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 #include <pthread.h> 85 86 #define EX_VKERNEL_REBOOT 32 87 88 vm_phystable_t phys_avail[16]; 89 vm_paddr_t Maxmem; 90 vm_paddr_t Maxmem_bytes; 91 long physmem; 92 int MemImageFd = -1; 93 struct vkdisk_info DiskInfo[VKDISK_MAX]; 94 int DiskNum; 95 struct vknetif_info NetifInfo[VKNETIF_MAX]; 96 int NetifNum; 97 char *pid_file; 98 vm_offset_t KvaStart; 99 vm_offset_t KvaEnd; 100 vm_offset_t KvaSize; 101 vm_offset_t virtual_start; 102 vm_offset_t virtual_end; 103 vm_offset_t virtual2_start; 104 vm_offset_t virtual2_end; 105 vm_offset_t kernel_vm_end; 106 vm_offset_t crashdumpmap; 107 vm_offset_t clean_sva; 108 vm_offset_t clean_eva; 109 struct msgbuf *msgbufp; 110 caddr_t ptvmmap; 111 vpte_t *KernelPTD; 112 vpte_t *KernelPTA; /* Warning: Offset for direct VA translation */ 113 void *dmap_min_address; 114 void *vkernel_stack; 115 u_int cpu_feature; /* XXX */ 116 int tsc_present; 117 int tsc_invariant; 118 int tsc_mpsync; 119 int optcpus; /* number of cpus - see mp_start() */ 120 int cpu_bits; 121 int lwp_cpu_lock; /* if/how to lock virtual CPUs to real CPUs */ 122 int real_ncpus; /* number of real CPUs */ 123 int next_cpu; /* next real CPU to lock a virtual CPU to */ 124 int vkernel_b_arg; /* no of logical CPU bits - only SMP */ 125 int vkernel_B_arg; /* no of core bits - only SMP */ 126 int vmm_enabled; /* VMM HW assisted enable */ 127 int use_precise_timer = 0; /* use a precise timer (more expensive) */ 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 'v': 323 bootverbose = 1; 324 break; 325 case 'i': 326 memImageFile = optarg; 327 break; 328 case 'I': 329 if (netifFileNum < VKNETIF_MAX) 330 netifFile[netifFileNum++] = strdup(optarg); 331 break; 332 case 'r': 333 case 'R': 334 if (bootOnDisk < 0) 335 bootOnDisk = 1; 336 if (diskFileNum + cdFileNum < VKDISK_MAX) { 337 diskFile[diskFileNum] = strdup(optarg); 338 diskFlags[diskFileNum] = (c == 'R'); 339 ++diskFileNum; 340 } 341 break; 342 case 'c': 343 if (bootOnDisk < 0) 344 bootOnDisk = 0; 345 if (diskFileNum + cdFileNum < VKDISK_MAX) 346 cdFile[cdFileNum++] = strdup(optarg); 347 break; 348 case 'm': 349 Maxmem_bytes = strtoull(optarg, &suffix, 0); 350 if (suffix) { 351 switch(*suffix) { 352 case 'g': 353 case 'G': 354 Maxmem_bytes <<= 30; 355 break; 356 case 'm': 357 case 'M': 358 Maxmem_bytes <<= 20; 359 break; 360 case 'k': 361 case 'K': 362 Maxmem_bytes <<= 10; 363 break; 364 default: 365 Maxmem_bytes = 0; 366 usage_err("Bad maxmem option"); 367 /* NOT REACHED */ 368 break; 369 } 370 } 371 break; 372 case 'l': 373 next_cpu = -1; 374 if (strncmp("map", optarg, 3) == 0) { 375 lwp_cpu_lock = LCL_PER_CPU; 376 if (optarg[3] == ',') { 377 next_cpu = strtol(optarg+4, &endp, 0); 378 if (*endp != '\0') 379 usage_err("Bad target CPU number at '%s'", endp); 380 } else { 381 next_cpu = 0; 382 } 383 if (next_cpu < 0 || next_cpu > real_ncpus - 1) 384 usage_err("Bad target CPU, valid range is 0-%d", real_ncpus - 1); 385 } else if (strncmp("any", optarg, 3) == 0) { 386 lwp_cpu_lock = LCL_NONE; 387 } else { 388 lwp_cpu_lock = LCL_SINGLE_CPU; 389 next_cpu = strtol(optarg, &endp, 0); 390 if (*endp != '\0') 391 usage_err("Bad target CPU number at '%s'", endp); 392 if (next_cpu < 0 || next_cpu > real_ncpus - 1) 393 usage_err("Bad target CPU, valid range is 0-%d", real_ncpus - 1); 394 } 395 break; 396 case 'n': 397 /* 398 * This value is set up by mp_start(), don't just 399 * set ncpus here. 400 */ 401 tok = strtok(optarg, ":"); 402 optcpus = strtol(tok, NULL, 0); 403 if (optcpus < 1 || optcpus > MAXCPU) 404 usage_err("Bad ncpus, valid range is 1-%d", MAXCPU); 405 cpu_bits = 1; 406 while ((1 << cpu_bits) < optcpus) 407 ++cpu_bits; 408 409 /* 410 * By default assume simple hyper-threading 411 */ 412 vkernel_b_arg = 1; 413 vkernel_B_arg = cpu_bits - vkernel_b_arg; 414 415 /* 416 * [:lbits[:cbits]] override # of cpu bits 417 * for logical and core extraction, supplying 418 * defaults for any omission. 419 */ 420 tok = strtok(NULL, ":"); 421 if (tok != NULL) { 422 vkernel_b_arg = strtol(tok, NULL, 0); 423 vkernel_B_arg = cpu_bits - vkernel_b_arg; 424 425 /* :cbits argument */ 426 tok = strtok(NULL, ":"); 427 if (tok != NULL) { 428 vkernel_B_arg = strtol(tok, NULL, 0); 429 } 430 } 431 break; 432 case 'p': 433 pid_file = optarg; 434 break; 435 case 'U': 436 kernel_mem_readonly = 0; 437 break; 438 case 'h': 439 usage_help(true); 440 break; 441 case 'z': 442 prezeromem = 1; 443 break; 444 default: 445 usage_help(false); 446 } 447 } 448 449 /* 450 * Check VMM presence 451 */ 452 vsize = sizeof(vmm_enabled); 453 sysctlbyname("hw.vmm.enable", &vmm_enabled, &vsize, NULL, 0); 454 vmm_enabled = (vmm_enabled && !dflag); 455 456 writepid(); 457 cpu_disable_intr(); 458 if (vmm_enabled) { 459 /* use a MAP_ANON directly */ 460 printf("VMM is available\n"); 461 init_kern_memory_vmm(); 462 } else { 463 printf("VMM is not available\n"); 464 init_sys_memory(memImageFile); 465 init_kern_memory(); 466 } 467 init_globaldata(); 468 init_vkernel(); 469 setrealcpu(); 470 init_kqueue(); 471 472 vmm_guest = VMM_GUEST_VKERNEL; 473 474 /* 475 * Check TSC 476 */ 477 vsize = sizeof(tsc_present); 478 sysctlbyname("hw.tsc_present", &tsc_present, &vsize, NULL, 0); 479 vsize = sizeof(tsc_invariant); 480 sysctlbyname("hw.tsc_invariant", &tsc_invariant, &vsize, NULL, 0); 481 vsize = sizeof(tsc_mpsync); 482 sysctlbyname("hw.tsc_mpsync", &tsc_mpsync, &vsize, NULL, 0); 483 vsize = sizeof(tsc_frequency); 484 sysctlbyname("hw.tsc_frequency", &tsc_frequency, &vsize, NULL, 0); 485 if (tsc_present) 486 cpu_feature |= CPUID_TSC; 487 tsc_oneus_approx = ((tsc_frequency|1) + 999999) / 1000000; 488 489 /* 490 * Check SSE 491 */ 492 vsize = sizeof(supports_sse); 493 supports_sse = 0; 494 sysctlbyname("hw.instruction_sse", &supports_sse, &vsize, NULL, 0); 495 sysctlbyname("hw.mxcsr_mask", &mxcsr_mask, &msize, NULL, 0); 496 init_fpu(supports_sse); 497 if (supports_sse) 498 cpu_feature |= CPUID_SSE | CPUID_FXSR; 499 500 /* 501 * We boot from the first installed disk. 502 */ 503 if (bootOnDisk == 1) { 504 init_disk(diskFile, diskFlags, diskFileNum, VKD_DISK); 505 init_disk(cdFile, NULL, cdFileNum, VKD_CD); 506 } else { 507 init_disk(cdFile, NULL, cdFileNum, VKD_CD); 508 init_disk(diskFile, diskFlags, diskFileNum, VKD_DISK); 509 } 510 511 init_netif(netifFile, netifFileNum); 512 init_exceptions(); 513 mi_startup(); 514 /* NOT REACHED */ 515 exit(EX_SOFTWARE); 516 } 517 518 /* SIGTERM handler */ 519 static 520 void 521 handle_term(int sig) 522 { 523 kill(childpid, sig); 524 } 525 526 /* 527 * Initialize system memory. This is the virtual kernel's 'RAM'. 528 */ 529 static 530 void 531 init_sys_memory(char *imageFile) 532 { 533 struct stat st; 534 int i; 535 int fd; 536 537 /* 538 * Figure out the system memory image size. If an image file was 539 * specified and -m was not specified, use the image file's size. 540 */ 541 if (imageFile && stat(imageFile, &st) == 0 && Maxmem_bytes == 0) 542 Maxmem_bytes = (vm_paddr_t)st.st_size; 543 if ((imageFile == NULL || stat(imageFile, &st) < 0) && 544 Maxmem_bytes == 0) { 545 errx(1, "Cannot create new memory file %s unless " 546 "system memory size is specified with -m", 547 imageFile); 548 /* NOT REACHED */ 549 } 550 551 /* 552 * Maxmem must be known at this time 553 */ 554 if (Maxmem_bytes < 64 * 1024 * 1024 || (Maxmem_bytes & SEG_MASK)) { 555 errx(1, "Bad maxmem specification: 64MB minimum, " 556 "multiples of %dMB only", 557 SEG_SIZE / 1024 / 1024); 558 /* NOT REACHED */ 559 } 560 561 /* 562 * Generate an image file name if necessary, then open/create the 563 * file exclusively locked. Do not allow multiple virtual kernels 564 * to use the same image file. 565 * 566 * Don't iterate through a million files if we do not have write 567 * access to the directory, stop if our open() failed on a 568 * non-existant file. Otherwise opens can fail for any number 569 */ 570 if (imageFile == NULL) { 571 for (i = 0; i < 1000000; ++i) { 572 asprintf(&imageFile, "/var/vkernel/memimg.%06d", i); 573 fd = open(imageFile, 574 O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644); 575 if (fd < 0 && stat(imageFile, &st) == 0) { 576 free(imageFile); 577 continue; 578 } 579 break; 580 } 581 } else { 582 fd = open(imageFile, O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644); 583 } 584 fprintf(stderr, "Using memory file: %s\n", imageFile); 585 if (fd < 0 || fstat(fd, &st) < 0) { 586 err(1, "Unable to open/create %s", imageFile); 587 /* NOT REACHED */ 588 } 589 590 /* 591 * Truncate or extend the file as necessary. Clean out the contents 592 * of the file, we want it to be full of holes so we don't waste 593 * time reading in data from an old file that we no longer care 594 * about. 595 */ 596 ftruncate(fd, 0); 597 ftruncate(fd, Maxmem_bytes); 598 599 MemImageFd = fd; 600 Maxmem = Maxmem_bytes >> PAGE_SHIFT; 601 physmem = Maxmem; 602 } 603 604 /* 605 * Initialize kernel memory. This reserves kernel virtual memory by using 606 * MAP_VPAGETABLE 607 */ 608 609 static 610 void 611 init_kern_memory(void) 612 { 613 void *base; 614 int i; 615 void *firstfree; 616 617 /* 618 * Memory map our kernel virtual memory space. Note that the 619 * kernel image itself is not made part of this memory for the 620 * moment. 621 * 622 * The memory map must be segment-aligned so we can properly 623 * offset KernelPTD. 624 * 625 * If the system kernel has a different MAXDSIZ, it might not 626 * be possible to map kernel memory in its prefered location. 627 * Try a number of different locations. 628 */ 629 630 base = mmap((void*)KERNEL_KVA_START, KERNEL_KVA_SIZE, 631 PROT_READ|PROT_WRITE|PROT_EXEC, 632 MAP_FILE|MAP_SHARED|MAP_VPAGETABLE|MAP_FIXED|MAP_TRYFIXED, 633 MemImageFd, (off_t)KERNEL_KVA_START); 634 635 if (base == MAP_FAILED) { 636 err(1, "Unable to mmap() kernel virtual memory!"); 637 /* NOT REACHED */ 638 } 639 madvise(base, KERNEL_KVA_SIZE, MADV_NOSYNC); 640 KvaStart = (vm_offset_t)base; 641 KvaSize = KERNEL_KVA_SIZE; 642 KvaEnd = KvaStart + KvaSize; 643 644 /* cannot use kprintf yet */ 645 printf("KVM mapped at %p-%p\n", (void *)KvaStart, (void *)KvaEnd); 646 647 /* MAP_FILE? */ 648 dmap_min_address = mmap(0, DMAP_SIZE, PROT_READ|PROT_WRITE, 649 MAP_NOCORE|MAP_NOSYNC|MAP_SHARED, 650 MemImageFd, 0); 651 if (dmap_min_address == MAP_FAILED) { 652 err(1, "Unable to mmap() kernel DMAP region!"); 653 /* NOT REACHED */ 654 } 655 656 /* 657 * Prefault the memory. The vkernel is going to fault it all in 658 * anyway, and faults on the backing store itself are very expensive 659 * once we go SMP (contend a lot). So do it now. 660 */ 661 if (prezeromem) 662 bzero(dmap_min_address, Maxmem_bytes); 663 664 /* 665 * Bootstrap the kernel_pmap 666 */ 667 firstfree = NULL; 668 pmap_bootstrap((vm_paddr_t *)&firstfree, (int64_t)base); 669 670 mcontrol(base, KERNEL_KVA_SIZE, MADV_SETMAP, 671 0 | VPTE_RW | VPTE_V); 672 673 /* 674 * phys_avail[] represents unallocated physical memory. MI code 675 * will use phys_avail[] to create the vm_page array. 676 */ 677 phys_avail[0].phys_beg = (vm_paddr_t)firstfree; 678 phys_avail[0].phys_beg = (phys_avail[0].phys_beg + PAGE_MASK) & 679 ~(vm_paddr_t)PAGE_MASK; 680 phys_avail[0].phys_end = Maxmem_bytes; 681 682 #if 0 /* JGV */ 683 /* 684 * (virtual_start, virtual_end) represent unallocated kernel virtual 685 * memory. MI code will create kernel_map using these parameters. 686 */ 687 virtual_start = KvaStart + (long)firstfree; 688 virtual_start = (virtual_start + PAGE_MASK) & ~(vm_offset_t)PAGE_MASK; 689 virtual_end = KvaStart + KERNEL_KVA_SIZE; 690 #endif 691 692 /* 693 * pmap_growkernel() will set the correct value. 694 */ 695 kernel_vm_end = 0; 696 697 /* 698 * Allocate space for process 0's UAREA. 699 */ 700 proc0paddr = (void *)virtual_start; 701 for (i = 0; i < UPAGES; ++i) { 702 pmap_kenter_quick(virtual_start, phys_avail[0].phys_beg); 703 virtual_start += PAGE_SIZE; 704 phys_avail[0].phys_beg += PAGE_SIZE; 705 } 706 707 /* 708 * crashdumpmap 709 */ 710 crashdumpmap = virtual_start; 711 virtual_start += MAXDUMPPGS * PAGE_SIZE; 712 713 /* 714 * msgbufp maps the system message buffer 715 */ 716 assert((MSGBUF_SIZE & PAGE_MASK) == 0); 717 msgbufp = (void *)virtual_start; 718 for (i = 0; i < (MSGBUF_SIZE >> PAGE_SHIFT); ++i) { 719 pmap_kenter_quick(virtual_start, phys_avail[0].phys_beg); 720 virtual_start += PAGE_SIZE; 721 phys_avail[0].phys_beg += PAGE_SIZE; 722 } 723 msgbufinit(msgbufp, MSGBUF_SIZE); 724 725 /* 726 * used by kern_memio for /dev/mem access 727 */ 728 ptvmmap = (caddr_t)virtual_start; 729 virtual_start += PAGE_SIZE; 730 } 731 732 static 733 void 734 init_kern_memory_vmm(void) 735 { 736 int i; 737 void *firstfree; 738 struct vmm_guest_options options; 739 void *dmap_address; 740 741 KvaStart = (vm_offset_t)KERNEL_KVA_START; 742 KvaSize = KERNEL_KVA_SIZE; 743 KvaEnd = KvaStart + KvaSize; 744 745 Maxmem = Maxmem_bytes >> PAGE_SHIFT; 746 physmem = Maxmem; 747 748 if (Maxmem_bytes < 64 * 1024 * 1024 || (Maxmem_bytes & SEG_MASK)) { 749 errx(1, "Bad maxmem specification: 64MB minimum, " 750 "multiples of %dMB only", 751 SEG_SIZE / 1024 / 1024); 752 /* NOT REACHED */ 753 } 754 755 /* Call the vmspace_create to allocate the internal 756 * vkernel structures. Won't do anything else (no new 757 * vmspace) 758 */ 759 if (vmspace_create(NULL, 0, NULL) < 0) 760 panic("vmspace_create() failed"); 761 762 763 /* 764 * MAP_ANON the region of the VKERNEL phyisical memory 765 * (known as GPA - Guest Physical Address 766 */ 767 dmap_address = mmap(NULL, Maxmem_bytes, 768 PROT_READ|PROT_WRITE|PROT_EXEC, 769 MAP_ANON|MAP_SHARED, -1, 0); 770 if (dmap_address == MAP_FAILED) { 771 err(1, "Unable to mmap() RAM region!"); 772 /* NOT REACHED */ 773 } 774 if (prezeromem) 775 bzero(dmap_address, Maxmem_bytes); 776 777 /* Alloc a new stack in the lowmem */ 778 vkernel_stack = mmap(NULL, KERNEL_STACK_SIZE, 779 PROT_READ|PROT_WRITE|PROT_EXEC, 780 MAP_ANON, -1, 0); 781 if (vkernel_stack == MAP_FAILED) { 782 err(1, "Unable to allocate stack\n"); 783 } 784 785 /* 786 * Bootstrap the kernel_pmap 787 */ 788 firstfree = dmap_address; 789 dmap_min_address = NULL; /* VIRT == PHYS in the first 512G */ 790 pmap_bootstrap((vm_paddr_t *)&firstfree, (uint64_t)KvaStart); 791 792 /* 793 * Enter VMM mode 794 */ 795 bzero(&options, sizeof(options)); 796 options.guest_cr3 = (register_t) KPML4phys; 797 options.new_stack = (uint64_t) vkernel_stack + KERNEL_STACK_SIZE; 798 options.master = 1; 799 if (vmm_guest_ctl(VMM_GUEST_RUN, &options)) { 800 err(1, "Unable to enter VMM mode."); 801 } 802 803 /* 804 * phys_avail[] represents unallocated physical memory. MI code 805 * will use phys_avail[] to create the vm_page array. 806 */ 807 phys_avail[0].phys_beg = (vm_paddr_t)firstfree; 808 phys_avail[0].phys_beg = (phys_avail[0].phys_beg + PAGE_MASK) & 809 ~(vm_paddr_t)PAGE_MASK; 810 phys_avail[0].phys_end = (vm_paddr_t)dmap_address + Maxmem_bytes; 811 812 /* 813 * pmap_growkernel() will set the correct value. 814 */ 815 kernel_vm_end = 0; 816 817 /* 818 * Allocate space for process 0's UAREA. 819 */ 820 proc0paddr = (void *)virtual_start; 821 for (i = 0; i < UPAGES; ++i) { 822 pmap_kenter_quick(virtual_start, phys_avail[0].phys_beg); 823 virtual_start += PAGE_SIZE; 824 phys_avail[0].phys_beg += PAGE_SIZE; 825 } 826 827 /* 828 * crashdumpmap 829 */ 830 crashdumpmap = virtual_start; 831 virtual_start += MAXDUMPPGS * PAGE_SIZE; 832 833 /* 834 * msgbufp maps the system message buffer 835 */ 836 assert((MSGBUF_SIZE & PAGE_MASK) == 0); 837 msgbufp = (void *)virtual_start; 838 for (i = 0; i < (MSGBUF_SIZE >> PAGE_SHIFT); ++i) { 839 840 pmap_kenter_quick(virtual_start, phys_avail[0].phys_beg); 841 virtual_start += PAGE_SIZE; 842 phys_avail[0].phys_beg += PAGE_SIZE; 843 } 844 845 msgbufinit(msgbufp, MSGBUF_SIZE); 846 847 /* 848 * used by kern_memio for /dev/mem access 849 */ 850 ptvmmap = (caddr_t)virtual_start; 851 virtual_start += PAGE_SIZE; 852 853 printf("vmm: Hardware pagetable enabled for guest\n"); 854 } 855 856 857 /* 858 * Map the per-cpu globaldata for cpu #0. Allocate the space using 859 * virtual_start and phys_avail[0] 860 */ 861 static 862 void 863 init_globaldata(void) 864 { 865 int i; 866 vm_paddr_t pa; 867 vm_offset_t va; 868 869 /* 870 * Reserve enough KVA to cover possible cpus. This is a considerable 871 * amount of KVA since the privatespace structure includes two 872 * whole page table mappings. 873 */ 874 virtual_start = (virtual_start + SEG_MASK) & ~(vm_offset_t)SEG_MASK; 875 CPU_prvspace = (void *)virtual_start; 876 virtual_start += sizeof(struct privatespace) * SMP_MAXCPU; 877 878 /* 879 * Allocate enough physical memory to cover the mdglobaldata 880 * portion of the space and the idle stack and map the pages 881 * into KVA. For cpu #0 only. 882 */ 883 for (i = 0; i < sizeof(struct mdglobaldata); i += PAGE_SIZE) { 884 pa = phys_avail[0].phys_beg; 885 va = (vm_offset_t)&CPU_prvspace[0].mdglobaldata + i; 886 pmap_kenter_quick(va, pa); 887 phys_avail[0].phys_beg += PAGE_SIZE; 888 } 889 for (i = 0; i < sizeof(CPU_prvspace[0].idlestack); i += PAGE_SIZE) { 890 pa = phys_avail[0].phys_beg; 891 va = (vm_offset_t)&CPU_prvspace[0].idlestack + i; 892 pmap_kenter_quick(va, pa); 893 phys_avail[0].phys_beg += PAGE_SIZE; 894 } 895 896 /* 897 * Setup the %gs for cpu #0. The mycpu macro works after this 898 * point. Note that %fs is used by pthreads. 899 */ 900 tls_set_gs(&CPU_prvspace[0], sizeof(struct privatespace)); 901 } 902 903 904 /* 905 * Initialize pool tokens and other necessary locks 906 */ 907 static void 908 init_locks(void) 909 { 910 911 /* 912 * Get the initial mplock with a count of 1 for the BSP. 913 * This uses a LOGICAL cpu ID, ie BSP == 0. 914 */ 915 cpu_get_initial_mplock(); 916 917 /* our token pool needs to work early */ 918 lwkt_token_pool_init(); 919 920 } 921 922 923 /* 924 * Initialize very low level systems including thread0, proc0, etc. 925 */ 926 static 927 void 928 init_vkernel(void) 929 { 930 struct mdglobaldata *gd; 931 932 gd = &CPU_prvspace[0].mdglobaldata; 933 bzero(gd, sizeof(*gd)); 934 935 gd->mi.gd_curthread = &thread0; 936 thread0.td_gd = &gd->mi; 937 ncpus = 1; 938 ncpus_fit = 1; /* rounded up power of 2 */ 939 /* ncpus_fit_mask are 0 */ 940 init_param1(); 941 gd->mi.gd_prvspace = &CPU_prvspace[0]; 942 mi_gdinit(&gd->mi, 0); 943 cpu_gdinit(gd, 0); 944 mi_proc0init(&gd->mi, proc0paddr); 945 lwp0.lwp_md.md_regs = &proc0_tf; 946 947 init_locks(); 948 cninit(); 949 rand_initialize(); 950 #if 0 /* #ifdef DDB */ 951 kdb_init(); 952 if (boothowto & RB_KDB) 953 Debugger("Boot flags requested debugger"); 954 #endif 955 identcpu(); 956 #if 0 957 initializecpu(); /* Initialize CPU registers */ 958 #endif 959 init_param2((phys_avail[0].phys_end - 960 phys_avail[0].phys_beg) / PAGE_SIZE); 961 962 #if 0 963 /* 964 * Map the message buffer 965 */ 966 for (off = 0; off < round_page(MSGBUF_SIZE); off += PAGE_SIZE) 967 pmap_kenter((vm_offset_t)msgbufp + off, avail_end + off); 968 msgbufinit(msgbufp, MSGBUF_SIZE); 969 #endif 970 #if 0 971 thread0.td_pcb_cr3 ... MMU 972 lwp0.lwp_md.md_regs = &proc0_tf; 973 #endif 974 } 975 976 /* 977 * Filesystem image paths for the virtual kernel are optional. 978 * If specified they each should point to a disk image, 979 * the first of which will become the root disk. 980 * 981 * The virtual kernel caches data from our 'disk' just like a normal kernel, 982 * so we do not really want the real kernel to cache the data too. Use 983 * O_DIRECT to remove the duplication. 984 */ 985 static 986 void 987 init_disk(char **diskExp, int *diskFlags, int diskFileNum, enum vkdisk_type type) 988 { 989 char *serno; 990 int i; 991 992 if (diskFileNum == 0) 993 return; 994 995 for (i=0; i < diskFileNum; i++){ 996 char *fname; 997 fname = diskExp[i]; 998 999 if (fname == NULL) { 1000 warnx("Invalid argument to '-r'"); 1001 continue; 1002 } 1003 /* 1004 * Check for a serial number for the virtual disk 1005 * passed from the command line. 1006 */ 1007 serno = fname; 1008 strsep(&serno, ":"); 1009 1010 if (DiskNum < VKDISK_MAX) { 1011 struct stat st; 1012 struct vkdisk_info *info = NULL; 1013 int fd; 1014 size_t l = 0; 1015 1016 if (type == VKD_DISK) 1017 fd = open(fname, O_RDWR|O_DIRECT, 0644); 1018 else 1019 fd = open(fname, O_RDONLY|O_DIRECT, 0644); 1020 if (fd < 0 || fstat(fd, &st) < 0) { 1021 err(1, "Unable to open/create %s", fname); 1022 /* NOT REACHED */ 1023 } 1024 if (S_ISREG(st.st_mode) && (diskFlags[i] & 1) == 0) { 1025 if (flock(fd, LOCK_EX|LOCK_NB) < 0) { 1026 errx(1, "Disk image %s is already " 1027 "in use\n", fname); 1028 /* NOT REACHED */ 1029 } 1030 } 1031 1032 info = &DiskInfo[DiskNum]; 1033 l = strlen(fname); 1034 1035 info->unit = i; 1036 info->fd = fd; 1037 info->type = type; 1038 info->flags = diskFlags[i]; 1039 memcpy(info->fname, fname, l); 1040 info->serno = NULL; 1041 if (serno) { 1042 if ((info->serno = malloc(SERNOLEN)) != NULL) 1043 strlcpy(info->serno, serno, SERNOLEN); 1044 else 1045 warnx("Couldn't allocate memory for the operation"); 1046 } 1047 1048 if (DiskNum == 0) { 1049 if (type == VKD_CD) { 1050 rootdevnames[0] = "cd9660:vcd0"; 1051 } else if (type == VKD_DISK) { 1052 rootdevnames[0] = "ufs:vkd0s0a"; 1053 rootdevnames[1] = "ufs:vkd0s1a"; 1054 } 1055 } 1056 1057 DiskNum++; 1058 } else { 1059 warnx("vkd%d (%s) > VKDISK_MAX", DiskNum, fname); 1060 continue; 1061 } 1062 } 1063 } 1064 1065 static 1066 int 1067 netif_set_tapflags(int tap_unit, int f, int s) 1068 { 1069 struct ifreq ifr; 1070 int flags; 1071 1072 bzero(&ifr, sizeof(ifr)); 1073 1074 snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit); 1075 if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0) { 1076 warn("tap%d: ioctl(SIOCGIFFLAGS) failed", tap_unit); 1077 return -1; 1078 } 1079 1080 /* 1081 * Adjust if_flags 1082 * 1083 * If the flags are already set/cleared, then we return 1084 * immediately to avoid extra syscalls 1085 */ 1086 flags = (ifr.ifr_flags & 0xffff) | (ifr.ifr_flagshigh << 16); 1087 if (f < 0) { 1088 /* Turn off flags */ 1089 f = -f; 1090 if ((flags & f) == 0) 1091 return 0; 1092 flags &= ~f; 1093 } else { 1094 /* Turn on flags */ 1095 if (flags & f) 1096 return 0; 1097 flags |= f; 1098 } 1099 1100 /* 1101 * Fix up ifreq.ifr_name, since it may be trashed 1102 * in previous ioctl(SIOCGIFFLAGS) 1103 */ 1104 snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit); 1105 1106 ifr.ifr_flags = flags & 0xffff; 1107 ifr.ifr_flagshigh = flags >> 16; 1108 if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0) { 1109 warn("tap%d: ioctl(SIOCSIFFLAGS) failed", tap_unit); 1110 return -1; 1111 } 1112 return 0; 1113 } 1114 1115 static 1116 int 1117 netif_set_tapaddr(int tap_unit, in_addr_t addr, in_addr_t mask, int s) 1118 { 1119 struct ifaliasreq ifra; 1120 struct sockaddr_in *in; 1121 1122 bzero(&ifra, sizeof(ifra)); 1123 snprintf(ifra.ifra_name, sizeof(ifra.ifra_name), "tap%d", tap_unit); 1124 1125 /* Setup address */ 1126 in = (struct sockaddr_in *)&ifra.ifra_addr; 1127 in->sin_family = AF_INET; 1128 in->sin_len = sizeof(*in); 1129 in->sin_addr.s_addr = addr; 1130 1131 if (mask != 0) { 1132 /* Setup netmask */ 1133 in = (struct sockaddr_in *)&ifra.ifra_mask; 1134 in->sin_len = sizeof(*in); 1135 in->sin_addr.s_addr = mask; 1136 } 1137 1138 if (ioctl(s, SIOCAIFADDR, &ifra) < 0) { 1139 warn("tap%d: ioctl(SIOCAIFADDR) failed", tap_unit); 1140 return -1; 1141 } 1142 return 0; 1143 } 1144 1145 static 1146 int 1147 netif_add_tap2brg(int tap_unit, const char *ifbridge, int s) 1148 { 1149 struct ifbreq ifbr; 1150 struct ifdrv ifd; 1151 1152 bzero(&ifbr, sizeof(ifbr)); 1153 snprintf(ifbr.ifbr_ifsname, sizeof(ifbr.ifbr_ifsname), 1154 "tap%d", tap_unit); 1155 1156 bzero(&ifd, sizeof(ifd)); 1157 strlcpy(ifd.ifd_name, ifbridge, sizeof(ifd.ifd_name)); 1158 ifd.ifd_cmd = BRDGADD; 1159 ifd.ifd_len = sizeof(ifbr); 1160 ifd.ifd_data = &ifbr; 1161 1162 if (ioctl(s, SIOCSDRVSPEC, &ifd) < 0) { 1163 /* 1164 * 'errno == EEXIST' means that the tap(4) is already 1165 * a member of the bridge(4) 1166 */ 1167 if (errno != EEXIST) { 1168 warn("ioctl(%s, SIOCSDRVSPEC) failed", ifbridge); 1169 return -1; 1170 } 1171 } 1172 return 0; 1173 } 1174 1175 #define TAPDEV_OFLAGS (O_RDWR | O_NONBLOCK) 1176 1177 /* 1178 * Locate the first unused tap(4) device file if auto mode is requested, 1179 * or open the user supplied device file, and bring up the corresponding 1180 * tap(4) interface. 1181 * 1182 * NOTE: Only tap(4) device file is supported currently 1183 */ 1184 static 1185 int 1186 netif_open_tap(const char *netif, int *tap_unit, int s) 1187 { 1188 char tap_dev[MAXPATHLEN]; 1189 int tap_fd, failed; 1190 struct stat st; 1191 char *dname; 1192 1193 *tap_unit = -1; 1194 1195 if (strcmp(netif, "auto") == 0) { 1196 /* 1197 * Find first unused tap(4) device file 1198 */ 1199 tap_fd = open("/dev/tap", TAPDEV_OFLAGS); 1200 if (tap_fd < 0) { 1201 warnc(errno, "Unable to find a free tap(4)"); 1202 return -1; 1203 } 1204 } else { 1205 /* 1206 * User supplied tap(4) device file or unix socket. 1207 */ 1208 if (netif[0] == '/') /* Absolute path */ 1209 strlcpy(tap_dev, netif, sizeof(tap_dev)); 1210 else 1211 snprintf(tap_dev, sizeof(tap_dev), "/dev/%s", netif); 1212 1213 tap_fd = open(tap_dev, TAPDEV_OFLAGS); 1214 1215 /* 1216 * If we cannot open normally try to connect to it. 1217 */ 1218 if (tap_fd < 0) 1219 tap_fd = unix_connect(tap_dev); 1220 1221 if (tap_fd < 0) { 1222 warn("Unable to open %s", tap_dev); 1223 return -1; 1224 } 1225 } 1226 1227 /* 1228 * Check whether the device file is a tap(4) 1229 */ 1230 if (fstat(tap_fd, &st) < 0) { 1231 failed = 1; 1232 } else if (S_ISCHR(st.st_mode)) { 1233 dname = fdevname(tap_fd); 1234 if (dname) 1235 dname = strstr(dname, "tap"); 1236 if (dname) { 1237 /* 1238 * Bring up the corresponding tap(4) interface 1239 */ 1240 *tap_unit = strtol(dname + 3, NULL, 10); 1241 printf("TAP UNIT %d\n", *tap_unit); 1242 if (netif_set_tapflags(*tap_unit, IFF_UP, s) == 0) 1243 failed = 0; 1244 else 1245 failed = 1; 1246 } else { 1247 failed = 1; 1248 } 1249 } else if (S_ISSOCK(st.st_mode)) { 1250 /* 1251 * Special socket connection (typically to vknet). We 1252 * do not have to do anything. 1253 */ 1254 failed = 0; 1255 } else { 1256 failed = 1; 1257 } 1258 1259 if (failed) { 1260 warnx("%s is not a tap(4) device or socket", tap_dev); 1261 close(tap_fd); 1262 tap_fd = -1; 1263 *tap_unit = -1; 1264 } 1265 return tap_fd; 1266 } 1267 1268 static int 1269 unix_connect(const char *path) 1270 { 1271 struct sockaddr_un sunx; 1272 int len; 1273 int net_fd; 1274 int sndbuf = 262144; 1275 struct stat st; 1276 1277 snprintf(sunx.sun_path, sizeof(sunx.sun_path), "%s", path); 1278 len = offsetof(struct sockaddr_un, sun_path[strlen(sunx.sun_path)]); 1279 ++len; /* include nul */ 1280 sunx.sun_family = AF_UNIX; 1281 sunx.sun_len = len; 1282 1283 net_fd = socket(AF_UNIX, SOCK_SEQPACKET, 0); 1284 if (net_fd < 0) 1285 return(-1); 1286 if (connect(net_fd, (void *)&sunx, len) < 0) { 1287 close(net_fd); 1288 return(-1); 1289 } 1290 setsockopt(net_fd, SOL_SOCKET, SO_SNDBUF, &sndbuf, sizeof(sndbuf)); 1291 if (fstat(net_fd, &st) == 0) 1292 printf("Network socket buffer: %ld bytes\n", st.st_blksize); 1293 fcntl(net_fd, F_SETFL, O_NONBLOCK); 1294 return(net_fd); 1295 } 1296 1297 #undef TAPDEV_MAJOR 1298 #undef TAPDEV_MINOR 1299 #undef TAPDEV_OFLAGS 1300 1301 /* 1302 * Following syntax is supported, 1303 * 1) x.x.x.x tap(4)'s address is x.x.x.x 1304 * 1305 * 2) x.x.x.x/z tap(4)'s address is x.x.x.x 1306 * tap(4)'s netmask len is z 1307 * 1308 * 3) x.x.x.x:y.y.y.y tap(4)'s address is x.x.x.x 1309 * pseudo netif's address is y.y.y.y 1310 * 1311 * 4) x.x.x.x:y.y.y.y/z tap(4)'s address is x.x.x.x 1312 * pseudo netif's address is y.y.y.y 1313 * tap(4) and pseudo netif's netmask len are z 1314 * 1315 * 5) bridgeX tap(4) will be added to bridgeX 1316 * 1317 * 6) bridgeX:y.y.y.y tap(4) will be added to bridgeX 1318 * pseudo netif's address is y.y.y.y 1319 * 1320 * 7) bridgeX:y.y.y.y/z tap(4) will be added to bridgeX 1321 * pseudo netif's address is y.y.y.y 1322 * pseudo netif's netmask len is z 1323 */ 1324 static 1325 int 1326 netif_init_tap(int tap_unit, in_addr_t *addr, in_addr_t *mask, int s) 1327 { 1328 in_addr_t tap_addr, netmask, netif_addr; 1329 int next_netif_addr; 1330 char *tok, *masklen_str, *ifbridge; 1331 1332 *addr = 0; 1333 *mask = 0; 1334 1335 tok = strtok(NULL, ":/"); 1336 if (tok == NULL) { 1337 /* 1338 * Nothing special, simply use tap(4) as backend 1339 */ 1340 return 0; 1341 } 1342 1343 if (inet_pton(AF_INET, tok, &tap_addr) > 0) { 1344 /* 1345 * tap(4)'s address is supplied 1346 */ 1347 ifbridge = NULL; 1348 1349 /* 1350 * If there is next token, then it may be pseudo 1351 * netif's address or netmask len for tap(4) 1352 */ 1353 next_netif_addr = 0; 1354 } else { 1355 /* 1356 * Not tap(4)'s address, assume it as a bridge(4) 1357 * iface name 1358 */ 1359 tap_addr = 0; 1360 ifbridge = tok; 1361 1362 /* 1363 * If there is next token, then it must be pseudo 1364 * netif's address 1365 */ 1366 next_netif_addr = 1; 1367 } 1368 1369 netmask = netif_addr = 0; 1370 1371 tok = strtok(NULL, ":/"); 1372 if (tok == NULL) 1373 goto back; 1374 1375 if (inet_pton(AF_INET, tok, &netif_addr) <= 0) { 1376 if (next_netif_addr) { 1377 warnx("Invalid pseudo netif address: %s", tok); 1378 return -1; 1379 } 1380 netif_addr = 0; 1381 1382 /* 1383 * Current token is not address, then it must be netmask len 1384 */ 1385 masklen_str = tok; 1386 } else { 1387 /* 1388 * Current token is pseudo netif address, if there is next token 1389 * it must be netmask len 1390 */ 1391 masklen_str = strtok(NULL, "/"); 1392 } 1393 1394 /* Calculate netmask */ 1395 if (masklen_str != NULL) { 1396 u_long masklen; 1397 1398 masklen = strtoul(masklen_str, NULL, 10); 1399 if (masklen < 32 && masklen > 0) { 1400 netmask = 1401 htonl(rounddown2(0xffffffff, 1LL << (32 - masklen))); 1402 } else { 1403 warnx("Invalid netmask len: %lu", masklen); 1404 return -1; 1405 } 1406 } 1407 1408 /* Make sure there is no more token left */ 1409 if (strtok(NULL, ":/") != NULL) { 1410 warnx("Invalid argument to '-I'"); 1411 return -1; 1412 } 1413 1414 back: 1415 if (tap_unit < 0) { 1416 /* Do nothing */ 1417 } else if (ifbridge == NULL) { 1418 /* Set tap(4) address/netmask */ 1419 if (netif_set_tapaddr(tap_unit, tap_addr, netmask, s) < 0) 1420 return -1; 1421 } else { 1422 /* Tie tap(4) to bridge(4) */ 1423 if (netif_add_tap2brg(tap_unit, ifbridge, s) < 0) 1424 return -1; 1425 } 1426 1427 *addr = netif_addr; 1428 *mask = netmask; 1429 return 0; 1430 } 1431 1432 /* 1433 * NetifInfo[] will be filled for pseudo netif initialization. 1434 * NetifNum will be bumped to reflect the number of valid entries 1435 * in NetifInfo[]. 1436 */ 1437 static 1438 void 1439 init_netif(char *netifExp[], int netifExpNum) 1440 { 1441 int i, s; 1442 char *tmp; 1443 1444 if (netifExpNum == 0) 1445 return; 1446 1447 s = socket(AF_INET, SOCK_DGRAM, 0); /* for ioctl(SIOC) */ 1448 if (s < 0) 1449 return; 1450 1451 for (i = 0; i < netifExpNum; ++i) { 1452 struct vknetif_info *info; 1453 in_addr_t netif_addr, netif_mask; 1454 int tap_fd, tap_unit; 1455 char *netif; 1456 1457 /* Extract MAC address if there is one */ 1458 tmp = netifExp[i]; 1459 strsep(&tmp, "="); 1460 1461 netif = strtok(netifExp[i], ":"); 1462 if (netif == NULL) { 1463 warnx("Invalid argument to '-I'"); 1464 continue; 1465 } 1466 1467 /* 1468 * Open tap(4) device file and bring up the 1469 * corresponding interface 1470 */ 1471 tap_fd = netif_open_tap(netif, &tap_unit, s); 1472 if (tap_fd < 0) 1473 continue; 1474 1475 /* 1476 * Initialize tap(4) and get address/netmask 1477 * for pseudo netif 1478 * 1479 * NB: Rest part of netifExp[i] is passed 1480 * to netif_init_tap() implicitly. 1481 */ 1482 if (netif_init_tap(tap_unit, &netif_addr, &netif_mask, s) < 0) { 1483 /* 1484 * NB: Closing tap(4) device file will bring 1485 * down the corresponding interface 1486 */ 1487 close(tap_fd); 1488 continue; 1489 } 1490 1491 info = &NetifInfo[NetifNum]; 1492 bzero(info, sizeof(*info)); 1493 info->tap_fd = tap_fd; 1494 info->tap_unit = tap_unit; 1495 info->netif_addr = netif_addr; 1496 info->netif_mask = netif_mask; 1497 /* 1498 * If tmp isn't NULL it means a MAC could have been 1499 * specified so attempt to convert it. 1500 * Setting enaddr to NULL will tell vke_attach() we 1501 * need a pseudo-random MAC address. 1502 */ 1503 if (tmp != NULL) { 1504 if ((info->enaddr = malloc(ETHER_ADDR_LEN)) == NULL) 1505 warnx("Couldn't allocate memory for the operation"); 1506 else { 1507 if ((kether_aton(tmp, info->enaddr)) == NULL) { 1508 free(info->enaddr); 1509 info->enaddr = NULL; 1510 } 1511 } 1512 } 1513 1514 NetifNum++; 1515 if (NetifNum >= VKNETIF_MAX) /* XXX will this happen? */ 1516 break; 1517 } 1518 close(s); 1519 } 1520 1521 /* 1522 * Create the pid file and leave it open and locked while the vkernel is 1523 * running. This allows a script to use /usr/bin/lockf to probe whether 1524 * a vkernel is still running (so as not to accidently kill an unrelated 1525 * process from a stale pid file). 1526 */ 1527 static 1528 void 1529 writepid(void) 1530 { 1531 char buf[32]; 1532 int fd; 1533 1534 if (pid_file != NULL) { 1535 snprintf(buf, sizeof(buf), "%ld\n", (long)getpid()); 1536 fd = open(pid_file, O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0666); 1537 if (fd < 0) { 1538 if (errno == EWOULDBLOCK) { 1539 perror("Failed to lock pidfile, " 1540 "vkernel already running"); 1541 } else { 1542 perror("Failed to create pidfile"); 1543 } 1544 exit(EX_SOFTWARE); 1545 } 1546 ftruncate(fd, 0); 1547 write(fd, buf, strlen(buf)); 1548 /* leave the file open to maintain the lock */ 1549 } 1550 } 1551 1552 static 1553 void 1554 cleanpid( void ) 1555 { 1556 if (pid_file != NULL) { 1557 if (unlink(pid_file) < 0) 1558 perror("Warning: couldn't remove pidfile"); 1559 } 1560 } 1561 1562 static 1563 void 1564 usage_err(const char *ctl, ...) 1565 { 1566 va_list va; 1567 1568 va_start(va, ctl); 1569 vfprintf(stderr, ctl, va); 1570 va_end(va); 1571 fprintf(stderr, "\n"); 1572 exit(EX_USAGE); 1573 } 1574 1575 static 1576 void 1577 usage_help(_Bool help) 1578 { 1579 fprintf(stderr, "Usage: %s [-hsUvdt] [-c file] [-e name=value:name=value:...]\n" 1580 "\t[-i file] [-I interface[:address1[:address2][/netmask]]] [-l cpulock]\n" 1581 "\t[-m size] [-n numcpus[:lbits[:cbits]]]\n" 1582 "\t[-p file] [-r file]\n", save_av[0]); 1583 1584 if (help) 1585 fprintf(stderr, "\nArguments:\n" 1586 "\t-c\tSpecify a readonly CD-ROM image file to be used by the kernel.\n" 1587 "\t-e\tSpecify an environment to be used by the kernel.\n" 1588 "\t-h\tThis list of options.\n" 1589 "\t-i\tSpecify a memory image file to be used by the virtual kernel.\n" 1590 "\t-I\tCreate a virtual network device.\n" 1591 "\t-l\tSpecify which, if any, real CPUs to lock virtual CPUs to.\n" 1592 "\t-m\tSpecify the amount of memory to be used by the kernel in bytes.\n" 1593 "\t-n\tSpecify the number of CPUs and the topology you wish to emulate:\n" 1594 "\t\t\tnumcpus - number of cpus\n" 1595 "\t\t\tlbits - specify the number of bits within APICID(=CPUID)\n" 1596 "\t\t\t needed for representing the logical ID.\n" 1597 "\t\t\t Controls the number of threads/core:\n" 1598 "\t\t\t (0 bits - 1 thread, 1 bit - 2 threads).\n" 1599 "\t\t\tcbits - specify the number of bits within APICID(=CPUID)\n" 1600 "\t\t\t needed for representing the core ID.\n" 1601 "\t\t\t Controls the number of cores/package:\n" 1602 "\t\t\t (0 bits - 1 core, 1 bit - 2 cores).\n" 1603 "\t-p\tSpecify a file in which to store the process ID.\n" 1604 "\t-r\tSpecify a R/W disk image file, iterates vkd0..n\n" 1605 "\t-R\tSpecify a COW disk image file, iterates vkd0..n\n" 1606 "\t-s\tBoot into single-user mode.\n" 1607 "\t-t\tUse a precise host timer when calculating clock values.\n" 1608 "\t-U\tEnable writing to kernel memory and module loading.\n" 1609 "\t-v\tTurn on verbose booting.\n"); 1610 1611 exit(EX_USAGE); 1612 } 1613 1614 void 1615 cpu_smp_stopped(void) 1616 { 1617 } 1618 1619 void 1620 cpu_reset(void) 1621 { 1622 kprintf("cpu reset, rebooting vkernel\n"); 1623 closefrom(3); 1624 cleanpid(); 1625 exit(EX_VKERNEL_REBOOT); 1626 } 1627 1628 void 1629 cpu_halt(void) 1630 { 1631 kprintf("cpu halt, exiting vkernel\n"); 1632 cleanpid(); 1633 exit(EX_OK); 1634 } 1635 1636 void 1637 setrealcpu(void) 1638 { 1639 switch(lwp_cpu_lock) { 1640 case LCL_PER_CPU: 1641 if (bootverbose) 1642 kprintf("Locking CPU%d to real cpu %d\n", 1643 mycpuid, next_cpu); 1644 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu)); 1645 next_cpu++; 1646 if (next_cpu >= real_ncpus) 1647 next_cpu = 0; 1648 break; 1649 case LCL_SINGLE_CPU: 1650 if (bootverbose) 1651 kprintf("Locking CPU%d to real cpu %d\n", 1652 mycpuid, next_cpu); 1653 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu)); 1654 break; 1655 default: 1656 /* do not map virtual cpus to real cpus */ 1657 break; 1658 } 1659 } 1660 1661 /* 1662 * Allocate and free memory for module loading. The loaded module 1663 * has to be placed somewhere near the current kernel binary load 1664 * point or the relocations will not work. 1665 * 1666 * I'm not sure why this isn't working. 1667 */ 1668 int 1669 vkernel_module_memory_alloc(vm_offset_t *basep, size_t bytes) 1670 { 1671 #if 1 1672 size_t xtra; 1673 xtra = (PAGE_SIZE - (vm_offset_t)sbrk(0)) & PAGE_MASK; 1674 *basep = (vm_offset_t)sbrk(xtra + bytes) + xtra; 1675 bzero((void *)*basep, bytes); 1676 #else 1677 *basep = (vm_offset_t)mmap((void *)0x000000000, bytes, 1678 PROT_READ|PROT_WRITE|PROT_EXEC, 1679 MAP_ANON|MAP_SHARED, -1, 0); 1680 if ((void *)*basep == MAP_FAILED) 1681 return ENOMEM; 1682 #endif 1683 return 0; 1684 } 1685 1686 void 1687 vkernel_module_memory_free(vm_offset_t base, size_t bytes) 1688 { 1689 #if 0 1690 #if 0 1691 munmap((void *)base, bytes); 1692 #endif 1693 #endif 1694 } 1695 1696 /* 1697 * VKERNEL64 implementation functions using ptrheads. 1698 */ 1699 void 1700 vkernel_yield(void) 1701 { 1702 pthread_yield(); 1703 } 1704