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 56 #include <machine/cpu.h> 57 #include <machine/globaldata.h> 58 #include <machine/tls.h> 59 #include <machine/md_var.h> 60 #include <machine/vmparam.h> 61 #include <cpu/specialreg.h> 62 63 #include <net/if.h> 64 #include <net/if_arp.h> 65 #include <net/ethernet.h> 66 #include <net/bridge/if_bridgevar.h> 67 #include <netinet/in.h> 68 #include <arpa/inet.h> 69 70 #include <stdio.h> 71 #include <stdlib.h> 72 #include <stdarg.h> 73 #include <stdbool.h> 74 #include <unistd.h> 75 #include <fcntl.h> 76 #include <string.h> 77 #include <err.h> 78 #include <errno.h> 79 #include <assert.h> 80 #include <sysexits.h> 81 82 vm_paddr_t phys_avail[16]; 83 vm_paddr_t Maxmem; 84 vm_paddr_t Maxmem_bytes; 85 long physmem; 86 int MemImageFd = -1; 87 struct vkdisk_info DiskInfo[VKDISK_MAX]; 88 int DiskNum; 89 struct vknetif_info NetifInfo[VKNETIF_MAX]; 90 int NetifNum; 91 char *pid_file; 92 vm_offset_t KvaStart; 93 vm_offset_t KvaEnd; 94 vm_offset_t KvaSize; 95 vm_offset_t virtual_start; 96 vm_offset_t virtual_end; 97 vm_offset_t virtual2_start; 98 vm_offset_t virtual2_end; 99 vm_offset_t kernel_vm_end; 100 vm_offset_t crashdumpmap; 101 vm_offset_t clean_sva; 102 vm_offset_t clean_eva; 103 struct msgbuf *msgbufp; 104 caddr_t ptvmmap; 105 vpte_t *KernelPTD; 106 vpte_t *KernelPTA; /* Warning: Offset for direct VA translation */ 107 void *dmap_min_address; 108 u_int cpu_feature; /* XXX */ 109 int tsc_present; 110 int64_t tsc_frequency; 111 int optcpus; /* number of cpus - see mp_start() */ 112 int lwp_cpu_lock; /* if/how to lock virtual CPUs to real CPUs */ 113 int real_ncpus; /* number of real CPUs */ 114 int next_cpu; /* next real CPU to lock a virtual CPU to */ 115 116 struct privatespace *CPU_prvspace; 117 118 static struct trapframe proc0_tf; 119 static void *proc0paddr; 120 121 static void init_sys_memory(char *imageFile); 122 static void init_kern_memory(void); 123 static void init_globaldata(void); 124 static void init_vkernel(void); 125 static void init_disk(char *diskExp[], int diskFileNum, enum vkdisk_type type); 126 static void init_netif(char *netifExp[], int netifFileNum); 127 static void writepid(void); 128 static void cleanpid(void); 129 static int unix_connect(const char *path); 130 static void usage_err(const char *ctl, ...); 131 static void usage_help(_Bool); 132 133 static int save_ac; 134 static char **save_av; 135 136 /* 137 * Kernel startup for virtual kernels - standard main() 138 */ 139 int 140 main(int ac, char **av) 141 { 142 char *memImageFile = NULL; 143 char *netifFile[VKNETIF_MAX]; 144 char *diskFile[VKDISK_MAX]; 145 char *cdFile[VKDISK_MAX]; 146 char *suffix; 147 char *endp; 148 char *tmp; 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 pos; 159 int eflag; 160 int real_vkernel_enable; 161 int supports_sse; 162 size_t vsize; 163 size_t kenv_size; 164 size_t kenv_size2; 165 166 save_ac = ac; 167 save_av = av; 168 eflag = 0; 169 pos = 0; 170 kenv_size = 0; 171 172 /* 173 * Process options 174 */ 175 kernel_mem_readonly = 1; 176 #ifdef SMP 177 optcpus = 2; 178 #endif 179 lwp_cpu_lock = LCL_NONE; 180 181 real_vkernel_enable = 0; 182 vsize = sizeof(real_vkernel_enable); 183 sysctlbyname("vm.vkernel_enable", &real_vkernel_enable, &vsize, NULL,0); 184 185 if (real_vkernel_enable == 0) { 186 errx(1, "vm.vkernel_enable is 0, must be set " 187 "to 1 to execute a vkernel!"); 188 } 189 190 real_ncpus = 1; 191 vsize = sizeof(real_ncpus); 192 sysctlbyname("hw.ncpu", &real_ncpus, &vsize, NULL, 0); 193 194 if (ac < 2) 195 usage_help(false); 196 197 while ((c = getopt(ac, av, "c:hsvl:m:n:r:e:i:p:I:U")) != -1) { 198 switch(c) { 199 case 'e': 200 /* 201 * name=value:name=value:name=value... 202 * name="value"... 203 * 204 * Allow values to be quoted but note that shells 205 * may remove the quotes, so using this feature 206 * to embed colons may require a backslash. 207 */ 208 n = strlen(optarg); 209 isq = 0; 210 211 if (eflag == 0) { 212 kenv_size = n + 2; 213 kern_envp = malloc(kenv_size); 214 if (kern_envp == NULL) 215 errx(1, "Couldn't allocate %zd bytes for kern_envp", kenv_size); 216 } else { 217 kenv_size2 = kenv_size + n + 1; 218 pos = kenv_size - 1; 219 if ((tmp = realloc(kern_envp, kenv_size2)) == NULL) 220 errx(1, "Couldn't reallocate %zd bytes for kern_envp", kenv_size2); 221 kern_envp = tmp; 222 kenv_size = kenv_size2; 223 } 224 225 for (i = 0, j = pos; i < n; ++i) { 226 if (optarg[i] == '"') 227 isq ^= 1; 228 else if (optarg[i] == '\'') 229 isq ^= 2; 230 else if (isq == 0 && optarg[i] == ':') 231 kern_envp[j++] = 0; 232 else 233 kern_envp[j++] = optarg[i]; 234 } 235 kern_envp[j++] = 0; 236 kern_envp[j++] = 0; 237 eflag++; 238 break; 239 case 's': 240 boothowto |= RB_SINGLE; 241 break; 242 case 'v': 243 bootverbose = 1; 244 break; 245 case 'i': 246 memImageFile = optarg; 247 break; 248 case 'I': 249 if (netifFileNum < VKNETIF_MAX) 250 netifFile[netifFileNum++] = strdup(optarg); 251 break; 252 case 'r': 253 if (bootOnDisk < 0) 254 bootOnDisk = 1; 255 if (diskFileNum + cdFileNum < VKDISK_MAX) 256 diskFile[diskFileNum++] = strdup(optarg); 257 break; 258 case 'c': 259 if (bootOnDisk < 0) 260 bootOnDisk = 0; 261 if (diskFileNum + cdFileNum < VKDISK_MAX) 262 cdFile[cdFileNum++] = strdup(optarg); 263 break; 264 case 'm': 265 Maxmem_bytes = strtoull(optarg, &suffix, 0); 266 if (suffix) { 267 switch(*suffix) { 268 case 'g': 269 case 'G': 270 Maxmem_bytes <<= 30; 271 break; 272 case 'm': 273 case 'M': 274 Maxmem_bytes <<= 20; 275 break; 276 case 'k': 277 case 'K': 278 Maxmem_bytes <<= 10; 279 break; 280 default: 281 Maxmem_bytes = 0; 282 usage_err("Bad maxmem option"); 283 /* NOT REACHED */ 284 break; 285 } 286 } 287 break; 288 case 'l': 289 next_cpu = -1; 290 if (strncmp("map", optarg, 3) == 0) { 291 lwp_cpu_lock = LCL_PER_CPU; 292 if (optarg[3] == ',') { 293 next_cpu = strtol(optarg+4, &endp, 0); 294 if (*endp != '\0') 295 usage_err("Bad target CPU number at '%s'", endp); 296 } else { 297 next_cpu = 0; 298 } 299 if (next_cpu < 0 || next_cpu > real_ncpus - 1) 300 usage_err("Bad target CPU, valid range is 0-%d", real_ncpus - 1); 301 } else if (strncmp("any", optarg, 3) == 0) { 302 lwp_cpu_lock = LCL_NONE; 303 } else { 304 lwp_cpu_lock = LCL_SINGLE_CPU; 305 next_cpu = strtol(optarg, &endp, 0); 306 if (*endp != '\0') 307 usage_err("Bad target CPU number at '%s'", endp); 308 if (next_cpu < 0 || next_cpu > real_ncpus - 1) 309 usage_err("Bad target CPU, valid range is 0-%d", real_ncpus - 1); 310 } 311 break; 312 case 'n': 313 /* 314 * This value is set up by mp_start(), don't just 315 * set ncpus here. 316 */ 317 #ifdef SMP 318 optcpus = strtol(optarg, NULL, 0); 319 if (optcpus < 1 || optcpus > MAXCPU) 320 usage_err("Bad ncpus, valid range is 1-%d", MAXCPU); 321 #else 322 if (strtol(optarg, NULL, 0) != 1) { 323 usage_err("You built a UP vkernel, only 1 cpu!"); 324 } 325 #endif 326 327 break; 328 case 'p': 329 pid_file = optarg; 330 break; 331 case 'U': 332 kernel_mem_readonly = 0; 333 break; 334 case 'h': 335 usage_help(true); 336 break; 337 default: 338 usage_help(false); 339 } 340 } 341 342 writepid(); 343 cpu_disable_intr(); 344 init_sys_memory(memImageFile); 345 init_kern_memory(); 346 init_globaldata(); 347 init_vkernel(); 348 setrealcpu(); 349 init_kqueue(); 350 351 /* 352 * Check TSC 353 */ 354 vsize = sizeof(tsc_present); 355 sysctlbyname("hw.tsc_present", &tsc_present, &vsize, NULL, 0); 356 vsize = sizeof(tsc_frequency); 357 sysctlbyname("hw.tsc_frequency", &tsc_frequency, &vsize, NULL, 0); 358 if (tsc_present) 359 cpu_feature |= CPUID_TSC; 360 361 /* 362 * Check SSE 363 */ 364 vsize = sizeof(supports_sse); 365 supports_sse = 0; 366 sysctlbyname("hw.instruction_sse", &supports_sse, &vsize, NULL, 0); 367 init_fpu(supports_sse); 368 if (supports_sse) 369 cpu_feature |= CPUID_SSE | CPUID_FXSR; 370 371 /* 372 * We boot from the first installed disk. 373 */ 374 if (bootOnDisk == 1) { 375 init_disk(diskFile, diskFileNum, VKD_DISK); 376 init_disk(cdFile, cdFileNum, VKD_CD); 377 } else { 378 init_disk(cdFile, cdFileNum, VKD_CD); 379 init_disk(diskFile, diskFileNum, VKD_DISK); 380 } 381 init_netif(netifFile, netifFileNum); 382 init_exceptions(); 383 mi_startup(); 384 /* NOT REACHED */ 385 exit(EX_SOFTWARE); 386 } 387 388 /* 389 * Initialize system memory. This is the virtual kernel's 'RAM'. 390 */ 391 static 392 void 393 init_sys_memory(char *imageFile) 394 { 395 struct stat st; 396 int i; 397 int fd; 398 399 /* 400 * Figure out the system memory image size. If an image file was 401 * specified and -m was not specified, use the image file's size. 402 */ 403 if (imageFile && stat(imageFile, &st) == 0 && Maxmem_bytes == 0) 404 Maxmem_bytes = (vm_paddr_t)st.st_size; 405 if ((imageFile == NULL || stat(imageFile, &st) < 0) && 406 Maxmem_bytes == 0) { 407 errx(1, "Cannot create new memory file %s unless " 408 "system memory size is specified with -m", 409 imageFile); 410 /* NOT REACHED */ 411 } 412 413 /* 414 * Maxmem must be known at this time 415 */ 416 if (Maxmem_bytes < 64 * 1024 * 1024 || (Maxmem_bytes & SEG_MASK)) { 417 errx(1, "Bad maxmem specification: 64MB minimum, " 418 "multiples of %dMB only", 419 SEG_SIZE / 1024 / 1024); 420 /* NOT REACHED */ 421 } 422 423 /* 424 * Generate an image file name if necessary, then open/create the 425 * file exclusively locked. Do not allow multiple virtual kernels 426 * to use the same image file. 427 * 428 * Don't iterate through a million files if we do not have write 429 * access to the directory, stop if our open() failed on a 430 * non-existant file. Otherwise opens can fail for any number 431 */ 432 if (imageFile == NULL) { 433 for (i = 0; i < 1000000; ++i) { 434 asprintf(&imageFile, "/var/vkernel/memimg.%06d", i); 435 fd = open(imageFile, 436 O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644); 437 if (fd < 0 && stat(imageFile, &st) == 0) { 438 free(imageFile); 439 continue; 440 } 441 break; 442 } 443 } else { 444 fd = open(imageFile, O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644); 445 } 446 fprintf(stderr, "Using memory file: %s\n", imageFile); 447 if (fd < 0 || fstat(fd, &st) < 0) { 448 err(1, "Unable to open/create %s", imageFile); 449 /* NOT REACHED */ 450 } 451 452 /* 453 * Truncate or extend the file as necessary. Clean out the contents 454 * of the file, we want it to be full of holes so we don't waste 455 * time reading in data from an old file that we no longer care 456 * about. 457 */ 458 ftruncate(fd, 0); 459 ftruncate(fd, Maxmem_bytes); 460 461 MemImageFd = fd; 462 Maxmem = Maxmem_bytes >> PAGE_SHIFT; 463 physmem = Maxmem; 464 } 465 466 /* 467 * Initialize kernel memory. This reserves kernel virtual memory by using 468 * MAP_VPAGETABLE 469 */ 470 471 static 472 void 473 init_kern_memory(void) 474 { 475 void *base; 476 void *try; 477 char dummy; 478 char *topofstack = &dummy; 479 int i; 480 void *firstfree; 481 482 /* 483 * Memory map our kernel virtual memory space. Note that the 484 * kernel image itself is not made part of this memory for the 485 * moment. 486 * 487 * The memory map must be segment-aligned so we can properly 488 * offset KernelPTD. 489 * 490 * If the system kernel has a different MAXDSIZ, it might not 491 * be possible to map kernel memory in its prefered location. 492 * Try a number of different locations. 493 */ 494 try = (void *)(512UL << 30); 495 base = NULL; 496 while ((char *)try + KERNEL_KVA_SIZE < topofstack) { 497 base = mmap(try, KERNEL_KVA_SIZE, PROT_READ|PROT_WRITE, 498 MAP_FILE|MAP_SHARED|MAP_VPAGETABLE, 499 MemImageFd, (off_t)try); 500 if (base == try) 501 break; 502 if (base != MAP_FAILED) 503 munmap(base, KERNEL_KVA_SIZE); 504 try = (char *)try + (512UL << 30); 505 } 506 if (base != try) { 507 err(1, "Unable to mmap() kernel virtual memory!"); 508 /* NOT REACHED */ 509 } 510 madvise(base, KERNEL_KVA_SIZE, MADV_NOSYNC); 511 KvaStart = (vm_offset_t)base; 512 KvaSize = KERNEL_KVA_SIZE; 513 KvaEnd = KvaStart + KvaSize; 514 515 /* cannot use kprintf yet */ 516 printf("KVM mapped at %p-%p\n", (void *)KvaStart, (void *)KvaEnd); 517 518 /* MAP_FILE? */ 519 dmap_min_address = mmap(0, DMAP_SIZE, PROT_READ|PROT_WRITE, 520 MAP_NOCORE|MAP_NOSYNC|MAP_SHARED, 521 MemImageFd, 0); 522 if (dmap_min_address == MAP_FAILED) { 523 err(1, "Unable to mmap() kernel DMAP region!"); 524 /* NOT REACHED */ 525 } 526 527 firstfree = NULL; 528 pmap_bootstrap((vm_paddr_t *)&firstfree, (int64_t)base); 529 530 mcontrol(base, KERNEL_KVA_SIZE, MADV_SETMAP, 531 0 | VPTE_R | VPTE_W | VPTE_V); 532 533 /* 534 * phys_avail[] represents unallocated physical memory. MI code 535 * will use phys_avail[] to create the vm_page array. 536 */ 537 phys_avail[0] = (vm_paddr_t)firstfree; 538 phys_avail[0] = (phys_avail[0] + PAGE_MASK) & ~(vm_paddr_t)PAGE_MASK; 539 phys_avail[1] = Maxmem_bytes; 540 541 #if JGV 542 /* 543 * (virtual_start, virtual_end) represent unallocated kernel virtual 544 * memory. MI code will create kernel_map using these parameters. 545 */ 546 virtual_start = KvaStart + (long)firstfree; 547 virtual_start = (virtual_start + PAGE_MASK) & ~(vm_offset_t)PAGE_MASK; 548 virtual_end = KvaStart + KERNEL_KVA_SIZE; 549 #endif 550 551 /* 552 * pmap_growkernel() will set the correct value. 553 */ 554 kernel_vm_end = 0; 555 556 /* 557 * Allocate space for process 0's UAREA. 558 */ 559 proc0paddr = (void *)virtual_start; 560 for (i = 0; i < UPAGES; ++i) { 561 pmap_kenter_quick(virtual_start, phys_avail[0]); 562 virtual_start += PAGE_SIZE; 563 phys_avail[0] += PAGE_SIZE; 564 } 565 566 /* 567 * crashdumpmap 568 */ 569 crashdumpmap = virtual_start; 570 virtual_start += MAXDUMPPGS * PAGE_SIZE; 571 572 /* 573 * msgbufp maps the system message buffer 574 */ 575 assert((MSGBUF_SIZE & PAGE_MASK) == 0); 576 msgbufp = (void *)virtual_start; 577 for (i = 0; i < (MSGBUF_SIZE >> PAGE_SHIFT); ++i) { 578 pmap_kenter_quick(virtual_start, phys_avail[0]); 579 virtual_start += PAGE_SIZE; 580 phys_avail[0] += PAGE_SIZE; 581 } 582 msgbufinit(msgbufp, MSGBUF_SIZE); 583 584 /* 585 * used by kern_memio for /dev/mem access 586 */ 587 ptvmmap = (caddr_t)virtual_start; 588 virtual_start += PAGE_SIZE; 589 590 /* 591 * Bootstrap the kernel_pmap 592 */ 593 #if JGV 594 pmap_bootstrap(); 595 #endif 596 } 597 598 /* 599 * Map the per-cpu globaldata for cpu #0. Allocate the space using 600 * virtual_start and phys_avail[0] 601 */ 602 static 603 void 604 init_globaldata(void) 605 { 606 int i; 607 vm_paddr_t pa; 608 vm_offset_t va; 609 610 /* 611 * Reserve enough KVA to cover possible cpus. This is a considerable 612 * amount of KVA since the privatespace structure includes two 613 * whole page table mappings. 614 */ 615 virtual_start = (virtual_start + SEG_MASK) & ~(vm_offset_t)SEG_MASK; 616 CPU_prvspace = (void *)virtual_start; 617 virtual_start += sizeof(struct privatespace) * SMP_MAXCPU; 618 619 /* 620 * Allocate enough physical memory to cover the mdglobaldata 621 * portion of the space and the idle stack and map the pages 622 * into KVA. For cpu #0 only. 623 */ 624 for (i = 0; i < sizeof(struct mdglobaldata); i += PAGE_SIZE) { 625 pa = phys_avail[0]; 626 va = (vm_offset_t)&CPU_prvspace[0].mdglobaldata + i; 627 pmap_kenter_quick(va, pa); 628 phys_avail[0] += PAGE_SIZE; 629 } 630 for (i = 0; i < sizeof(CPU_prvspace[0].idlestack); i += PAGE_SIZE) { 631 pa = phys_avail[0]; 632 va = (vm_offset_t)&CPU_prvspace[0].idlestack + i; 633 pmap_kenter_quick(va, pa); 634 phys_avail[0] += PAGE_SIZE; 635 } 636 637 /* 638 * Setup the %gs for cpu #0. The mycpu macro works after this 639 * point. Note that %fs is used by pthreads. 640 */ 641 tls_set_gs(&CPU_prvspace[0], sizeof(struct privatespace)); 642 } 643 644 /* 645 * Initialize very low level systems including thread0, proc0, etc. 646 */ 647 static 648 void 649 init_vkernel(void) 650 { 651 struct mdglobaldata *gd; 652 653 gd = &CPU_prvspace[0].mdglobaldata; 654 bzero(gd, sizeof(*gd)); 655 656 gd->mi.gd_curthread = &thread0; 657 thread0.td_gd = &gd->mi; 658 ncpus = 1; 659 ncpus2 = 1; /* rounded down power of 2 */ 660 ncpus_fit = 1; /* rounded up power of 2 */ 661 /* ncpus2_mask and ncpus_fit_mask are 0 */ 662 init_param1(); 663 gd->mi.gd_prvspace = &CPU_prvspace[0]; 664 mi_gdinit(&gd->mi, 0); 665 cpu_gdinit(gd, 0); 666 mi_proc0init(&gd->mi, proc0paddr); 667 lwp0.lwp_md.md_regs = &proc0_tf; 668 669 /*init_locks();*/ 670 #ifdef SMP 671 /* 672 * Get the initial mplock with a count of 1 for the BSP. 673 * This uses a LOGICAL cpu ID, ie BSP == 0. 674 */ 675 cpu_get_initial_mplock(); 676 #endif 677 cninit(); 678 rand_initialize(); 679 #if 0 /* #ifdef DDB */ 680 kdb_init(); 681 if (boothowto & RB_KDB) 682 Debugger("Boot flags requested debugger"); 683 #endif 684 identcpu(); 685 #if 0 686 initializecpu(); /* Initialize CPU registers */ 687 #endif 688 init_param2((phys_avail[1] - phys_avail[0]) / PAGE_SIZE); 689 690 #if 0 691 /* 692 * Map the message buffer 693 */ 694 for (off = 0; off < round_page(MSGBUF_SIZE); off += PAGE_SIZE) 695 pmap_kenter((vm_offset_t)msgbufp + off, avail_end + off); 696 msgbufinit(msgbufp, MSGBUF_SIZE); 697 #endif 698 #if 0 699 thread0.td_pcb_cr3 ... MMU 700 lwp0.lwp_md.md_regs = &proc0_tf; 701 #endif 702 } 703 704 /* 705 * Filesystem image paths for the virtual kernel are optional. 706 * If specified they each should point to a disk image, 707 * the first of which will become the root disk. 708 * 709 * The virtual kernel caches data from our 'disk' just like a normal kernel, 710 * so we do not really want the real kernel to cache the data too. Use 711 * O_DIRECT to remove the duplication. 712 */ 713 static 714 void 715 init_disk(char *diskExp[], int diskFileNum, enum vkdisk_type type) 716 { 717 int i; 718 719 if (diskFileNum == 0) 720 return; 721 722 for(i=0; i < diskFileNum; i++){ 723 char *fname; 724 fname = diskExp[i]; 725 726 if (fname == NULL) { 727 warnx("Invalid argument to '-r'"); 728 continue; 729 } 730 731 if (DiskNum < VKDISK_MAX) { 732 struct stat st; 733 struct vkdisk_info* info = NULL; 734 int fd; 735 size_t l = 0; 736 737 if (type == VKD_DISK) 738 fd = open(fname, O_RDWR|O_DIRECT, 0644); 739 else 740 fd = open(fname, O_RDONLY|O_DIRECT, 0644); 741 if (fd < 0 || fstat(fd, &st) < 0) { 742 err(1, "Unable to open/create %s", fname); 743 /* NOT REACHED */ 744 } 745 if (S_ISREG(st.st_mode)) { 746 if (flock(fd, LOCK_EX|LOCK_NB) < 0) { 747 errx(1, "Disk image %s is already " 748 "in use\n", fname); 749 /* NOT REACHED */ 750 } 751 } 752 753 info = &DiskInfo[DiskNum]; 754 l = strlen(fname); 755 756 info->unit = i; 757 info->fd = fd; 758 info->type = type; 759 memcpy(info->fname, fname, l); 760 761 if (DiskNum == 0) { 762 if (type == VKD_CD) { 763 rootdevnames[0] = "cd9660:vcd0a"; 764 } else if (type == VKD_DISK) { 765 rootdevnames[0] = "ufs:vkd0s0a"; 766 rootdevnames[1] = "ufs:vkd0s1a"; 767 } 768 } 769 770 DiskNum++; 771 } else { 772 warnx("vkd%d (%s) > VKDISK_MAX", DiskNum, fname); 773 continue; 774 } 775 } 776 } 777 778 static 779 int 780 netif_set_tapflags(int tap_unit, int f, int s) 781 { 782 struct ifreq ifr; 783 int flags; 784 785 bzero(&ifr, sizeof(ifr)); 786 787 snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit); 788 if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0) { 789 warn("tap%d: ioctl(SIOCGIFFLAGS) failed", tap_unit); 790 return -1; 791 } 792 793 /* 794 * Adjust if_flags 795 * 796 * If the flags are already set/cleared, then we return 797 * immediately to avoid extra syscalls 798 */ 799 flags = (ifr.ifr_flags & 0xffff) | (ifr.ifr_flagshigh << 16); 800 if (f < 0) { 801 /* Turn off flags */ 802 f = -f; 803 if ((flags & f) == 0) 804 return 0; 805 flags &= ~f; 806 } else { 807 /* Turn on flags */ 808 if (flags & f) 809 return 0; 810 flags |= f; 811 } 812 813 /* 814 * Fix up ifreq.ifr_name, since it may be trashed 815 * in previous ioctl(SIOCGIFFLAGS) 816 */ 817 snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit); 818 819 ifr.ifr_flags = flags & 0xffff; 820 ifr.ifr_flagshigh = flags >> 16; 821 if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0) { 822 warn("tap%d: ioctl(SIOCSIFFLAGS) failed", tap_unit); 823 return -1; 824 } 825 return 0; 826 } 827 828 static 829 int 830 netif_set_tapaddr(int tap_unit, in_addr_t addr, in_addr_t mask, int s) 831 { 832 struct ifaliasreq ifra; 833 struct sockaddr_in *in; 834 835 bzero(&ifra, sizeof(ifra)); 836 snprintf(ifra.ifra_name, sizeof(ifra.ifra_name), "tap%d", tap_unit); 837 838 /* Setup address */ 839 in = (struct sockaddr_in *)&ifra.ifra_addr; 840 in->sin_family = AF_INET; 841 in->sin_len = sizeof(*in); 842 in->sin_addr.s_addr = addr; 843 844 if (mask != 0) { 845 /* Setup netmask */ 846 in = (struct sockaddr_in *)&ifra.ifra_mask; 847 in->sin_len = sizeof(*in); 848 in->sin_addr.s_addr = mask; 849 } 850 851 if (ioctl(s, SIOCAIFADDR, &ifra) < 0) { 852 warn("tap%d: ioctl(SIOCAIFADDR) failed", tap_unit); 853 return -1; 854 } 855 return 0; 856 } 857 858 static 859 int 860 netif_add_tap2brg(int tap_unit, const char *ifbridge, int s) 861 { 862 struct ifbreq ifbr; 863 struct ifdrv ifd; 864 865 bzero(&ifbr, sizeof(ifbr)); 866 snprintf(ifbr.ifbr_ifsname, sizeof(ifbr.ifbr_ifsname), 867 "tap%d", tap_unit); 868 869 bzero(&ifd, sizeof(ifd)); 870 strlcpy(ifd.ifd_name, ifbridge, sizeof(ifd.ifd_name)); 871 ifd.ifd_cmd = BRDGADD; 872 ifd.ifd_len = sizeof(ifbr); 873 ifd.ifd_data = &ifbr; 874 875 if (ioctl(s, SIOCSDRVSPEC, &ifd) < 0) { 876 /* 877 * 'errno == EEXIST' means that the tap(4) is already 878 * a member of the bridge(4) 879 */ 880 if (errno != EEXIST) { 881 warn("ioctl(%s, SIOCSDRVSPEC) failed", ifbridge); 882 return -1; 883 } 884 } 885 return 0; 886 } 887 888 #define TAPDEV_OFLAGS (O_RDWR | O_NONBLOCK) 889 890 /* 891 * Locate the first unused tap(4) device file if auto mode is requested, 892 * or open the user supplied device file, and bring up the corresponding 893 * tap(4) interface. 894 * 895 * NOTE: Only tap(4) device file is supported currently 896 */ 897 static 898 int 899 netif_open_tap(const char *netif, int *tap_unit, int s) 900 { 901 char tap_dev[MAXPATHLEN]; 902 int tap_fd, failed; 903 struct stat st; 904 char *dname; 905 906 *tap_unit = -1; 907 908 if (strcmp(netif, "auto") == 0) { 909 /* 910 * Find first unused tap(4) device file 911 */ 912 tap_fd = open("/dev/tap", TAPDEV_OFLAGS); 913 if (tap_fd < 0) { 914 warnc(errno, "Unable to find a free tap(4)"); 915 return -1; 916 } 917 } else { 918 /* 919 * User supplied tap(4) device file or unix socket. 920 */ 921 if (netif[0] == '/') /* Absolute path */ 922 strlcpy(tap_dev, netif, sizeof(tap_dev)); 923 else 924 snprintf(tap_dev, sizeof(tap_dev), "/dev/%s", netif); 925 926 tap_fd = open(tap_dev, TAPDEV_OFLAGS); 927 928 /* 929 * If we cannot open normally try to connect to it. 930 */ 931 if (tap_fd < 0) 932 tap_fd = unix_connect(tap_dev); 933 934 if (tap_fd < 0) { 935 warn("Unable to open %s", tap_dev); 936 return -1; 937 } 938 } 939 940 /* 941 * Check whether the device file is a tap(4) 942 */ 943 if (fstat(tap_fd, &st) < 0) { 944 failed = 1; 945 } else if (S_ISCHR(st.st_mode)) { 946 dname = fdevname(tap_fd); 947 if (dname) 948 dname = strstr(dname, "tap"); 949 if (dname) { 950 /* 951 * Bring up the corresponding tap(4) interface 952 */ 953 *tap_unit = strtol(dname + 3, NULL, 10); 954 printf("TAP UNIT %d\n", *tap_unit); 955 if (netif_set_tapflags(*tap_unit, IFF_UP, s) == 0) 956 failed = 0; 957 else 958 failed = 1; 959 } else { 960 failed = 1; 961 } 962 } else if (S_ISSOCK(st.st_mode)) { 963 /* 964 * Special socket connection (typically to vknet). We 965 * do not have to do anything. 966 */ 967 failed = 0; 968 } else { 969 failed = 1; 970 } 971 972 if (failed) { 973 warnx("%s is not a tap(4) device or socket", tap_dev); 974 close(tap_fd); 975 tap_fd = -1; 976 *tap_unit = -1; 977 } 978 return tap_fd; 979 } 980 981 static int 982 unix_connect(const char *path) 983 { 984 struct sockaddr_un sunx; 985 int len; 986 int net_fd; 987 int sndbuf = 262144; 988 struct stat st; 989 990 snprintf(sunx.sun_path, sizeof(sunx.sun_path), "%s", path); 991 len = offsetof(struct sockaddr_un, sun_path[strlen(sunx.sun_path)]); 992 ++len; /* include nul */ 993 sunx.sun_family = AF_UNIX; 994 sunx.sun_len = len; 995 996 net_fd = socket(AF_UNIX, SOCK_SEQPACKET, 0); 997 if (net_fd < 0) 998 return(-1); 999 if (connect(net_fd, (void *)&sunx, len) < 0) { 1000 close(net_fd); 1001 return(-1); 1002 } 1003 setsockopt(net_fd, SOL_SOCKET, SO_SNDBUF, &sndbuf, sizeof(sndbuf)); 1004 if (fstat(net_fd, &st) == 0) 1005 printf("Network socket buffer: %d bytes\n", st.st_blksize); 1006 fcntl(net_fd, F_SETFL, O_NONBLOCK); 1007 return(net_fd); 1008 } 1009 1010 #undef TAPDEV_MAJOR 1011 #undef TAPDEV_MINOR 1012 #undef TAPDEV_OFLAGS 1013 1014 /* 1015 * Following syntax is supported, 1016 * 1) x.x.x.x tap(4)'s address is x.x.x.x 1017 * 1018 * 2) x.x.x.x/z tap(4)'s address is x.x.x.x 1019 * tap(4)'s netmask len is z 1020 * 1021 * 3) x.x.x.x:y.y.y.y tap(4)'s address is x.x.x.x 1022 * pseudo netif's address is y.y.y.y 1023 * 1024 * 4) x.x.x.x:y.y.y.y/z tap(4)'s address is x.x.x.x 1025 * pseudo netif's address is y.y.y.y 1026 * tap(4) and pseudo netif's netmask len are z 1027 * 1028 * 5) bridgeX tap(4) will be added to bridgeX 1029 * 1030 * 6) bridgeX:y.y.y.y tap(4) will be added to bridgeX 1031 * pseudo netif's address is y.y.y.y 1032 * 1033 * 7) bridgeX:y.y.y.y/z tap(4) will be added to bridgeX 1034 * pseudo netif's address is y.y.y.y 1035 * pseudo netif's netmask len is z 1036 */ 1037 static 1038 int 1039 netif_init_tap(int tap_unit, in_addr_t *addr, in_addr_t *mask, int s) 1040 { 1041 in_addr_t tap_addr, netmask, netif_addr; 1042 int next_netif_addr; 1043 char *tok, *masklen_str, *ifbridge; 1044 1045 *addr = 0; 1046 *mask = 0; 1047 1048 tok = strtok(NULL, ":/"); 1049 if (tok == NULL) { 1050 /* 1051 * Nothing special, simply use tap(4) as backend 1052 */ 1053 return 0; 1054 } 1055 1056 if (inet_pton(AF_INET, tok, &tap_addr) > 0) { 1057 /* 1058 * tap(4)'s address is supplied 1059 */ 1060 ifbridge = NULL; 1061 1062 /* 1063 * If there is next token, then it may be pseudo 1064 * netif's address or netmask len for tap(4) 1065 */ 1066 next_netif_addr = 0; 1067 } else { 1068 /* 1069 * Not tap(4)'s address, assume it as a bridge(4) 1070 * iface name 1071 */ 1072 tap_addr = 0; 1073 ifbridge = tok; 1074 1075 /* 1076 * If there is next token, then it must be pseudo 1077 * netif's address 1078 */ 1079 next_netif_addr = 1; 1080 } 1081 1082 netmask = netif_addr = 0; 1083 1084 tok = strtok(NULL, ":/"); 1085 if (tok == NULL) 1086 goto back; 1087 1088 if (inet_pton(AF_INET, tok, &netif_addr) <= 0) { 1089 if (next_netif_addr) { 1090 warnx("Invalid pseudo netif address: %s", tok); 1091 return -1; 1092 } 1093 netif_addr = 0; 1094 1095 /* 1096 * Current token is not address, then it must be netmask len 1097 */ 1098 masklen_str = tok; 1099 } else { 1100 /* 1101 * Current token is pseudo netif address, if there is next token 1102 * it must be netmask len 1103 */ 1104 masklen_str = strtok(NULL, "/"); 1105 } 1106 1107 /* Calculate netmask */ 1108 if (masklen_str != NULL) { 1109 u_long masklen; 1110 1111 masklen = strtoul(masklen_str, NULL, 10); 1112 if (masklen < 32 && masklen > 0) { 1113 netmask = htonl(~((1LL << (32 - masklen)) - 1) 1114 & 0xffffffff); 1115 } else { 1116 warnx("Invalid netmask len: %lu", masklen); 1117 return -1; 1118 } 1119 } 1120 1121 /* Make sure there is no more token left */ 1122 if (strtok(NULL, ":/") != NULL) { 1123 warnx("Invalid argument to '-I'"); 1124 return -1; 1125 } 1126 1127 back: 1128 if (tap_unit < 0) { 1129 /* Do nothing */ 1130 } else if (ifbridge == NULL) { 1131 /* Set tap(4) address/netmask */ 1132 if (netif_set_tapaddr(tap_unit, tap_addr, netmask, s) < 0) 1133 return -1; 1134 } else { 1135 /* Tie tap(4) to bridge(4) */ 1136 if (netif_add_tap2brg(tap_unit, ifbridge, s) < 0) 1137 return -1; 1138 } 1139 1140 *addr = netif_addr; 1141 *mask = netmask; 1142 return 0; 1143 } 1144 1145 /* 1146 * NetifInfo[] will be filled for pseudo netif initialization. 1147 * NetifNum will be bumped to reflect the number of valid entries 1148 * in NetifInfo[]. 1149 */ 1150 static 1151 void 1152 init_netif(char *netifExp[], int netifExpNum) 1153 { 1154 int i, s; 1155 1156 if (netifExpNum == 0) 1157 return; 1158 1159 s = socket(AF_INET, SOCK_DGRAM, 0); /* for ioctl(SIOC) */ 1160 if (s < 0) 1161 return; 1162 1163 for (i = 0; i < netifExpNum; ++i) { 1164 struct vknetif_info *info; 1165 in_addr_t netif_addr, netif_mask; 1166 int tap_fd, tap_unit; 1167 char *netif; 1168 1169 netif = strtok(netifExp[i], ":"); 1170 if (netif == NULL) { 1171 warnx("Invalid argument to '-I'"); 1172 continue; 1173 } 1174 1175 /* 1176 * Open tap(4) device file and bring up the 1177 * corresponding interface 1178 */ 1179 tap_fd = netif_open_tap(netif, &tap_unit, s); 1180 if (tap_fd < 0) 1181 continue; 1182 1183 /* 1184 * Initialize tap(4) and get address/netmask 1185 * for pseudo netif 1186 * 1187 * NB: Rest part of netifExp[i] is passed 1188 * to netif_init_tap() implicitly. 1189 */ 1190 if (netif_init_tap(tap_unit, &netif_addr, &netif_mask, s) < 0) { 1191 /* 1192 * NB: Closing tap(4) device file will bring 1193 * down the corresponding interface 1194 */ 1195 close(tap_fd); 1196 continue; 1197 } 1198 1199 info = &NetifInfo[NetifNum]; 1200 info->tap_fd = tap_fd; 1201 info->tap_unit = tap_unit; 1202 info->netif_addr = netif_addr; 1203 info->netif_mask = netif_mask; 1204 1205 NetifNum++; 1206 if (NetifNum >= VKNETIF_MAX) /* XXX will this happen? */ 1207 break; 1208 } 1209 close(s); 1210 } 1211 1212 /* 1213 * Create the pid file and leave it open and locked while the vkernel is 1214 * running. This allows a script to use /usr/bin/lockf to probe whether 1215 * a vkernel is still running (so as not to accidently kill an unrelated 1216 * process from a stale pid file). 1217 */ 1218 static 1219 void 1220 writepid(void) 1221 { 1222 char buf[32]; 1223 int fd; 1224 1225 if (pid_file != NULL) { 1226 snprintf(buf, sizeof(buf), "%ld\n", (long)getpid()); 1227 fd = open(pid_file, O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0666); 1228 if (fd < 0) { 1229 if (errno == EWOULDBLOCK) { 1230 perror("Failed to lock pidfile, " 1231 "vkernel already running"); 1232 } else { 1233 perror("Failed to create pidfile"); 1234 } 1235 exit(EX_SOFTWARE); 1236 } 1237 ftruncate(fd, 0); 1238 write(fd, buf, strlen(buf)); 1239 /* leave the file open to maintain the lock */ 1240 } 1241 } 1242 1243 static 1244 void 1245 cleanpid( void ) 1246 { 1247 if (pid_file != NULL) { 1248 if (unlink(pid_file) < 0) 1249 perror("Warning: couldn't remove pidfile"); 1250 } 1251 } 1252 1253 static 1254 void 1255 usage_err(const char *ctl, ...) 1256 { 1257 va_list va; 1258 1259 va_start(va, ctl); 1260 vfprintf(stderr, ctl, va); 1261 va_end(va); 1262 fprintf(stderr, "\n"); 1263 exit(EX_USAGE); 1264 } 1265 1266 static 1267 void 1268 usage_help(_Bool help) 1269 { 1270 fprintf(stderr, "Usage: %s [-hsUv] [-c file] [-e name=value:name=value:...]\n" 1271 "\t[-i file] [-I interface[:address1[:address2][/netmask]]] [-l cpulock]\n" 1272 "\t[-m size] [-n numcpus] [-p file] [-r file]\n", save_av[0]); 1273 1274 if (help) 1275 fprintf(stderr, "\nArguments:\n" 1276 "\t-c\tSpecify a readonly CD-ROM image file to be used by the kernel.\n" 1277 "\t-e\tSpecify an environment to be used by the kernel.\n" 1278 "\t-h\tThis list of options.\n" 1279 "\t-i\tSpecify a memory image file to be used by the virtual kernel.\n" 1280 "\t-I\tCreate a virtual network device.\n" 1281 "\t-l\tSpecify which, if any, real CPUs to lock virtual CPUs to.\n" 1282 "\t-m\tSpecify the amount of memory to be used by the kernel in bytes.\n" 1283 "\t-n\tSpecify the number of CPUs you wish to emulate.\n" 1284 "\t-p\tSpecify a file in which to store the process ID.\n" 1285 "\t-r\tSpecify a R/W disk image file to be used by the kernel.\n" 1286 "\t-s\tBoot into single-user mode.\n" 1287 "\t-U\tEnable writing to kernel memory and module loading.\n" 1288 "\t-v\tTurn on verbose booting.\n"); 1289 1290 exit(EX_USAGE); 1291 } 1292 1293 void 1294 cpu_reset(void) 1295 { 1296 kprintf("cpu reset, rebooting vkernel\n"); 1297 closefrom(3); 1298 cleanpid(); 1299 execv(save_av[0], save_av); 1300 } 1301 1302 void 1303 cpu_halt(void) 1304 { 1305 kprintf("cpu halt, exiting vkernel\n"); 1306 cleanpid(); 1307 exit(EX_OK); 1308 } 1309 1310 void 1311 setrealcpu(void) 1312 { 1313 switch(lwp_cpu_lock) { 1314 case LCL_PER_CPU: 1315 if (bootverbose) 1316 kprintf("Locking CPU%d to real cpu %d\n", 1317 mycpuid, next_cpu); 1318 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu)); 1319 next_cpu++; 1320 if (next_cpu >= real_ncpus) 1321 next_cpu = 0; 1322 break; 1323 case LCL_SINGLE_CPU: 1324 if (bootverbose) 1325 kprintf("Locking CPU%d to real cpu %d\n", 1326 mycpuid, next_cpu); 1327 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu)); 1328 break; 1329 default: 1330 /* do not map virtual cpus to real cpus */ 1331 break; 1332 } 1333 } 1334 1335 /* 1336 * Allocate and free memory for module loading. The loaded module 1337 * has to be placed somewhere near the current kernel binary load 1338 * point or the relocations will not work. 1339 * 1340 * I'm not sure why this isn't working. 1341 */ 1342 int 1343 vkernel_module_memory_alloc(vm_offset_t *basep, size_t bytes) 1344 { 1345 kprintf("module loading for vkernel64's not currently supported\n"); 1346 *basep = 0; 1347 return ENOMEM; 1348 #if 0 1349 #if 1 1350 size_t xtra; 1351 xtra = (PAGE_SIZE - (vm_offset_t)sbrk(0)) & PAGE_MASK; 1352 *basep = (vm_offset_t)sbrk(xtra + bytes) + xtra; 1353 bzero((void *)*basep, bytes); 1354 #else 1355 *basep = (vm_offset_t)mmap((void *)0x000000000, bytes, 1356 PROT_READ|PROT_WRITE|PROT_EXEC, 1357 MAP_ANON|MAP_SHARED, -1, 0); 1358 if ((void *)*basep == MAP_FAILED) 1359 return ENOMEM; 1360 #endif 1361 kprintf("basep %p %p %zd\n", 1362 (void *)vkernel_module_memory_alloc, (void *)*basep, bytes); 1363 return 0; 1364 #endif 1365 } 1366 1367 void 1368 vkernel_module_memory_free(vm_offset_t base, size_t bytes) 1369 { 1370 #if 0 1371 #if 0 1372 munmap((void *)base, bytes); 1373 #endif 1374 #endif 1375 } 1376