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 * $DragonFly: src/sys/platform/vkernel/platform/init.c,v 1.56 2008/05/27 07:48:00 dillon Exp $ 35 */ 36 37 #include <sys/types.h> 38 #include <sys/systm.h> 39 #include <sys/kernel.h> 40 #include <sys/stat.h> 41 #include <sys/mman.h> 42 #include <sys/cons.h> 43 #include <sys/random.h> 44 #include <sys/vkernel.h> 45 #include <sys/tls.h> 46 #include <sys/reboot.h> 47 #include <sys/proc.h> 48 #include <sys/msgbuf.h> 49 #include <sys/vmspace.h> 50 #include <sys/socket.h> 51 #include <sys/sockio.h> 52 #include <sys/sysctl.h> 53 #include <sys/un.h> 54 #include <vm/vm_page.h> 55 #include <sys/mplock2.h> 56 57 #include <machine/cpu.h> 58 #include <machine/globaldata.h> 59 #include <machine/tls.h> 60 #include <machine/md_var.h> 61 #include <machine/vmparam.h> 62 #include <cpu/specialreg.h> 63 64 #include <net/if.h> 65 #include <net/if_arp.h> 66 #include <net/ethernet.h> 67 #include <net/bridge/if_bridgevar.h> 68 #include <netinet/in.h> 69 #include <arpa/inet.h> 70 71 #include <stdio.h> 72 #include <stdlib.h> 73 #include <stdarg.h> 74 #include <stdbool.h> 75 #include <unistd.h> 76 #include <fcntl.h> 77 #include <string.h> 78 #include <err.h> 79 #include <errno.h> 80 #include <assert.h> 81 #include <sysexits.h> 82 83 vm_paddr_t phys_avail[16]; 84 vm_paddr_t Maxmem; 85 vm_paddr_t Maxmem_bytes; 86 int physmem; 87 int MemImageFd = -1; 88 struct vkdisk_info DiskInfo[VKDISK_MAX]; 89 int DiskNum; 90 struct vknetif_info NetifInfo[VKNETIF_MAX]; 91 int NetifNum; 92 char *pid_file; 93 vm_offset_t KvaStart; 94 vm_offset_t KvaEnd; 95 vm_offset_t KvaSize; 96 vm_offset_t virtual_start; 97 vm_offset_t virtual_end; 98 vm_offset_t virtual2_start; 99 vm_offset_t virtual2_end; 100 vm_offset_t kernel_vm_end; 101 vm_offset_t crashdumpmap; 102 vm_offset_t clean_sva; 103 vm_offset_t clean_eva; 104 struct msgbuf *msgbufp; 105 caddr_t ptvmmap; 106 vpte_t *KernelPTD; 107 vpte_t *KernelPTA; /* Warning: Offset for direct VA translation */ 108 void *dmap_min_address; 109 u_int cpu_feature; /* XXX */ 110 int tsc_present; 111 int64_t tsc_frequency; 112 int optcpus; /* number of cpus - see mp_start() */ 113 int lwp_cpu_lock; /* if/how to lock virtual CPUs to real CPUs */ 114 int real_ncpus; /* number of real CPUs */ 115 int next_cpu; /* next real CPU to lock a virtual CPU to */ 116 117 struct privatespace *CPU_prvspace; 118 119 static struct trapframe proc0_tf; 120 static void *proc0paddr; 121 122 static void init_sys_memory(char *imageFile); 123 static void init_kern_memory(void); 124 static void init_globaldata(void); 125 static void init_vkernel(void); 126 static void init_disk(char *diskExp[], int diskFileNum, enum vkdisk_type type); 127 static void init_netif(char *netifExp[], int netifFileNum); 128 static void writepid(void); 129 static void cleanpid(void); 130 static int unix_connect(const char *path); 131 static void usage_err(const char *ctl, ...); 132 static void usage_help(_Bool); 133 134 static int save_ac; 135 static char **save_av; 136 137 /* 138 * Kernel startup for virtual kernels - standard main() 139 */ 140 int 141 main(int ac, char **av) 142 { 143 char *memImageFile = NULL; 144 char *netifFile[VKNETIF_MAX]; 145 char *diskFile[VKDISK_MAX]; 146 char *cdFile[VKDISK_MAX]; 147 char *suffix; 148 char *endp; 149 int netifFileNum = 0; 150 int diskFileNum = 0; 151 int cdFileNum = 0; 152 int bootOnDisk = -1; /* set below to vcd (0) or vkd (1) */ 153 int c; 154 int i; 155 int j; 156 int n; 157 int isq; 158 int real_vkernel_enable; 159 int supports_sse; 160 size_t vsize; 161 162 save_ac = ac; 163 save_av = av; 164 165 /* 166 * Process options 167 */ 168 kernel_mem_readonly = 1; 169 #ifdef SMP 170 optcpus = 2; 171 #endif 172 lwp_cpu_lock = LCL_NONE; 173 174 real_vkernel_enable = 0; 175 vsize = sizeof(real_vkernel_enable); 176 sysctlbyname("vm.vkernel_enable", &real_vkernel_enable, &vsize, NULL,0); 177 178 if (real_vkernel_enable == 0) { 179 errx(1, "vm.vkernel_enable is 0, must be set " 180 "to 1 to execute a vkernel!"); 181 } 182 183 real_ncpus = 1; 184 vsize = sizeof(real_ncpus); 185 sysctlbyname("hw.ncpu", &real_ncpus, &vsize, NULL, 0); 186 187 if (ac < 2) 188 usage_help(false); 189 190 while ((c = getopt(ac, av, "c:hsvl:m:n:r:e:i:p:I:U")) != -1) { 191 switch(c) { 192 case 'e': 193 /* 194 * name=value:name=value:name=value... 195 * name="value"... 196 * 197 * Allow values to be quoted but note that shells 198 * may remove the quotes, so using this feature 199 * to embed colons may require a backslash. 200 */ 201 n = strlen(optarg); 202 isq = 0; 203 kern_envp = malloc(n + 2); 204 for (i = j = 0; i < n; ++i) { 205 if (optarg[i] == '"') 206 isq ^= 1; 207 else if (optarg[i] == '\'') 208 isq ^= 2; 209 else if (isq == 0 && optarg[i] == ':') 210 kern_envp[j++] = 0; 211 else 212 kern_envp[j++] = optarg[i]; 213 } 214 kern_envp[j++] = 0; 215 kern_envp[j++] = 0; 216 break; 217 case 's': 218 boothowto |= RB_SINGLE; 219 break; 220 case 'v': 221 bootverbose = 1; 222 break; 223 case 'i': 224 memImageFile = optarg; 225 break; 226 case 'I': 227 if (netifFileNum < VKNETIF_MAX) 228 netifFile[netifFileNum++] = strdup(optarg); 229 break; 230 case 'r': 231 if (bootOnDisk < 0) 232 bootOnDisk = 1; 233 if (diskFileNum + cdFileNum < VKDISK_MAX) 234 diskFile[diskFileNum++] = strdup(optarg); 235 break; 236 case 'c': 237 if (bootOnDisk < 0) 238 bootOnDisk = 0; 239 if (diskFileNum + cdFileNum < VKDISK_MAX) 240 cdFile[cdFileNum++] = strdup(optarg); 241 break; 242 case 'm': 243 Maxmem_bytes = strtoull(optarg, &suffix, 0); 244 if (suffix) { 245 switch(*suffix) { 246 case 'g': 247 case 'G': 248 Maxmem_bytes <<= 30; 249 break; 250 case 'm': 251 case 'M': 252 Maxmem_bytes <<= 20; 253 break; 254 case 'k': 255 case 'K': 256 Maxmem_bytes <<= 10; 257 break; 258 default: 259 Maxmem_bytes = 0; 260 usage_err("Bad maxmem option"); 261 /* NOT REACHED */ 262 break; 263 } 264 } 265 break; 266 case 'l': 267 next_cpu = -1; 268 if (strncmp("map", optarg, 3) == 0) { 269 lwp_cpu_lock = LCL_PER_CPU; 270 if (optarg[3] == ',') { 271 next_cpu = strtol(optarg+4, &endp, 0); 272 if (*endp != '\0') 273 usage_err("Bad target CPU number at '%s'", endp); 274 } else { 275 next_cpu = 0; 276 } 277 if (next_cpu < 0 || next_cpu > real_ncpus - 1) 278 usage_err("Bad target CPU, valid range is 0-%d", real_ncpus - 1); 279 } else if (strncmp("any", optarg, 3) == 0) { 280 lwp_cpu_lock = LCL_NONE; 281 } else { 282 lwp_cpu_lock = LCL_SINGLE_CPU; 283 next_cpu = strtol(optarg, &endp, 0); 284 if (*endp != '\0') 285 usage_err("Bad target CPU number at '%s'", endp); 286 if (next_cpu < 0 || next_cpu > real_ncpus - 1) 287 usage_err("Bad target CPU, valid range is 0-%d", real_ncpus - 1); 288 } 289 break; 290 case 'n': 291 /* 292 * This value is set up by mp_start(), don't just 293 * set ncpus here. 294 */ 295 #ifdef SMP 296 optcpus = strtol(optarg, NULL, 0); 297 if (optcpus < 1 || optcpus > MAXCPU) 298 usage_err("Bad ncpus, valid range is 1-%d", MAXCPU); 299 #else 300 if (strtol(optarg, NULL, 0) != 1) { 301 usage_err("You built a UP vkernel, only 1 cpu!"); 302 } 303 #endif 304 305 break; 306 case 'p': 307 pid_file = optarg; 308 break; 309 case 'U': 310 kernel_mem_readonly = 0; 311 break; 312 case 'h': 313 usage_help(true); 314 break; 315 default: 316 usage_help(false); 317 } 318 } 319 320 writepid(); 321 cpu_disable_intr(); 322 init_sys_memory(memImageFile); 323 init_kern_memory(); 324 init_globaldata(); 325 init_vkernel(); 326 setrealcpu(); 327 init_kqueue(); 328 329 /* 330 * Check TSC 331 */ 332 vsize = sizeof(tsc_present); 333 sysctlbyname("hw.tsc_present", &tsc_present, &vsize, NULL, 0); 334 vsize = sizeof(tsc_frequency); 335 sysctlbyname("hw.tsc_frequency", &tsc_frequency, &vsize, NULL, 0); 336 if (tsc_present) 337 cpu_feature |= CPUID_TSC; 338 339 /* 340 * Check SSE 341 */ 342 vsize = sizeof(supports_sse); 343 supports_sse = 0; 344 sysctlbyname("hw.instruction_sse", &supports_sse, &vsize, NULL, 0); 345 init_fpu(supports_sse); 346 if (supports_sse) 347 cpu_feature |= CPUID_SSE | CPUID_FXSR; 348 349 /* 350 * We boot from the first installed disk. 351 */ 352 if (bootOnDisk == 1) { 353 init_disk(diskFile, diskFileNum, VKD_DISK); 354 init_disk(cdFile, cdFileNum, VKD_CD); 355 } else { 356 init_disk(cdFile, cdFileNum, VKD_CD); 357 init_disk(diskFile, diskFileNum, VKD_DISK); 358 } 359 init_netif(netifFile, netifFileNum); 360 init_exceptions(); 361 mi_startup(); 362 /* NOT REACHED */ 363 exit(EX_SOFTWARE); 364 } 365 366 /* 367 * Initialize system memory. This is the virtual kernel's 'RAM'. 368 */ 369 static 370 void 371 init_sys_memory(char *imageFile) 372 { 373 struct stat st; 374 int i; 375 int fd; 376 377 /* 378 * Figure out the system memory image size. If an image file was 379 * specified and -m was not specified, use the image file's size. 380 */ 381 382 if (imageFile && stat(imageFile, &st) == 0 && Maxmem_bytes == 0) 383 Maxmem_bytes = (vm_paddr_t)st.st_size; 384 if ((imageFile == NULL || stat(imageFile, &st) < 0) && 385 Maxmem_bytes == 0) { 386 err(1, "Cannot create new memory file %s unless " 387 "system memory size is specified with -m", 388 imageFile); 389 /* NOT REACHED */ 390 } 391 392 /* 393 * Maxmem must be known at this time 394 */ 395 if (Maxmem_bytes < 32 * 1024 * 1024 || (Maxmem_bytes & SEG_MASK)) { 396 err(1, "Bad maxmem specification: 32MB minimum, " 397 "multiples of %dMB only", 398 SEG_SIZE / 1024 / 1024); 399 /* NOT REACHED */ 400 } 401 402 /* 403 * Generate an image file name if necessary, then open/create the 404 * file exclusively locked. Do not allow multiple virtual kernels 405 * to use the same image file. 406 */ 407 if (imageFile == NULL) { 408 for (i = 0; i < 1000000; ++i) { 409 asprintf(&imageFile, "/var/vkernel/memimg.%06d", i); 410 fd = open(imageFile, 411 O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644); 412 if (fd < 0 && errno == EWOULDBLOCK) { 413 free(imageFile); 414 continue; 415 } 416 break; 417 } 418 } else { 419 fd = open(imageFile, O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644); 420 } 421 fprintf(stderr, "Using memory file: %s\n", imageFile); 422 if (fd < 0 || fstat(fd, &st) < 0) { 423 err(1, "Unable to open/create %s", imageFile); 424 /* NOT REACHED */ 425 } 426 427 /* 428 * Truncate or extend the file as necessary. Clean out the contents 429 * of the file, we want it to be full of holes so we don't waste 430 * time reading in data from an old file that we no longer care 431 * about. 432 */ 433 ftruncate(fd, 0); 434 ftruncate(fd, Maxmem_bytes); 435 436 MemImageFd = fd; 437 Maxmem = Maxmem_bytes >> PAGE_SHIFT; 438 physmem = Maxmem; 439 } 440 441 /* 442 * Initialize kernel memory. This reserves kernel virtual memory by using 443 * MAP_VPAGETABLE 444 */ 445 446 static 447 void 448 init_kern_memory(void) 449 { 450 void *base; 451 void *try; 452 char dummy; 453 char *topofstack = &dummy; 454 int i; 455 void *firstfree; 456 457 /* 458 * Memory map our kernel virtual memory space. Note that the 459 * kernel image itself is not made part of this memory for the 460 * moment. 461 * 462 * The memory map must be segment-aligned so we can properly 463 * offset KernelPTD. 464 * 465 * If the system kernel has a different MAXDSIZ, it might not 466 * be possible to map kernel memory in its prefered location. 467 * Try a number of different locations. 468 */ 469 try = (void *)(512UL << 30); 470 base = NULL; 471 while ((char *)try + KERNEL_KVA_SIZE < topofstack) { 472 base = mmap(try, KERNEL_KVA_SIZE, PROT_READ|PROT_WRITE, 473 MAP_FILE|MAP_SHARED|MAP_VPAGETABLE, 474 MemImageFd, (off_t)try); 475 if (base == try) 476 break; 477 if (base != MAP_FAILED) 478 munmap(base, KERNEL_KVA_SIZE); 479 try = (char *)try + (512UL << 30); 480 } 481 if (base != try) { 482 err(1, "Unable to mmap() kernel virtual memory!"); 483 /* NOT REACHED */ 484 } 485 madvise(base, KERNEL_KVA_SIZE, MADV_NOSYNC); 486 KvaStart = (vm_offset_t)base; 487 KvaSize = KERNEL_KVA_SIZE; 488 KvaEnd = KvaStart + KvaSize; 489 490 /* cannot use kprintf yet */ 491 printf("KVM mapped at %p-%p\n", (void *)KvaStart, (void *)KvaEnd); 492 493 /* MAP_FILE? */ 494 dmap_min_address = mmap(0, DMAP_SIZE, PROT_READ|PROT_WRITE, 495 MAP_NOCORE|MAP_NOSYNC|MAP_SHARED, 496 MemImageFd, 0); 497 if (dmap_min_address == MAP_FAILED) { 498 err(1, "Unable to mmap() kernel DMAP region!"); 499 /* NOT REACHED */ 500 } 501 502 firstfree = 0; 503 pmap_bootstrap((vm_paddr_t *)&firstfree, (int64_t)base); 504 505 mcontrol(base, KERNEL_KVA_SIZE, MADV_SETMAP, 506 0 | VPTE_R | VPTE_W | VPTE_V); 507 508 /* 509 * phys_avail[] represents unallocated physical memory. MI code 510 * will use phys_avail[] to create the vm_page array. 511 */ 512 phys_avail[0] = (vm_paddr_t)firstfree; 513 phys_avail[0] = (phys_avail[0] + PAGE_MASK) & ~(vm_paddr_t)PAGE_MASK; 514 phys_avail[1] = Maxmem_bytes; 515 516 #if JGV 517 /* 518 * (virtual_start, virtual_end) represent unallocated kernel virtual 519 * memory. MI code will create kernel_map using these parameters. 520 */ 521 virtual_start = KvaStart + (long)firstfree; 522 virtual_start = (virtual_start + PAGE_MASK) & ~(vm_offset_t)PAGE_MASK; 523 virtual_end = KvaStart + KERNEL_KVA_SIZE; 524 #endif 525 526 /* 527 * pmap_growkernel() will set the correct value. 528 */ 529 kernel_vm_end = 0; 530 531 /* 532 * Allocate space for process 0's UAREA. 533 */ 534 proc0paddr = (void *)virtual_start; 535 for (i = 0; i < UPAGES; ++i) { 536 pmap_kenter_quick(virtual_start, phys_avail[0]); 537 virtual_start += PAGE_SIZE; 538 phys_avail[0] += PAGE_SIZE; 539 } 540 541 /* 542 * crashdumpmap 543 */ 544 crashdumpmap = virtual_start; 545 virtual_start += MAXDUMPPGS * PAGE_SIZE; 546 547 /* 548 * msgbufp maps the system message buffer 549 */ 550 assert((MSGBUF_SIZE & PAGE_MASK) == 0); 551 msgbufp = (void *)virtual_start; 552 for (i = 0; i < (MSGBUF_SIZE >> PAGE_SHIFT); ++i) { 553 pmap_kenter_quick(virtual_start, phys_avail[0]); 554 virtual_start += PAGE_SIZE; 555 phys_avail[0] += PAGE_SIZE; 556 } 557 msgbufinit(msgbufp, MSGBUF_SIZE); 558 559 /* 560 * used by kern_memio for /dev/mem access 561 */ 562 ptvmmap = (caddr_t)virtual_start; 563 virtual_start += PAGE_SIZE; 564 565 /* 566 * Bootstrap the kernel_pmap 567 */ 568 #if JGV 569 pmap_bootstrap(); 570 #endif 571 } 572 573 /* 574 * Map the per-cpu globaldata for cpu #0. Allocate the space using 575 * virtual_start and phys_avail[0] 576 */ 577 static 578 void 579 init_globaldata(void) 580 { 581 int i; 582 vm_paddr_t pa; 583 vm_offset_t va; 584 585 /* 586 * Reserve enough KVA to cover possible cpus. This is a considerable 587 * amount of KVA since the privatespace structure includes two 588 * whole page table mappings. 589 */ 590 virtual_start = (virtual_start + SEG_MASK) & ~(vm_offset_t)SEG_MASK; 591 CPU_prvspace = (void *)virtual_start; 592 virtual_start += sizeof(struct privatespace) * SMP_MAXCPU; 593 594 /* 595 * Allocate enough physical memory to cover the mdglobaldata 596 * portion of the space and the idle stack and map the pages 597 * into KVA. For cpu #0 only. 598 */ 599 for (i = 0; i < sizeof(struct mdglobaldata); i += PAGE_SIZE) { 600 pa = phys_avail[0]; 601 va = (vm_offset_t)&CPU_prvspace[0].mdglobaldata + i; 602 pmap_kenter_quick(va, pa); 603 phys_avail[0] += PAGE_SIZE; 604 } 605 for (i = 0; i < sizeof(CPU_prvspace[0].idlestack); i += PAGE_SIZE) { 606 pa = phys_avail[0]; 607 va = (vm_offset_t)&CPU_prvspace[0].idlestack + i; 608 pmap_kenter_quick(va, pa); 609 phys_avail[0] += PAGE_SIZE; 610 } 611 612 /* 613 * Setup the %gs for cpu #0. The mycpu macro works after this 614 * point. Note that %fs is used by pthreads. 615 */ 616 tls_set_gs(&CPU_prvspace[0], sizeof(struct privatespace)); 617 } 618 619 /* 620 * Initialize very low level systems including thread0, proc0, etc. 621 */ 622 static 623 void 624 init_vkernel(void) 625 { 626 struct mdglobaldata *gd; 627 628 gd = &CPU_prvspace[0].mdglobaldata; 629 bzero(gd, sizeof(*gd)); 630 631 gd->mi.gd_curthread = &thread0; 632 thread0.td_gd = &gd->mi; 633 ncpus = 1; 634 ncpus2 = 1; /* rounded down power of 2 */ 635 ncpus_fit = 1; /* rounded up power of 2 */ 636 /* ncpus2_mask and ncpus_fit_mask are 0 */ 637 init_param1(); 638 gd->mi.gd_prvspace = &CPU_prvspace[0]; 639 mi_gdinit(&gd->mi, 0); 640 cpu_gdinit(gd, 0); 641 mi_proc0init(&gd->mi, proc0paddr); 642 lwp0.lwp_md.md_regs = &proc0_tf; 643 644 /*init_locks();*/ 645 #ifdef SMP 646 /* 647 * Get the initial mplock with a count of 1 for the BSP. 648 * This uses a LOGICAL cpu ID, ie BSP == 0. 649 */ 650 cpu_get_initial_mplock(); 651 #endif 652 cninit(); 653 rand_initialize(); 654 #if 0 /* #ifdef DDB */ 655 kdb_init(); 656 if (boothowto & RB_KDB) 657 Debugger("Boot flags requested debugger"); 658 #endif 659 identcpu(); 660 #if 0 661 initializecpu(); /* Initialize CPU registers */ 662 #endif 663 init_param2((phys_avail[1] - phys_avail[0]) / PAGE_SIZE); 664 665 #if 0 666 /* 667 * Map the message buffer 668 */ 669 for (off = 0; off < round_page(MSGBUF_SIZE); off += PAGE_SIZE) 670 pmap_kenter((vm_offset_t)msgbufp + off, avail_end + off); 671 msgbufinit(msgbufp, MSGBUF_SIZE); 672 #endif 673 #if 0 674 thread0.td_pcb_cr3 ... MMU 675 lwp0.lwp_md.md_regs = &proc0_tf; 676 #endif 677 } 678 679 /* 680 * Filesystem image paths for the virtual kernel are optional. 681 * If specified they each should point to a disk image, 682 * the first of which will become the root disk. 683 * 684 * The virtual kernel caches data from our 'disk' just like a normal kernel, 685 * so we do not really want the real kernel to cache the data too. Use 686 * O_DIRECT to remove the duplication. 687 */ 688 static 689 void 690 init_disk(char *diskExp[], int diskFileNum, enum vkdisk_type type) 691 { 692 int i; 693 694 if (diskFileNum == 0) 695 return; 696 697 for(i=0; i < diskFileNum; i++){ 698 char *fname; 699 fname = diskExp[i]; 700 701 if (fname == NULL) { 702 warnx("Invalid argument to '-r'"); 703 continue; 704 } 705 706 if (DiskNum < VKDISK_MAX) { 707 struct stat st; 708 struct vkdisk_info* info = NULL; 709 int fd; 710 size_t l = 0; 711 712 if (type == VKD_DISK) 713 fd = open(fname, O_RDWR|O_DIRECT, 0644); 714 else 715 fd = open(fname, O_RDONLY|O_DIRECT, 0644); 716 if (fd < 0 || fstat(fd, &st) < 0) { 717 err(1, "Unable to open/create %s", fname); 718 /* NOT REACHED */ 719 } 720 if (S_ISREG(st.st_mode)) { 721 if (flock(fd, LOCK_EX|LOCK_NB) < 0) { 722 errx(1, "Disk image %s is already " 723 "in use\n", fname); 724 /* NOT REACHED */ 725 } 726 } 727 728 info = &DiskInfo[DiskNum]; 729 l = strlen(fname); 730 731 info->unit = i; 732 info->fd = fd; 733 info->type = type; 734 memcpy(info->fname, fname, l); 735 736 if (DiskNum == 0) { 737 if (type == VKD_CD) { 738 rootdevnames[0] = "cd9660:vcd0a"; 739 } else if (type == VKD_DISK) { 740 rootdevnames[0] = "ufs:vkd0s0a"; 741 rootdevnames[1] = "ufs:vkd0s1a"; 742 } 743 } 744 745 DiskNum++; 746 } else { 747 warnx("vkd%d (%s) > VKDISK_MAX", DiskNum, fname); 748 continue; 749 } 750 } 751 } 752 753 static 754 int 755 netif_set_tapflags(int tap_unit, int f, int s) 756 { 757 struct ifreq ifr; 758 int flags; 759 760 bzero(&ifr, sizeof(ifr)); 761 762 snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit); 763 if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0) { 764 warn("tap%d: ioctl(SIOCGIFFLAGS) failed", tap_unit); 765 return -1; 766 } 767 768 /* 769 * Adjust if_flags 770 * 771 * If the flags are already set/cleared, then we return 772 * immediately to avoid extra syscalls 773 */ 774 flags = (ifr.ifr_flags & 0xffff) | (ifr.ifr_flagshigh << 16); 775 if (f < 0) { 776 /* Turn off flags */ 777 f = -f; 778 if ((flags & f) == 0) 779 return 0; 780 flags &= ~f; 781 } else { 782 /* Turn on flags */ 783 if (flags & f) 784 return 0; 785 flags |= f; 786 } 787 788 /* 789 * Fix up ifreq.ifr_name, since it may be trashed 790 * in previous ioctl(SIOCGIFFLAGS) 791 */ 792 snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit); 793 794 ifr.ifr_flags = flags & 0xffff; 795 ifr.ifr_flagshigh = flags >> 16; 796 if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0) { 797 warn("tap%d: ioctl(SIOCSIFFLAGS) failed", tap_unit); 798 return -1; 799 } 800 return 0; 801 } 802 803 static 804 int 805 netif_set_tapaddr(int tap_unit, in_addr_t addr, in_addr_t mask, int s) 806 { 807 struct ifaliasreq ifra; 808 struct sockaddr_in *in; 809 810 bzero(&ifra, sizeof(ifra)); 811 snprintf(ifra.ifra_name, sizeof(ifra.ifra_name), "tap%d", tap_unit); 812 813 /* Setup address */ 814 in = (struct sockaddr_in *)&ifra.ifra_addr; 815 in->sin_family = AF_INET; 816 in->sin_len = sizeof(*in); 817 in->sin_addr.s_addr = addr; 818 819 if (mask != 0) { 820 /* Setup netmask */ 821 in = (struct sockaddr_in *)&ifra.ifra_mask; 822 in->sin_len = sizeof(*in); 823 in->sin_addr.s_addr = mask; 824 } 825 826 if (ioctl(s, SIOCAIFADDR, &ifra) < 0) { 827 warn("tap%d: ioctl(SIOCAIFADDR) failed", tap_unit); 828 return -1; 829 } 830 return 0; 831 } 832 833 static 834 int 835 netif_add_tap2brg(int tap_unit, const char *ifbridge, int s) 836 { 837 struct ifbreq ifbr; 838 struct ifdrv ifd; 839 840 bzero(&ifbr, sizeof(ifbr)); 841 snprintf(ifbr.ifbr_ifsname, sizeof(ifbr.ifbr_ifsname), 842 "tap%d", tap_unit); 843 844 bzero(&ifd, sizeof(ifd)); 845 strlcpy(ifd.ifd_name, ifbridge, sizeof(ifd.ifd_name)); 846 ifd.ifd_cmd = BRDGADD; 847 ifd.ifd_len = sizeof(ifbr); 848 ifd.ifd_data = &ifbr; 849 850 if (ioctl(s, SIOCSDRVSPEC, &ifd) < 0) { 851 /* 852 * 'errno == EEXIST' means that the tap(4) is already 853 * a member of the bridge(4) 854 */ 855 if (errno != EEXIST) { 856 warn("ioctl(%s, SIOCSDRVSPEC) failed", ifbridge); 857 return -1; 858 } 859 } 860 return 0; 861 } 862 863 #define TAPDEV_OFLAGS (O_RDWR | O_NONBLOCK) 864 865 /* 866 * Locate the first unused tap(4) device file if auto mode is requested, 867 * or open the user supplied device file, and bring up the corresponding 868 * tap(4) interface. 869 * 870 * NOTE: Only tap(4) device file is supported currently 871 */ 872 static 873 int 874 netif_open_tap(const char *netif, int *tap_unit, int s) 875 { 876 char tap_dev[MAXPATHLEN]; 877 int tap_fd, failed; 878 struct stat st; 879 char *dname; 880 881 *tap_unit = -1; 882 883 if (strcmp(netif, "auto") == 0) { 884 /* 885 * Find first unused tap(4) device file 886 */ 887 tap_fd = open("/dev/tap", TAPDEV_OFLAGS); 888 if (tap_fd < 0) { 889 warnc(errno, "Unable to find a free tap(4)"); 890 return -1; 891 } 892 } else { 893 /* 894 * User supplied tap(4) device file or unix socket. 895 */ 896 if (netif[0] == '/') /* Absolute path */ 897 strlcpy(tap_dev, netif, sizeof(tap_dev)); 898 else 899 snprintf(tap_dev, sizeof(tap_dev), "/dev/%s", netif); 900 901 tap_fd = open(tap_dev, TAPDEV_OFLAGS); 902 903 /* 904 * If we cannot open normally try to connect to it. 905 */ 906 if (tap_fd < 0) 907 tap_fd = unix_connect(tap_dev); 908 909 if (tap_fd < 0) { 910 warn("Unable to open %s", tap_dev); 911 return -1; 912 } 913 } 914 915 /* 916 * Check whether the device file is a tap(4) 917 */ 918 if (fstat(tap_fd, &st) < 0) { 919 failed = 1; 920 } else if (S_ISCHR(st.st_mode)) { 921 dname = fdevname(tap_fd); 922 if (dname) 923 dname = strstr(dname, "tap"); 924 if (dname) { 925 /* 926 * Bring up the corresponding tap(4) interface 927 */ 928 *tap_unit = strtol(dname + 3, NULL, 10); 929 printf("TAP UNIT %d\n", *tap_unit); 930 if (netif_set_tapflags(*tap_unit, IFF_UP, s) == 0) 931 failed = 0; 932 else 933 failed = 1; 934 } else { 935 failed = 1; 936 } 937 } else if (S_ISSOCK(st.st_mode)) { 938 /* 939 * Special socket connection (typically to vknet). We 940 * do not have to do anything. 941 */ 942 failed = 0; 943 } else { 944 failed = 1; 945 } 946 947 if (failed) { 948 warnx("%s is not a tap(4) device or socket", tap_dev); 949 close(tap_fd); 950 tap_fd = -1; 951 *tap_unit = -1; 952 } 953 return tap_fd; 954 } 955 956 static int 957 unix_connect(const char *path) 958 { 959 struct sockaddr_un sunx; 960 int len; 961 int net_fd; 962 int sndbuf = 262144; 963 struct stat st; 964 965 snprintf(sunx.sun_path, sizeof(sunx.sun_path), "%s", path); 966 len = offsetof(struct sockaddr_un, sun_path[strlen(sunx.sun_path)]); 967 ++len; /* include nul */ 968 sunx.sun_family = AF_UNIX; 969 sunx.sun_len = len; 970 971 net_fd = socket(AF_UNIX, SOCK_SEQPACKET, 0); 972 if (net_fd < 0) 973 return(-1); 974 if (connect(net_fd, (void *)&sunx, len) < 0) { 975 close(net_fd); 976 return(-1); 977 } 978 setsockopt(net_fd, SOL_SOCKET, SO_SNDBUF, &sndbuf, sizeof(sndbuf)); 979 if (fstat(net_fd, &st) == 0) 980 printf("Network socket buffer: %d bytes\n", st.st_blksize); 981 fcntl(net_fd, F_SETFL, O_NONBLOCK); 982 return(net_fd); 983 } 984 985 #undef TAPDEV_MAJOR 986 #undef TAPDEV_MINOR 987 #undef TAPDEV_OFLAGS 988 989 /* 990 * Following syntax is supported, 991 * 1) x.x.x.x tap(4)'s address is x.x.x.x 992 * 993 * 2) x.x.x.x/z tap(4)'s address is x.x.x.x 994 * tap(4)'s netmask len is z 995 * 996 * 3) x.x.x.x:y.y.y.y tap(4)'s address is x.x.x.x 997 * pseudo netif's address is y.y.y.y 998 * 999 * 4) x.x.x.x:y.y.y.y/z tap(4)'s address is x.x.x.x 1000 * pseudo netif's address is y.y.y.y 1001 * tap(4) and pseudo netif's netmask len are z 1002 * 1003 * 5) bridgeX tap(4) will be added to bridgeX 1004 * 1005 * 6) bridgeX:y.y.y.y tap(4) will be added to bridgeX 1006 * pseudo netif's address is y.y.y.y 1007 * 1008 * 7) bridgeX:y.y.y.y/z tap(4) will be added to bridgeX 1009 * pseudo netif's address is y.y.y.y 1010 * pseudo netif's netmask len is z 1011 */ 1012 static 1013 int 1014 netif_init_tap(int tap_unit, in_addr_t *addr, in_addr_t *mask, int s) 1015 { 1016 in_addr_t tap_addr, netmask, netif_addr; 1017 int next_netif_addr; 1018 char *tok, *masklen_str, *ifbridge; 1019 1020 *addr = 0; 1021 *mask = 0; 1022 1023 tok = strtok(NULL, ":/"); 1024 if (tok == NULL) { 1025 /* 1026 * Nothing special, simply use tap(4) as backend 1027 */ 1028 return 0; 1029 } 1030 1031 if (inet_pton(AF_INET, tok, &tap_addr) > 0) { 1032 /* 1033 * tap(4)'s address is supplied 1034 */ 1035 ifbridge = NULL; 1036 1037 /* 1038 * If there is next token, then it may be pseudo 1039 * netif's address or netmask len for tap(4) 1040 */ 1041 next_netif_addr = 0; 1042 } else { 1043 /* 1044 * Not tap(4)'s address, assume it as a bridge(4) 1045 * iface name 1046 */ 1047 tap_addr = 0; 1048 ifbridge = tok; 1049 1050 /* 1051 * If there is next token, then it must be pseudo 1052 * netif's address 1053 */ 1054 next_netif_addr = 1; 1055 } 1056 1057 netmask = netif_addr = 0; 1058 1059 tok = strtok(NULL, ":/"); 1060 if (tok == NULL) 1061 goto back; 1062 1063 if (inet_pton(AF_INET, tok, &netif_addr) <= 0) { 1064 if (next_netif_addr) { 1065 warnx("Invalid pseudo netif address: %s", tok); 1066 return -1; 1067 } 1068 netif_addr = 0; 1069 1070 /* 1071 * Current token is not address, then it must be netmask len 1072 */ 1073 masklen_str = tok; 1074 } else { 1075 /* 1076 * Current token is pseudo netif address, if there is next token 1077 * it must be netmask len 1078 */ 1079 masklen_str = strtok(NULL, "/"); 1080 } 1081 1082 /* Calculate netmask */ 1083 if (masklen_str != NULL) { 1084 u_long masklen; 1085 1086 masklen = strtoul(masklen_str, NULL, 10); 1087 if (masklen < 32 && masklen > 0) { 1088 netmask = htonl(~((1LL << (32 - masklen)) - 1) 1089 & 0xffffffff); 1090 } else { 1091 warnx("Invalid netmask len: %lu", masklen); 1092 return -1; 1093 } 1094 } 1095 1096 /* Make sure there is no more token left */ 1097 if (strtok(NULL, ":/") != NULL) { 1098 warnx("Invalid argument to '-I'"); 1099 return -1; 1100 } 1101 1102 back: 1103 if (tap_unit < 0) { 1104 /* Do nothing */ 1105 } else if (ifbridge == NULL) { 1106 /* Set tap(4) address/netmask */ 1107 if (netif_set_tapaddr(tap_unit, tap_addr, netmask, s) < 0) 1108 return -1; 1109 } else { 1110 /* Tie tap(4) to bridge(4) */ 1111 if (netif_add_tap2brg(tap_unit, ifbridge, s) < 0) 1112 return -1; 1113 } 1114 1115 *addr = netif_addr; 1116 *mask = netmask; 1117 return 0; 1118 } 1119 1120 /* 1121 * NetifInfo[] will be filled for pseudo netif initialization. 1122 * NetifNum will be bumped to reflect the number of valid entries 1123 * in NetifInfo[]. 1124 */ 1125 static 1126 void 1127 init_netif(char *netifExp[], int netifExpNum) 1128 { 1129 int i, s; 1130 1131 if (netifExpNum == 0) 1132 return; 1133 1134 s = socket(AF_INET, SOCK_DGRAM, 0); /* for ioctl(SIOC) */ 1135 if (s < 0) 1136 return; 1137 1138 for (i = 0; i < netifExpNum; ++i) { 1139 struct vknetif_info *info; 1140 in_addr_t netif_addr, netif_mask; 1141 int tap_fd, tap_unit; 1142 char *netif; 1143 1144 netif = strtok(netifExp[i], ":"); 1145 if (netif == NULL) { 1146 warnx("Invalid argument to '-I'"); 1147 continue; 1148 } 1149 1150 /* 1151 * Open tap(4) device file and bring up the 1152 * corresponding interface 1153 */ 1154 tap_fd = netif_open_tap(netif, &tap_unit, s); 1155 if (tap_fd < 0) 1156 continue; 1157 1158 /* 1159 * Initialize tap(4) and get address/netmask 1160 * for pseudo netif 1161 * 1162 * NB: Rest part of netifExp[i] is passed 1163 * to netif_init_tap() implicitly. 1164 */ 1165 if (netif_init_tap(tap_unit, &netif_addr, &netif_mask, s) < 0) { 1166 /* 1167 * NB: Closing tap(4) device file will bring 1168 * down the corresponding interface 1169 */ 1170 close(tap_fd); 1171 continue; 1172 } 1173 1174 info = &NetifInfo[NetifNum]; 1175 info->tap_fd = tap_fd; 1176 info->tap_unit = tap_unit; 1177 info->netif_addr = netif_addr; 1178 info->netif_mask = netif_mask; 1179 1180 NetifNum++; 1181 if (NetifNum >= VKNETIF_MAX) /* XXX will this happen? */ 1182 break; 1183 } 1184 close(s); 1185 } 1186 1187 static 1188 void 1189 writepid( void ) 1190 { 1191 pid_t self; 1192 FILE *fp; 1193 1194 if (pid_file != NULL) { 1195 self = getpid(); 1196 fp = fopen(pid_file, "w"); 1197 1198 if (fp != NULL) { 1199 fprintf(fp, "%ld\n", (long)self); 1200 fclose(fp); 1201 } 1202 else { 1203 perror("Warning: couldn't open pidfile"); 1204 } 1205 } 1206 } 1207 1208 static 1209 void 1210 cleanpid( void ) 1211 { 1212 if (pid_file != NULL) { 1213 if ( unlink(pid_file) != 0 ) 1214 perror("Warning: couldn't remove pidfile"); 1215 } 1216 } 1217 1218 static 1219 void 1220 usage_err(const char *ctl, ...) 1221 { 1222 va_list va; 1223 1224 va_start(va, ctl); 1225 vfprintf(stderr, ctl, va); 1226 va_end(va); 1227 fprintf(stderr, "\n"); 1228 exit(EX_USAGE); 1229 } 1230 1231 static 1232 void 1233 usage_help(_Bool help) 1234 { 1235 fprintf(stderr, "Usage: %s [-hsUv] [-c file] [-e name=value:name=value:...]\n" 1236 "\t[-i file] [-I interface[:address1[:address2][/netmask]]] [-l cpulock]\n" 1237 "\t[-m size] [-n numcpus] [-p file] [-r file]\n", save_av[0]); 1238 1239 if (help) 1240 fprintf(stderr, "\nArguments:\n" 1241 "\t-c\tSpecify a readonly CD-ROM image file to be used by the kernel.\n" 1242 "\t-e\tSpecify an environment to be used by the kernel.\n" 1243 "\t-h\tThis list of options.\n" 1244 "\t-i\tSpecify a memory image file to be used by the virtual kernel.\n" 1245 "\t-I\tCreate a virtual network device.\n" 1246 "\t-l\tSpecify which, if any, real CPUs to lock virtual CPUs to.\n" 1247 "\t-m\tSpecify the amount of memory to be used by the kernel in bytes.\n" 1248 "\t-n\tSpecify the number of CPUs you wish to emulate.\n" 1249 "\t-p\tSpecify a file in which to store the process ID.\n" 1250 "\t-r\tSpecify a R/W disk image file to be used by the kernel.\n" 1251 "\t-s\tBoot into single-user mode.\n" 1252 "\t-U\tEnable writing to kernel memory and module loading.\n" 1253 "\t-v\tTurn on verbose booting.\n"); 1254 1255 exit(EX_USAGE); 1256 } 1257 1258 void 1259 cpu_reset(void) 1260 { 1261 kprintf("cpu reset, rebooting vkernel\n"); 1262 closefrom(3); 1263 cleanpid(); 1264 execv(save_av[0], save_av); 1265 } 1266 1267 void 1268 cpu_halt(void) 1269 { 1270 kprintf("cpu halt, exiting vkernel\n"); 1271 cleanpid(); 1272 exit(EX_OK); 1273 } 1274 1275 void 1276 setrealcpu(void) 1277 { 1278 switch(lwp_cpu_lock) { 1279 case LCL_PER_CPU: 1280 if (bootverbose) 1281 kprintf("Locking CPU%d to real cpu %d\n", 1282 mycpuid, next_cpu); 1283 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu)); 1284 next_cpu++; 1285 if (next_cpu >= real_ncpus) 1286 next_cpu = 0; 1287 break; 1288 case LCL_SINGLE_CPU: 1289 if (bootverbose) 1290 kprintf("Locking CPU%d to real cpu %d\n", 1291 mycpuid, next_cpu); 1292 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu)); 1293 break; 1294 default: 1295 /* do not map virtual cpus to real cpus */ 1296 break; 1297 } 1298 } 1299