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