus, &vsize, NULL, 0);
 267
 268         if (ac < 2)
 269                 usage_help(false);
 270
 271         while ((c = getopt(ac, av, "c:hsvztTl:m:n:r:R:e:i:p:I:Ud")) != -1) {
 272                 switch(c) {
 273                 case 'd':
 274                         dflag = 1;
 275                         break;
 276                 case 'e':
 277                         /*
 278                          * name=value:name=value:name=value...
 279                          * name="value"...
 280                          *
 281                          * Allow values to be quoted but note that shells
 282                          * may remove the quotes, so using this feature
 283                          * to embed colons may require a backslash.
 284                          */
 285                         n = strlen(optarg);
 286                         isq = 0;
 287
 288                         if (eflag == 0) {
 289                                 kenv_size = n + 2;
 290                                 kern_envp = malloc(kenv_size);
 291                                 if (kern_envp == NULL)
 292                                         errx(1, "Couldn't allocate %zd bytes for kern_envp", kenv_size);
 293                         } else {
 294                                 kenv_size2 = kenv_size + n + 1;
 295                                 pos = kenv_size - 1;
 296                                 if ((tmp = realloc(kern_envp, kenv_size2)) == NULL)
 297                                         errx(1, "Couldn't reallocate %zd bytes for kern_envp", kenv_size2);
 298                                 kern_envp = tmp;
 299                                 kenv_size = kenv_size2;
 300                         }
 301
 302                         for (i = 0, j = pos; i < n; ++i) {
 303                                 if (optarg[i] == '"')
 304                                         isq ^= 1;
 305                                 else if (optarg[i] == '\'')
 306                                         isq ^= 2;
 307                                 else if (isq == 0 && optarg[i] == ':')
 308                                         kern_envp[j++] = 0;
 309                                 else
 310                                         kern_envp[j++] = optarg[i];
 311                         }
 312                         kern_envp[j++] = 0;
 313                         kern_envp[j++] = 0;
 314                         eflag++;
 315                         break;
 316                 case 's':
 317                         boothowto |= RB_SINGLE;
 318                         break;
 319                 case 't':
 320                         use_precise_timer = 1;
 321                         break;
 322                 case 'T':
 323                         allow_tsc_timer = 0;
 324                         break;
 325                 case 'v':
 326                         bootverbose = 1;
 327                         break;
 328                 case 'i':
 329                         memImageFile = optarg;
 330                         break;
 331                 case 'I':
 332                         if (netifFileNum < VKNETIF_MAX)
 333                                 netifFile[netifFileNum++] = strdup(optarg);
 334                         break;
 335                 case 'r':
 336                 case 'R':
 337                         if (bootOnDisk < 0)
 338                                 bootOnDisk = 1;
 339                         if (diskFileNum + cdFileNum < VKDISK_MAX) {
 340                                 diskFile[diskFileNum] = strdup(optarg);
 341                                 diskFlags[diskFileNum] = (c == 'R');
 342                                 ++diskFileNum;
 343                         }
 344                         break;
 345                 case 'c':
 346                         if (bootOnDisk < 0)
 347                                 bootOnDisk = 0;
 348                         if (diskFileNum + cdFileNum < VKDISK_MAX)
 349                                 cdFile[cdFileNum++] = strdup(optarg);
 350                         break;
 351                 case 'm':
 352                         Maxmem_bytes = strtoull(optarg, &suffix, 0);
 353                         if (suffix) {
 354                                 switch(*suffix) {
 355                                 case 'g':
 356                                 case 'G':
 357                                         Maxmem_bytes <<= 30;
 358                                         break;
 359                                 case 'm':
 360                                 case 'M':
 361                                         Maxmem_bytes <<= 20;
 362                                         break;
 363                                 case 'k':
 364                                 case 'K':
 365                                         Maxmem_bytes <<= 10;
 366                                         break;
 367                                 default:
 368                                         Maxmem_bytes = 0;
 369                                         usage_err("Bad maxmem option");
 370                                         /* NOT REACHED */
 371                                         break;
 372                                 }
 373                         }
 374                         break;
 375                 case 'l':
 376                         next_cpu = -1;
 377                         if (strncmp("map", optarg, 3) == 0) {
 378                                 lwp_cpu_lock = LCL_PER_CPU;
 379                                 if (optarg[3] == ',') {
 380                                         next_cpu = strtol(optarg+4, &endp, 0);
 381                                         if (*endp != '\0')
 382                                                 usage_err("Bad target CPU number at '%s'", endp);
 383                                 } else {
 384                                         next_cpu = 0;
 385                                 }
 386                                 if (next_cpu < 0 || next_cpu > real_ncpus - 1)
 387                                         usage_err("Bad target CPU, valid range is 0-%d", real_ncpus - 1);
 388                         } else if (strncmp("any", optarg, 3) == 0) {
 389                                 lwp_cpu_lock = LCL_NONE;
 390                         } else {
 391                                 lwp_cpu_lock = LCL_SINGLE_CPU;
 392                                 next_cpu = strtol(optarg, &endp, 0);
 393                                 if (*endp != '\0')
 394                                         usage_err("Bad target CPU number at '%s'", endp);
 395                                 if (next_cpu < 0 || next_cpu > real_ncpus - 1)
 396                                         usage_err("Bad target CPU, valid range is 0-%d", real_ncpus - 1);
 397                         }
 398                         break;
 399                 case 'n':
 400                         /*
 401                          * This value is set up by mp_start(), don't just
 402                          * set ncpus here.
 403                          */
 404                         tok = strtok(optarg, ":");
 405                         optcpus = strtol(tok, NULL, 0);
 406                         if (optcpus < 1 || optcpus > MAXCPU)
 407                                 usage_err("Bad ncpus, valid range is 1-%d", MAXCPU);
 408                         cpu_bits = 1;
 409                         while ((1 << cpu_bits) < optcpus)
 410                                 ++cpu_bits;
 411
 412                         /*
 413                          * By default assume simple hyper-threading
 414                          */
 415                         vkernel_b_arg = 1;
 416                         vkernel_B_arg = cpu_bits - vkernel_b_arg;
 417
 418                         /*
 419                          * [:lbits[:cbits]] override # of cpu bits
 420                          * for logical and core extraction, supplying
 421                          * defaults for any omission.
 422                          */
 423                         tok = strtok(NULL, ":");
 424                         if (tok != NULL) {
 425                                 vkernel_b_arg = strtol(tok, NULL, 0);
 426                                 vkernel_B_arg = cpu_bits - vkernel_b_arg;
 427
 428                                 /* :cbits argument */
 429                                 tok = strtok(NULL, ":");
 430                                 if (tok != NULL) {
 431                                         vkernel_B_arg = strtol(tok, NULL, 0);
 432                                 }
 433                         }
 434                         break;
 435                 case 'p':
 436                         pid_file = optarg;
 437                         break;
 438                 case 'U':
 439                         kernel_mem_readonly = 0;
 440                         break;
 441                 case 'h':
 442                         usage_help(true);
 443                         break;
 444                 case 'z':
 445                         prezeromem = 1;
 446                         break;
 447                 default:
 448                         usage_help(false);
 449                 }
 450         }
 451
 452         /*
 453          * Check VMM presence
 454          */
 455         vsize = sizeof(vmm_enabled);
 456         sysctlbyname("hw.vmm.enable", &vmm_enabled, &vsize, NULL, 0);
 457         vmm_enabled = (vmm_enabled && !dflag);
 458
 459         writepid();
 460         cpu_disable_intr();
 461         if (vmm_enabled) {
 462                 /* use a MAP_ANON directly */
 463                 printf("VMM is available\n");
 464                 init_kern_memory_vmm();
 465         } else {
 466                 printf("VMM is not available\n");
 467                 init_sys_memory(memImageFile);
 468                 init_kern_memory();
 469         }
 470         init_globaldata();
 471         init_vkernel();
 472         setrealcpu();
 473         init_kqueue();
 474
 475         vmm_guest = VMM_GUEST_VKERNEL;
 476
 477         /*
 478          * Check TSC
 479          */
 480         vsize = sizeof(tsc_present);
 481         sysctlbyname("hw.tsc_present", &tsc_present, &vsize, NULL, 0);
 482         vsize = sizeof(tsc_invariant);
 483         sysctlbyname("hw.tsc_invariant", &tsc_invariant, &vsize, NULL, 0);
 484         vsize = sizeof(tsc_mpsync);
 485         sysctlbyname("hw.tsc_mpsync", &tsc_mpsync, &vsize, NULL, 0);
 486         vsize = sizeof(tsc_frequency);
 487         sysctlbyname("hw.tsc_frequency", &tsc_frequency, &vsize, NULL, 0);
 488         if (tsc_present)
 489                 cpu_feature |= CPUID_TSC;
 490         tsc_oneus_approx = ((tsc_frequency|1) + 999999) / 1000000;
 491
 492         /*
 493          * Check SSE
 494          */
 495         vsize = sizeof(supports_sse);
 496         supports_sse = 0;
 497         sysctlbyname("hw.instruction_sse", &supports_sse, &vsize, NULL, 0);
 498         sysctlbyname("hw.mxcsr_mask", &mxcsr_mask, &msize, NULL, 0);
 499         init_fpu(supports_sse);
 500         if (supports_sse)
 501                 cpu_feature |= CPUID_SSE | CPUID_FXSR;
 502
 503         /*
 504          * We boot from the first installed disk.
 505          */
 506         if (bootOnDisk == 1) {
 507                 init_disk(diskFile, diskFlags, diskFileNum, VKD_DISK);
 508                 init_disk(cdFile, NULL, cdFileNum, VKD_CD);
 509         } else {
 510                 init_disk(cdFile, NULL, cdFileNum, VKD_CD);
 511                 init_disk(diskFile, diskFlags, diskFileNum, VKD_DISK);
 512         }
 513
 514         init_netif(netifFile, netifFileNum);
 515         init_exceptions();
 516         mi_startup();
 517         /* NOT REACHED */
 518         exit(EX_SOFTWARE);
 519 }
 520
 521 /* SIGTERM handler */
 522 static
 523 void
 524 handle_term(int sig)
 525 {
 526         kill(childpid, sig);
 527 }
 528
 529 /*
 530  * Initialize system memory.  This is the virtual kernel's 'RAM'.
 531  */
 532 static
 533 void
 534 init_sys_memory(char *imageFile)
 535 {
 536         struct stat st;
 537         int i;
 538         int fd;
 539
 540         /*
 541          * Figure out the system memory image size.  If an image file was
 542          * specified and -m was not specified, use the image file's size.
 543          */
 544         if (imageFile && stat(imageFile, &st) == 0 && Maxmem_bytes == 0)
 545                 Maxmem_bytes = (vm_paddr_t)st.st_size;
 546         if ((imageFile == NULL || stat(imageFile, &st) < 0) &&
 547             Maxmem_bytes == 0) {
 548                 errx(1, "Cannot create new memory file %s unless "
 549                        "system memory size is specified with -m",
 550                        imageFile);
 551                 /* NOT REACHED */
 552         }
 553
 554         /*
 555          * Maxmem must be known at this time
 556          */
 557         if (Maxmem_bytes < 64 * 1024 * 1024 || (Maxmem_bytes & SEG_MASK)) {
 558                 errx(1, "Bad maxmem specification: 64MB minimum, "
 559                        "multiples of %dMB only",
 560                        SEG_SIZE / 1024 / 1024);
 561                 /* NOT REACHED */
 562         }
 563
 564         /*
 565          * Generate an image file name if necessary, then open/create the
 566          * file exclusively locked.  Do not allow multiple virtual kernels
 567          * to use the same image file.
 568          *
 569          * Don't iterate through a million files if we do not have write
 570          * access to the directory, stop if our open() failed on a
 571          * non-existant file.  Otherwise opens can fail for any number
 572          */
 573         if (imageFile == NULL) {
 574                 for (i = 0; i < 1000000; ++i) {
 575                         asprintf(&imageFile, "/var/vkernel/memimg.%06d", i);
 576                         fd = open(imageFile,
 577                                   O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644);
 578                         if (fd < 0 && stat(imageFile, &st) == 0) {
 579                                 free(imageFile);
 580                                 continue;
 581                         }
 582                         break;
 583                 }
 584         } else {
 585                 fd = open(imageFile, O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644);
 586         }
 587         fprintf(stderr, "Using memory file: %s\n", imageFile);
 588         if (fd < 0 || fstat(fd, &st) < 0) {
 589                 err(1, "Unable to open/create %s", imageFile);
 590                 /* NOT REACHED */
 591         }
 592
 593         /*
 594          * Truncate or extend the file as necessary.  Clean out the contents
 595          * of the file, we want it to be full of holes so we don't waste
 596          * time reading in data from an old file that we no longer care
 597          * about.
 598          */
 599         ftruncate(fd, 0);
 600         ftruncate(fd, Maxmem_bytes);
 601
 602         MemImageFd = fd;
 603         Maxmem = Maxmem_bytes >> PAGE_SHIFT;
 604         physmem = Maxmem;
 605 }
 606
 607 /*
 608  * Initialize kernel memory.  This reserves kernel virtual memory by using
 609  * MAP_VPAGETABLE
 610  */
 611
 612 static
 613 void
 614 init_kern_memory(void)
 615 {
 616         void *base;
 617         int i;
 618         void *firstfree;
 619
 620         /*
 621          * Memory map our kernel virtual memory space.  Note that the
 622          * kernel image itself is not made part of this memory for the
 623          * moment.
 624          *
 625          * The memory map must be segment-aligned so we can properly
 626          * offset KernelPTD.
 627          *
 628          * If the system kernel has a different MAXDSIZ, it might not
 629          * be possible to map kernel memory in its prefered location.
 630          * Try a number of different locations.
 631          */
 632
 633         base = mmap((void*)KERNEL_KVA_START, KERNEL_KVA_SIZE,
 634                     PROT_READ|PROT_WRITE|PROT_EXEC,
 635                     MAP_FILE|MAP_SHARED|MAP_VPAGETABLE|MAP_FIXED|MAP_TRYFIXED,
 636                     MemImageFd, (off_t)KERNEL_KVA_START);
 637
 638         if (base == MAP_FAILED) {
 639                 err(1, "Unable to mmap() kernel virtual memory!");
 640                 /* NOT REACHED */
 641         }
 642         madvise(base, KERNEL_KVA_SIZE, MADV_NOSYNC);
 643         KvaStart = (vm_offset_t)base;
 644         KvaSize = KERNEL_KVA_SIZE;
 645         KvaEnd = KvaStart + KvaSize;
 646
 647         /* cannot use kprintf yet */
 648         printf("KVM mapped at %p-%p\n", (void *)KvaStart, (void *)KvaEnd);
 649
 650         /* MAP_FILE? */
 651         dmap_min_address = mmap(0, DMAP_SIZE, PROT_READ|PROT_WRITE,
 652                                 MAP_NOCORE|MAP_NOSYNC|MAP_SHARED,
 653                                 MemImageFd, 0);
 654         if (dmap_min_address == MAP_FAILED) {
 655                 err(1, "Unable to mmap() kernel DMAP region!");
 656                 /* NOT REACHED */
 657         }
 658
 659         /*
 660          * Prefault the memory.  The vkernel is going to fault it all in
 661          * anyway, and faults on the backing store itself are very expensive
 662          * once we go SMP (contend a lot).  So do it now.
 663          */
 664         if (prezeromem)
 665                 bzero(dmap_min_address, Maxmem_bytes);
 666
 667         /*
 668          * Bootstrap the kernel_pmap
 669          */
 670         firstfree = NULL;
 671         pmap_bootstrap((vm_paddr_t *)&firstfree, (int64_t)base);
 672
 673         mcontrol(base, KERNEL_KVA_SIZE, MADV_SETMAP,
 674                  0 | VPTE_RW | VPTE_V);
 675
 676         /*
 677          * phys_avail[] represents unallocated physical memory.  MI code
 678          * will use phys_avail[] to create the vm_page array.
 679          */
 680         phys_avail[0].phys_beg = (vm_paddr_t)firstfree;
 681         phys_avail[0].phys_beg = (phys_avail[0].phys_beg + PAGE_MASK) &
 682                                  ~(vm_paddr_t)PAGE_MASK;
 683         phys_avail[0].phys_end = Maxmem_bytes;
 684
 685 #if 0 /* JGV */
 686         /*
 687          * (virtual_start, virtual_end) represent unallocated kernel virtual
 688          * memory.  MI code will create kernel_map using these parameters.
 689          */
 690         virtual_start = KvaStart + (long)firstfree;
 691         virtual_start = (virtual_start + PAGE_MASK) & ~(vm_offset_t)PAGE_MASK;
 692         virtual_end = KvaStart + KERNEL_KVA_SIZE;
 693 #endif
 694
 695         /*
 696          * pmap_growkernel() will set the correct value.
 697          */
 698         kernel_vm_end = 0;
 699
 700         /*
 701          * Allocate space for process 0's UAREA.
 702          */
 703         proc0paddr = (void *)virtual_start;
 704         for (i = 0; i < UPAGES; ++i) {
 705                 pmap_kenter_quick(virtual_start, phys_avail[0].phys_beg);
 706                 virtual_start += PAGE_SIZE;
 707                 phys_avail[0].phys_beg += PAGE_SIZE;
 708         }
 709
 710         /*
 711          * crashdumpmap
 712          */
 713         crashdumpmap = virtual_start;
 714         virtual_start += MAXDUMPPGS * PAGE_SIZE;
 715
 716         /*
 717          * msgbufp maps the system message buffer
 718          */
 719         assert((MSGBUF_SIZE & PAGE_MASK) == 0);
 720         msgbufp = (void *)virtual_start;
 721         for (i = 0; i < (MSGBUF_SIZE >> PAGE_SHIFT); ++i) {
 722                 pmap_kenter_quick(virtual_start, phys_avail[0].phys_beg);
 723                 virtual_start += PAGE_SIZE;
 724                 phys_avail[0].phys_beg += PAGE_SIZE;
 725         }
 726         msgbufinit(msgbufp, MSGBUF_SIZE);
 727
 728         /*
 729          * used by kern_memio for /dev/mem access
 730          */
 731         ptvmmap = (caddr_t)virtual_start;
 732         virtual_start += PAGE_SIZE;
 733 }
 734
 735 static
 736 void
 737 init_kern_memory_vmm(void)
 738 {
 739         int i;
 740         void *firstfree;
 741         struct vmm_guest_options options;
 742         void *dmap_address;
 743
 744         KvaStart = (vm_offset_t)KERNEL_KVA_START;
 745         KvaSize = KERNEL_KVA_SIZE;
 746         KvaEnd = KvaStart + KvaSize;
 747
 748         Maxmem = Maxmem_bytes >> PAGE_SHIFT;
 749         physmem = Maxmem;
 750
 751         if (Maxmem_bytes < 64 * 1024 * 1024 || (Maxmem_bytes & SEG_MASK)) {
 752                 errx(1, "Bad maxmem specification: 64MB minimum, "
 753                        "multiples of %dMB only",
 754                        SEG_SIZE / 1024 / 1024);
 755                 /* NOT REACHED */
 756         }
 757
 758         /* Call the vmspace_create to allocate the internal
 759          * vkernel structures. Won't do anything else (no new
 760          * vmspace)
 761          */
 762         if (vmspace_create(NULL, 0, NULL) < 0)
 763                 panic("vmspace_create() failed");
 764
 765
 766         /*
 767          * MAP_ANON the region of the VKERNEL phyisical memory
 768          * (known as GPA - Guest Physical Address
 769          */
 770         dmap_address = mmap(NULL, Maxmem_bytes,
 771                             PROT_READ|PROT_WRITE|PROT_EXEC,
 772                             MAP_ANON|MAP_SHARED, -1, 0);
 773         if (dmap_address == MAP_FAILED) {
 774                 err(1, "Unable to mmap() RAM region!");
 775                 /* NOT REACHED */
 776         }
 777         if (prezeromem)
 778                 bzero(dmap_address, Maxmem_bytes);
 779
 780         /* Alloc a new stack in the lowmem */
 781         vkernel_stack = mmap(NULL, KERNEL_STACK_SIZE,
 782                              PROT_READ|PROT_WRITE|PROT_EXEC,
 783                              MAP_ANON, -1, 0);
 784         if (vkernel_stack == MAP_FAILED) {
 785                 err(1, "Unable to allocate stack\n");
 786         }
 787
 788         /*
 789          * Bootstrap the kernel_pmap
 790          */
 791         firstfree = dmap_address;
 792         dmap_min_address = NULL; /* VIRT == PHYS in the first 512G */
 793         pmap_bootstrap((vm_paddr_t *)&firstfree, (uint64_t)KvaStart);
 794
 795         /*
 796          * Enter VMM mode
 797          */
 798         bzero(&options, sizeof(options));
 799         options.guest_cr3 = (register_t) KPML4phys;
 800         options.new_stack = (uint64_t) vkernel_stack + KERNEL_STACK_SIZE;
 801         options.master = 1;
 802         if (vmm_guest_ctl(VMM_GUEST_RUN, &options)) {
 803                 err(1, "Unable to enter VMM mode.");
 804         }
 805
 806         /*
 807          * phys_avail[] represents unallocated physical memory.  MI code
 808          * will use phys_avail[] to create the vm_page array.
 809          */
 810         phys_avail[0].phys_beg = (vm_paddr_t)firstfree;
 811         phys_avail[0].phys_beg = (phys_avail[0].phys_beg + PAGE_MASK) &
 812                                  ~(vm_paddr_t)PAGE_MASK;
 813         phys_avail[0].phys_end = (vm_paddr_t)dmap_address + Maxmem_bytes;
 814
 815         /*
 816          * pmap_growkernel() will set the correct value.
 817          */
 818         kernel_vm_end = 0;
 819
 820         /*
 821          * Allocate space for process 0's UAREA.
 822          */
 823         proc0paddr = (void *)virtual_start;
 824         for (i = 0; i < UPAGES; ++i) {
 825                 pmap_kenter_quick(virtual_start, phys_avail[0].phys_beg);
 826                 virtual_start += PAGE_SIZE;
 827                 phys_avail[0].phys_beg += PAGE_SIZE;
 828         }
 829
 830         /*
 831          * crashdumpmap
 832          */
 833         crashdumpmap = virtual_start;
 834         virtual_start += MAXDUMPPGS * PAGE_SIZE;
 835
 836         /*
 837          * msgbufp maps the system message buffer
 838          */
 839         assert((MSGBUF_SIZE & PAGE_MASK) == 0);
 840         msgbufp = (void *)virtual_start;
 841         for (i = 0; i < (MSGBUF_SIZE >> PAGE_SHIFT); ++i) {
 842
 843                 pmap_kenter_quick(virtual_start, phys_avail[0].phys_beg);
 844                 virtual_start += PAGE_SIZE;
 845                 phys_avail[0].phys_beg += PAGE_SIZE;
 846         }
 847
 848         msgbufinit(msgbufp, MSGBUF_SIZE);
 849
 850         /*
 851          * used by kern_memio for /dev/mem access
 852          */
 853         ptvmmap = (caddr_t)virtual_start;
 854         virtual_start += PAGE_SIZE;
 855
 856         printf("vmm: Hardware pagetable enabled for guest\n");
 857 }
 858
 859
 860 /*
 861  * Map the per-cpu globaldata for cpu #0.  Allocate the space using
 862  * virtual_start and phys_avail[0]
 863  */
 864 static
 865 void
 866 init_globaldata(void)
 867 {
 868         int i;
 869         vm_paddr_t pa;
 870         vm_offset_t va;
 871
 872         /*
 873          * Reserve enough KVA to cover possible cpus.  This is a considerable
 874          * amount of KVA since the privatespace structure includes two
 875          * whole page table mappings.
 876          */
 877         virtual_start = (virtual_start + SEG_MASK) & ~(vm_offset_t)SEG_MASK;
 878         CPU_prvspace = (void *)virtual_start;
 879         virtual_start += sizeof(struct privatespace) * SMP_MAXCPU;
 880
 881         /*
 882          * Allocate enough physical memory to cover the mdglobaldata
 883          * portion of the space and the idle stack and map the pages
 884          * into KVA.  For cpu #0 only.
 885          */
 886         for (i = 0; i < sizeof(struct mdglobaldata); i += PAGE_SIZE) {
 887                 pa = phys_avail[0].phys_beg;
 888                 va = (vm_offset_t)&CPU_prvspace[0].mdglobaldata + i;
 889                 pmap_kenter_quick(va, pa);
 890                 phys_avail[0].phys_beg += PAGE_SIZE;
 891         }
 892         for (i = 0; i < sizeof(CPU_prvspace[0].idlestack); i += PAGE_SIZE) {
 893                 pa = phys_avail[0].phys_beg;
 894                 va = (vm_offset_t)&CPU_prvspace[0].idlestack + i;
 895                 pmap_kenter_quick(va, pa);
 896                 phys_avail[0].phys_beg += PAGE_SIZE;
 897         }
 898
 899         /*
 900          * Setup the %gs for cpu #0.  The mycpu macro works after this
 901          * point.  Note that %fs is used by pthreads.
 902          */
 903         tls_set_gs(&CPU_prvspace[0], sizeof(struct privatespace));
 904 }
 905
 906
 907 /*
 908  * Initialize pool tokens and other necessary locks
 909  */
 910 static void
 911 init_locks(void)
 912 {
 913
 914         /*
 915          * Get the initial mplock with a count of 1 for the BSP.
 916          * This uses a LOGICAL cpu ID, ie BSP == 0.
 917          */
 918         cpu_get_initial_mplock();
 919
 920         /* our token pool needs to work early */
 921         lwkt_token_pool_init();
 922
 923 }
 924
 925
 926 /*
 927  * Initialize very low level systems including thread0, proc0, etc.
 928  */
 929 static
 930 void
 931 init_vkernel(void)
 932 {
 933         struct mdglobaldata *gd;
 934
 935         gd = &CPU_prvspace[0].mdglobaldata;
 936         bzero(gd, sizeof(*gd));
 937
 938         gd->mi.gd_curthread = &thread0;
 939         thread0.td_gd = &gd->mi;
 940         ncpus = 1;
 941         ncpus_fit = 1;  /* rounded up power of 2 */
 942         /* ncpus_fit_mask are 0 */
 943         init_param1();
 944         gd->mi.gd_prvspace = &CPU_prvspace[0];
 945         mi_gdinit(&gd->mi, 0);
 946         cpu_gdinit(gd, 0);
 947         mi_proc0init(&gd->mi, proc0paddr);
 948         lwp0.lwp_md.md_regs = &proc0_tf;
 949
 950         init_locks();
 951         cninit();
 952         rand_initialize();
 953 #if 0   /* #ifdef DDB */
 954         kdb_init();
 955         if (boothowto & RB_KDB)
 956                 Debugger("Boot flags requested debugger");
 957 #endif
 958         identcpu();
 959 #if 0
 960         initializecpu();        /* Initialize CPU registers */
 961 #endif
 962         init_param2((phys_avail[0].phys_end -
 963                      phys_avail[0].phys_beg) / PAGE_SIZE);
 964
 965 #if 0
 966         /*
 967          * Map the message buffer
 968          */
 969         for (off = 0; off < round_page(MSGBUF_SIZE); off += PAGE_SIZE)
 970                 pmap_kenter((vm_offset_t)msgbufp + off, avail_end + off);
 971         msgbufinit(msgbufp, MSGBUF_SIZE);
 972 #endif
 973 #if 0
 974         thread0.td_pcb_cr3 ... MMU
 975         lwp0.lwp_md.md_regs = &proc0_tf;
 976 #endif
 977 }
 978
 979 /*
 980  * Filesystem image paths for the virtual kernel are optional.
 981  * If specified they each should point to a disk image,
 982  * the first of which will become the root disk.
 983  *
 984  * The virtual kernel caches data from our 'disk' just like a normal kernel,
 985  * so we do not really want the real kernel to cache the data too.  Use
 986  * O_DIRECT to remove the duplication.
 987  */
 988 static
 989 void
 990 init_disk(char **diskExp, int *diskFlags, int diskFileNum, enum vkdisk_type type)
 991 {
 992         char *serno;
 993         int i;
 994
 995         if (diskFileNum == 0)
 996                 return;
 997
 998         for (i=0; i < diskFileNum; i++){
 999                 char *fname;
1000                 fname = diskExp[i];
1001
1002                 if (fname == NULL) {
1003                         warnx("Invalid argument to '-r'");
1004                         continue;
1005                 }
1006                 /*
1007                  * Check for a serial number for the virtual disk
1008                  * passed from the command line.
1009                  */
1010                 serno = fname;
1011                 strsep(&serno, ":");
1012
1013                 if (DiskNum < VKDISK_MAX) {
1014                         struct stat st;
1015                         struct vkdisk_info *info = NULL;
1016                         int fd;
1017                         size_t l = 0;
1018
1019                         if (type == VKD_DISK)
1020                             fd = open(fname, O_RDWR|O_DIRECT, 0644);
1021                         else
1022                             fd = open(fname, O_RDONLY|O_DIRECT, 0644);
1023                         if (fd < 0 || fstat(fd, &st) < 0) {
1024                                 err(1, "Unable to open/create %s", fname);
1025                                 /* NOT REACHED */
1026                         }
1027                         if (S_ISREG(st.st_mode) && (diskFlags[i] & 1) == 0) {
1028                                 if (flock(fd, LOCK_EX|LOCK_NB) < 0) {
1029                                         errx(1, "Disk image %s is already "
1030                                                 "in use\n", fname);
1031                                         /* NOT REACHED */
1032                                 }
1033                         }
1034
1035                         info = &DiskInfo[DiskNum];
1036                         l = strlen(fname);
1037
1038                         info->unit = i;
1039                         info->fd = fd;
1040                         info->type = type;
1041                         info->flags = diskFlags[i];
1042                         memcpy(info->fname, fname, l);
1043                         info->serno = NULL;
1044                         if (serno) {
1045                                 if ((info->serno = malloc(SERNOLEN)) != NULL)
1046                                         strlcpy(info->serno, serno, SERNOLEN);
1047                                 else
1048                                         warnx("Couldn't allocate memory for the operation");
1049                         }
1050
1051                         if (DiskNum == 0) {
1052                                 if (type == VKD_CD) {
1053                                         rootdevnames[0] = "cd9660:vcd0";
1054                                 } else if (type == VKD_DISK) {
1055                                         rootdevnames[0] = "ufs:vkd0s0a";
1056                                         rootdevnames[1] = "ufs:vkd0s1a";
1057                                 }
1058                         }
1059
1060                         DiskNum++;
1061                 } else {
1062                         warnx("vkd%d (%s) > VKDISK_MAX", DiskNum, fname);
1063                         continue;
1064                 }
1065         }
1066 }
1067
1068 static
1069 int
1070 netif_set_tapflags(int tap_unit, int f, int s)
1071 {
1072         struct ifreq ifr;
1073         int flags;
1074
1075         bzero(&ifr, sizeof(ifr));
1076
1077         snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit);
1078         if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0) {
1079                 warn("tap%d: ioctl(SIOCGIFFLAGS) failed", tap_unit);
1080                 return -1;
1081         }
1082
1083         /*
1084          * Adjust if_flags
1085          *
1086          * If the flags are already set/cleared, then we return
1087          * immediately to avoid extra syscalls
1088          */
1089         flags = (ifr.ifr_flags & 0xffff) | (ifr.ifr_flagshigh << 16);
1090         if (f < 0) {
1091                 /* Turn off flags */
1092                 f = -f;
1093                 if ((flags & f) == 0)
1094                         return 0;
1095                 flags &= ~f;
1096         } else {
1097                 /* Turn on flags */
1098                 if (flags & f)
1099                         return 0;
1100                 flags |= f;
1101         }
1102
1103         /*
1104          * Fix up ifreq.ifr_name, since it may be trashed
1105          * in previous ioctl(SIOCGIFFLAGS)
1106          */
1107         snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit);
1108
1109         ifr.ifr_flags = flags & 0xffff;
1110         ifr.ifr_flagshigh = flags >> 16;
1111         if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0) {
1112                 warn("tap%d: ioctl(SIOCSIFFLAGS) failed", tap_unit);
1113                 return -1;
1114         }
1115         return 0;
1116 }
1117
1118 static
1119 int
1120 netif_set_tapaddr(int tap_unit, in_addr_t addr, in_addr_t mask, int s)
1121 {
1122         struct ifaliasreq ifra;
1123         struct sockaddr_in *in;
1124
1125         bzero(&ifra, sizeof(ifra));
1126         snprintf(ifra.ifra_name, sizeof(ifra.ifra_name), "tap%d", tap_unit);
1127
1128         /* Setup address */
1129         in = (struct sockaddr_in *)&ifra.ifra_addr;
1130         in->sin_family = AF_INET;
1131         in->sin_len = sizeof(*in);
1132         in->sin_addr.s_addr = addr;
1133
1134         if (mask != 0) {
1135                 /* Setup netmask */
1136                 in = (struct sockaddr_in *)&ifra.ifra_mask;
1137                 in->sin_len = sizeof(*in);
1138                 in->sin_addr.s_addr = mask;
1139         }
1140
1141         if (ioctl(s, SIOCAIFADDR, &ifra) < 0) {
1142                 warn("tap%d: ioctl(SIOCAIFADDR) failed", tap_unit);
1143                 return -1;
1144         }
1145         return 0;
1146 }
1147
1148 static
1149 int
1150 netif_add_tap2brg(int tap_unit, const char *ifbridge, int s)
1151 {
1152         struct ifbreq ifbr;
1153         struct ifdrv ifd;
1154
1155         bzero(&ifbr, sizeof(ifbr));
1156         snprintf(ifbr.ifbr_ifsname, sizeof(ifbr.ifbr_ifsname),
1157                  "tap%d", tap_unit);
1158
1159         bzero(&ifd, sizeof(ifd));
1160         strlcpy(ifd.ifd_name, ifbridge, sizeof(ifd.ifd_name));
1161         ifd.ifd_cmd = BRDGADD;
1162         ifd.ifd_len = sizeof(ifbr);
1163         ifd.ifd_data = &ifbr;
1164
1165         if (ioctl(s, SIOCSDRVSPEC, &ifd) < 0) {
1166                 /*
1167                  * 'errno == EEXIST' means that the tap(4) is already
1168                  * a member of the bridge(4)
1169                  */
1170                 if (errno != EEXIST) {
1171                         warn("ioctl(%s, SIOCSDRVSPEC) failed", ifbridge);
1172                         return -1;
1173                 }
1174         }
1175         return 0;
1176 }
1177
1178 #define TAPDEV_OFLAGS   (O_RDWR | O_NONBLOCK)
1179
1180 /*
1181  * Locate the first unused tap(4) device file if auto mode is requested,
1182  * or open the user supplied device file, and bring up the corresponding
1183  * tap(4) interface.
1184  *
1185  * NOTE: Only tap(4) device file is supported currently
1186  */
1187 static
1188 int
1189 netif_open_tap(const char *netif, int *tap_unit, int s)
1190 {
1191         char tap_dev[MAXPATHLEN];
1192         int tap_fd, failed;
1193         struct stat st;
1194         char *dname;
1195
1196         *tap_unit = -1;
1197
1198         if (strcmp(netif, "auto") == 0) {
1199                 /*
1200                  * Find first unused tap(4) device file
1201                  */
1202                 tap_fd = open("/dev/tap", TAPDEV_OFLAGS);
1203                 if (tap_fd < 0) {
1204                         warnc(errno, "Unable to find a free tap(4)");
1205                         return -1;
1206                 }
1207         } else {
1208                 /*
1209                  * User supplied tap(4) device file or unix socket.
1210                  */
1211                 if (netif[0] == '/')    /* Absolute path */
1212                         strlcpy(tap_dev, netif, sizeof(tap_dev));
1213                 else
1214                         snprintf(tap_dev, sizeof(tap_dev), "/dev/%s", netif);
1215
1216                 tap_fd = open(tap_dev, TAPDEV_OFLAGS);
1217
1218                 /*
1219                  * If we cannot open normally try to connect to it.
1220                  */
1221                 if (tap_fd < 0)
1222                         tap_fd = unix_connect(tap_dev);
1223
1224                 if (tap_fd < 0) {
1225                         warn("Unable to open %s", tap_dev);
1226                         return -1;
1227                 }
1228         }
1229
1230         /*
1231          * Check whether the device file is a tap(4)
1232          */
1233         if (fstat(tap_fd, &st) < 0) {
1234                 failed = 1;
1235         } else if (S_ISCHR(st.st_mode)) {
1236                 dname = fdevname(tap_fd);
1237                 if (dname)
1238                         dname = strstr(dname, "tap");
1239                 if (dname) {
1240                         /*
1241                          * Bring up the corresponding tap(4) interface
1242                          */
1243                         *tap_unit = strtol(dname + 3, NULL, 10);
1244                         printf("TAP UNIT %d\n", *tap_unit);
1245                         if (netif_set_tapflags(*tap_unit, IFF_UP, s) == 0)
1246                                 failed = 0;
1247                         else
1248                                 failed = 1;
1249                 } else {
1250                         failed = 1;
1251                 }
1252         } else if (S_ISSOCK(st.st_mode)) {
1253                 /*
1254                  * Special socket connection (typically to vknet).  We
1255                  * do not have to do anything.
1256                  */
1257                 failed = 0;
1258         } else {
1259                 failed = 1;
1260         }
1261
1262         if (failed) {
1263                 warnx("%s is not a tap(4) device or socket", tap_dev);
1264                 close(tap_fd);
1265                 tap_fd = -1;
1266                 *tap_unit = -1;
1267         }
1268         return tap_fd;
1269 }
1270
1271 static int
1272 unix_connect(const char *path)
1273 {
1274         struct sockaddr_un sunx;
1275         int len;
1276         int net_fd;
1277         int sndbuf = 262144;
1278         struct stat st;
1279
1280         snprintf(sunx.sun_path, sizeof(sunx.sun_path), "%s", path);
1281         len = offsetof(struct sockaddr_un, sun_path[strlen(sunx.sun_path)]);
1282         ++len;  /* include nul */
1283         sunx.sun_family = AF_UNIX;
1284         sunx.sun_len = len;
1285
1286         net_fd = socket(AF_UNIX, SOCK_SEQPACKET, 0);
1287         if (net_fd < 0)
1288                 return(-1);
1289         if (connect(net_fd, (void *)&sunx, len) < 0) {
1290                 close(net_fd);
1291                 return(-1);
1292         }
1293         setsockopt(net_fd, SOL_SOCKET, SO_SNDBUF, &sndbuf, sizeof(sndbuf));
1294         if (fstat(net_fd, &st) == 0)
1295                 printf("Network socket buffer: %d bytes\n", st.st_blksize);
1296         fcntl(net_fd, F_SETFL, O_NONBLOCK);
1297         return(net_fd);
1298 }
1299
1300 #undef TAPDEV_MAJOR
1301 #undef TAPDEV_MINOR
1302 #undef TAPDEV_OFLAGS
1303
1304 /*
1305  * Following syntax is supported,
1306  * 1) x.x.x.x             tap(4)'s address is x.x.x.x
1307  *
1308  * 2) x.x.x.x/z           tap(4)'s address is x.x.x.x
1309  *                        tap(4)'s netmask len is z
1310  *
1311  * 3) x.x.x.x:y.y.y.y     tap(4)'s address is x.x.x.x
1312  *                        pseudo netif's address is y.y.y.y
1313  *
1314  * 4) x.x.x.x:y.y.y.y/z   tap(4)'s address is x.x.x.x
1315  *                        pseudo netif's address is y.y.y.y
1316  *                        tap(4) and pseudo netif's netmask len are z
1317  *
1318  * 5) bridgeX             tap(4) will be added to bridgeX
1319  *
1320  * 6) bridgeX:y.y.y.y     tap(4) will be added to bridgeX
1321  *                        pseudo netif's address is y.y.y.y
1322  *
1323  * 7) bridgeX:y.y.y.y/z   tap(4) will be added to bridgeX
1324  *                        pseudo netif's address is y.y.y.y
1325  *                        pseudo netif's netmask len is z
1326  */
1327 static
1328 int
1329 netif_init_tap(int tap_unit, in_addr_t *addr, in_addr_t *mask, int s)
1330 {
1331         in_addr_t tap_addr, netmask, netif_addr;
1332         int next_netif_addr;
1333         char *tok, *masklen_str, *ifbridge;
1334
1335         *addr = 0;
1336         *mask = 0;
1337
1338         tok = strtok(NULL, ":/");
1339         if (tok == NULL) {
1340                 /*
1341                  * Nothing special, simply use tap(4) as backend
1342                  */
1343                 return 0;
1344         }
1345
1346         if (inet_pton(AF_INET, tok, &tap_addr) > 0) {
1347                 /*
1348                  * tap(4)'s address is supplied
1349                  */
1350                 ifbridge = NULL;
1351
1352                 /*
1353                  * If there is next token, then it may be pseudo
1354                  * netif's address or netmask len for tap(4)
1355                  */
1356                 next_netif_addr = 0;
1357         } else {
1358                 /*
1359                  * Not tap(4)'s address, assume it as a bridge(4)
1360                  * iface name
1361                  */
1362                 tap_addr = 0;
1363                 ifbridge = tok;
1364
1365                 /*
1366                  * If there is next token, then it must be pseudo
1367                  * netif's address
1368                  */
1369                 next_netif_addr = 1;
1370         }
1371
1372         netmask = netif_addr = 0;
1373
1374         tok = strtok(NULL, ":/");
1375         if (tok == NULL)
1376                 goto back;
1377
1378         if (inet_pton(AF_INET, tok, &netif_addr) <= 0) {
1379                 if (next_netif_addr) {
1380                         warnx("Invalid pseudo netif address: %s", tok);
1381                         return -1;
1382                 }
1383                 netif_addr = 0;
1384
1385                 /*
1386                  * Current token is not address, then it must be netmask len
1387                  */
1388                 masklen_str = tok;
1389         } else {
1390                 /*
1391                  * Current token is pseudo netif address, if there is next token
1392                  * it must be netmask len
1393                  */
1394                 masklen_str = strtok(NULL, "/");
1395         }
1396
1397         /* Calculate netmask */
1398         if (masklen_str != NULL) {
1399                 u_long masklen;
1400
1401                 masklen = strtoul(masklen_str, NULL, 10);
1402                 if (masklen < 32 && masklen > 0) {
1403                         netmask = htonl(~((1LL << (32 - masklen)) - 1)
1404                                         & 0xffffffff);
1405                 } else {
1406                         warnx("Invalid netmask len: %lu", masklen);
1407                         return -1;
1408                 }
1409         }
1410
1411         /* Make sure there is no more token left */
1412         if (strtok(NULL, ":/") != NULL) {
1413                 warnx("Invalid argument to '-I'");
1414                 return -1;
1415         }
1416
1417 back:
1418         if (tap_unit < 0) {
1419                 /* Do nothing */
1420         } else if (ifbridge == NULL) {
1421                 /* Set tap(4) address/netmask */
1422                 if (netif_set_tapaddr(tap_unit, tap_addr, netmask, s) < 0)
1423                         return -1;
1424         } else {
1425                 /* Tie tap(4) to bridge(4) */
1426                 if (netif_add_tap2brg(tap_unit, ifbridge, s) < 0)
1427                         return -1;
1428         }
1429
1430         *addr = netif_addr;
1431         *mask = netmask;
1432         return 0;
1433 }
1434
1435 /*
1436  * NetifInfo[] will be filled for pseudo netif initialization.
1437  * NetifNum will be bumped to reflect the number of valid entries
1438  * in NetifInfo[].
1439  */
1440 static
1441 void
1442 init_netif(char *netifExp[], int netifExpNum)
1443 {
1444         int i, s;
1445         char *tmp;
1446
1447         if (netifExpNum == 0)
1448                 return;
1449
1450         s = socket(AF_INET, SOCK_DGRAM, 0);     /* for ioctl(SIOC) */
1451         if (s < 0)
1452                 return;
1453
1454         for (i = 0; i < netifExpNum; ++i) {
1455                 struct vknetif_info *info;
1456                 in_addr_t netif_addr, netif_mask;
1457                 int tap_fd, tap_unit;
1458                 char *netif;
1459
1460                 /* Extract MAC address if there is one */
1461                 tmp = netifExp[i];
1462                 strsep(&tmp, "=");
1463
1464                 netif = strtok(netifExp[i], ":");
1465                 if (netif == NULL) {
1466                         warnx("Invalid argument to '-I'");
1467                         continue;
1468                 }
1469
1470                 /*
1471                  * Open tap(4) device file and bring up the
1472                  * corresponding interface
1473                  */
1474                 tap_fd = netif_open_tap(netif, &tap_unit, s);
1475                 if (tap_fd < 0)
1476                         continue;
1477
1478                 /*
1479                  * Initialize tap(4) and get address/netmask
1480                  * for pseudo netif
1481                  *
1482                  * NB: Rest part of netifExp[i] is passed
1483                  *     to netif_init_tap() implicitly.
1484                  */
1485                 if (netif_init_tap(tap_unit, &netif_addr, &netif_mask, s) < 0) {
1486                         /*
1487                          * NB: Closing tap(4) device file will bring
1488                          *     down the corresponding interface
1489                          */
1490                         close(tap_fd);
1491                         continue;
1492                 }
1493
1494                 info = &NetifInfo[NetifNum];
1495                 bzero(info, sizeof(*info));
1496                 info->tap_fd = tap_fd;
1497                 info->tap_unit = tap_unit;
1498                 info->netif_addr = netif_addr;
1499                 info->netif_mask = netif_mask;
1500                 /*
1501                  * If tmp isn't NULL it means a MAC could have been
1502                  * specified so attempt to convert it.
1503                  * Setting enaddr to NULL will tell vke_attach() we
1504                  * need a pseudo-random MAC address.
1505                  */
1506                 if (tmp != NULL) {
1507                         if ((info->enaddr = malloc(ETHER_ADDR_LEN)) == NULL)
1508                                 warnx("Couldn't allocate memory for the operation");
1509                         else {
1510                                 if ((kether_aton(tmp, info->enaddr)) == NULL) {
1511                                         free(info->enaddr);
1512                                         info->enaddr = NULL;
1513                                 }
1514                         }
1515                 }
1516
1517                 NetifNum++;
1518                 if (NetifNum >= VKNETIF_MAX)    /* XXX will this happen? */
1519                         break;
1520         }
1521         close(s);
1522 }
1523
1524 /*
1525  * Create the pid file and leave it open and locked while the vkernel is
1526  * running.  This allows a script to use /usr/bin/lockf to probe whether
1527  * a vkernel is still running (so as not to accidently kill an unrelated
1528  * process from a stale pid file).
1529  */
1530 static
1531 void
1532 writepid(void)
1533 {
1534         char buf[32];
1535         int fd;
1536
1537         if (pid_file != NULL) {
1538                 snprintf(buf, sizeof(buf), "%ld\n", (long)getpid());
1539                 fd = open(pid_file, O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0666);
1540                 if (fd < 0) {
1541                         if (errno == EWOULDBLOCK) {
1542                                 perror("Failed to lock pidfile, "
1543                                        "vkernel already running");
1544                         } else {
1545                                 perror("Failed to create pidfile");
1546                         }
1547                         exit(EX_SOFTWARE);
1548                 }
1549                 ftruncate(fd, 0);
1550                 write(fd, buf, strlen(buf));
1551                 /* leave the file open to maintain the lock */
1552         }
1553 }
1554
1555 static
1556 void
1557 cleanpid( void )
1558 {
1559         if (pid_file != NULL) {
1560                 if (unlink(pid_file) < 0)
1561                         perror("Warning: couldn't remove pidfile");
1562         }
1563 }
1564
1565 static
1566 void
1567 usage_err(const char *ctl, ...)
1568 {
1569         va_list va;
1570
1571         va_start(va, ctl);
1572         vfprintf(stderr, ctl, va);
1573         va_end(va);
1574         fprintf(stderr, "\n");
1575         exit(EX_USAGE);
1576 }
1577
1578 static
1579 void
1580 usage_help(_Bool help)
1581 {
1582         fprintf(stderr, "Usage: %s [-hsUvdtT] [-c file] [-e name=value:name=value:...]\n"
1583             "\t[-i file] [-I interface[:address1[:address2][/netmask]]] [-l cpulock]\n"
1584             "\t[-m size] [-n numcpus[:lbits[:cbits]]]\n"
1585             "\t[-p file] [-r file]\n", save_av[0]);
1586
1587         if (help)
1588                 fprintf(stderr, "\nArguments:\n"
1589                     "\t-c\tSpecify a readonly CD-ROM image file to be used by the kernel.\n"
1590                     "\t-e\tSpecify an environment to be used by the kernel.\n"
1591                     "\t-h\tThis list of options.\n"
1592                     "\t-i\tSpecify a memory image file to be used by the virtual kernel.\n"
1593                     "\t-I\tCreate a virtual network device.\n"
1594                     "\t-l\tSpecify which, if any, real CPUs to lock virtual CPUs to.\n"
1595                     "\t-m\tSpecify the amount of memory to be used by the kernel in bytes.\n"
1596                     "\t-n\tSpecify the number of CPUs and the topology you wish to emulate:\n"
1597                     "\t\t\tnumcpus - number of cpus\n"
1598                     "\t\t\tlbits - specify the number of bits within APICID(=CPUID)\n"
1599                     "\t\t\t        needed for representing the logical ID.\n"
1600                     "\t\t\t        Controls the number of threads/core:\n"
1601                     "\t\t\t        (0 bits - 1 thread, 1 bit - 2 threads).\n"
1602                     "\t\t\tcbits - specify the number of bits within APICID(=CPUID)\n"
1603                     "\t\t\t        needed for representing the core ID.\n"
1604                     "\t\t\t        Controls the number of cores/package:\n"
1605                     "\t\t\t        (0 bits - 1 core, 1 bit - 2 cores).\n"
1606                     "\t-p\tSpecify a file in which to store the process ID.\n"
1607                     "\t-r\tSpecify a R/W disk image file, iterates vkd0..n\n"
1608                     "\t-R\tSpecify a COW disk image file, iterates vkd0..n\n"
1609                     "\t-s\tBoot into single-user mode.\n"
1610                     "\t-t\tUse a precise host timer when calculating clock values.\n"
1611                     "\t-T\tDisallow use of the TSC cpu timer as a clock.\n"
1612                     "\t-U\tEnable writing to kernel memory and module loading.\n"
1613                     "\t-v\tTurn on verbose booting.\n");
1614
1615         exit(EX_USAGE);
1616 }
1617
1618 void
1619 cpu_smp_stopped(void)
1620 {
1621 }
1622
1623 void
1624 cpu_reset(void)
1625 {
1626         kprintf("cpu reset, rebooting vkernel\n");
1627         closefrom(3);
1628         cleanpid();
1629         exit(EX_VKERNEL_REBOOT);
1630 }
1631
1632 void
1633 cpu_halt(void)
1634 {
1635         kprintf("cpu halt, exiting vkernel\n");
1636         cleanpid();
1637         exit(EX_OK);
1638 }
1639
1640 void
1641 setrealcpu(void)
1642 {
1643         switch(lwp_cpu_lock) {
1644         case LCL_PER_CPU:
1645                 if (bootverbose)
1646                         kprintf("Locking CPU%d to real cpu %d\n",
1647                                 mycpuid, next_cpu);
1648                 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu));
1649                 next_cpu++;
1650                 if (next_cpu >= real_ncpus)
1651                         next_cpu = 0;
1652                 break;
1653         case LCL_SINGLE_CPU:
1654                 if (bootverbose)
1655                         kprintf("Locking CPU%d to real cpu %d\n",
1656                                 mycpuid, next_cpu);
1657                 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu));
1658                 break;
1659         default:
1660                 /* do not map virtual cpus to real cpus */
1661                 break;
1662         }
1663 }
1664
1665 /*
1666  * Allocate and free memory for module loading.  The loaded module
1667  * has to be placed somewhere near the current kernel binary load
1668  * point or the relocations will not work.
1669  *
1670  * I'm not sure why this isn't working.
1671  */
1672 int
1673 vkernel_module_memory_alloc(vm_offset_t *basep, size_t bytes)
1674 {
1675 #if 1
1676         size_t xtra;
1677         xtra = (PAGE_SIZE - (vm_offset_t)sbrk(0)) & PAGE_MASK;
1678         *basep = (vm_offset_t)sbrk(xtra + bytes) + xtra;
1679         bzero((void *)*basep, bytes);
1680 #else
1681         *basep = (vm_offset_t)mmap((void *)0x000000000, bytes,
1682                                    PROT_READ|PROT_WRITE|PROT_EXEC,
1683                                    MAP_ANON|MAP_SHARED, -1, 0);
1684         if ((void *)*basep == MAP_FAILED)
1685                 return ENOMEM;
1686 #endif
1687         return 0;
1688 }
1689
1690 void
1691 vkernel_module_memory_free(vm_offset_t base, size_t bytes)
1692 {
1693 #if 0
1694 #if 0
1695         munmap((void *)base, bytes);
1696 #endif
1697 #endif
1698 }