1 /* $NetBSD: loadfile.c,v 1.10 2000/12/03 02:53:04 tsutsui Exp $ */ 2 /* $OpenBSD: loadfile_elf.c,v 1.38 2021/04/05 18:09:48 dv Exp $ */ 3 4 /*- 5 * Copyright (c) 1997 The NetBSD Foundation, Inc. 6 * All rights reserved. 7 * 8 * This code is derived from software contributed to The NetBSD Foundation 9 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility, 10 * NASA Ames Research Center and by Christos Zoulas. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 23 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 24 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 25 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 26 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 27 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 28 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 29 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 30 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 31 * POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 /* 35 * Copyright (c) 1992, 1993 36 * The Regents of the University of California. All rights reserved. 37 * 38 * This code is derived from software contributed to Berkeley by 39 * Ralph Campbell. 40 * 41 * Redistribution and use in source and binary forms, with or without 42 * modification, are permitted provided that the following conditions 43 * are met: 44 * 1. Redistributions of source code must retain the above copyright 45 * notice, this list of conditions and the following disclaimer. 46 * 2. Redistributions in binary form must reproduce the above copyright 47 * notice, this list of conditions and the following disclaimer in the 48 * documentation and/or other materials provided with the distribution. 49 * 3. Neither the name of the University nor the names of its contributors 50 * may be used to endorse or promote products derived from this software 51 * without specific prior written permission. 52 * 53 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 54 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 55 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 56 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 57 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 58 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 59 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 60 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 61 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 62 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 63 * SUCH DAMAGE. 64 * 65 * @(#)boot.c 8.1 (Berkeley) 6/10/93 66 */ 67 68 /* 69 * Copyright (c) 2015 Mike Larkin <mlarkin@openbsd.org> 70 * 71 * Permission to use, copy, modify, and distribute this software for any 72 * purpose with or without fee is hereby granted, provided that the above 73 * copyright notice and this permission notice appear in all copies. 74 * 75 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 76 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 77 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 78 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 79 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 80 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 81 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 82 */ 83 84 #include <sys/param.h> /* PAGE_SIZE PAGE_MASK roundup */ 85 #include <sys/ioctl.h> 86 #include <sys/reboot.h> 87 #include <sys/exec.h> 88 89 #include <elf.h> 90 #include <stdio.h> 91 #include <string.h> 92 #include <errno.h> 93 #include <stdlib.h> 94 #include <unistd.h> 95 #include <fcntl.h> 96 #include <err.h> 97 #include <errno.h> 98 #include <stddef.h> 99 100 #include <machine/vmmvar.h> 101 #include <machine/biosvar.h> 102 #include <machine/segments.h> 103 #include <machine/specialreg.h> 104 #include <machine/pte.h> 105 106 #include "loadfile.h" 107 #include "vmd.h" 108 109 #define LOADADDR(a) ((((u_long)(a)) + offset)&0xfffffff) 110 111 union { 112 Elf32_Ehdr elf32; 113 Elf64_Ehdr elf64; 114 } hdr; 115 116 static void setsegment(struct mem_segment_descriptor *, uint32_t, 117 size_t, int, int, int, int); 118 static int elf32_exec(gzFile, Elf32_Ehdr *, u_long *, int); 119 static int elf64_exec(gzFile, Elf64_Ehdr *, u_long *, int); 120 static size_t create_bios_memmap(struct vm_create_params *, bios_memmap_t *); 121 static uint32_t push_bootargs(bios_memmap_t *, size_t); 122 static size_t push_stack(uint32_t, uint32_t); 123 static void push_gdt(void); 124 static void push_pt_32(void); 125 static void push_pt_64(void); 126 static void marc4random_buf(paddr_t, int); 127 static void mbzero(paddr_t, int); 128 static void mbcopy(void *, paddr_t, int); 129 130 extern char *__progname; 131 extern int vm_id; 132 133 /* 134 * setsegment 135 * 136 * Initializes a segment selector entry with the provided descriptor. 137 * For the purposes of the bootloader mimiced by vmd(8), we only need 138 * memory-type segment descriptor support. 139 * 140 * This function was copied from machdep.c 141 * 142 * Parameters: 143 * sd: Address of the entry to initialize 144 * base: base of the segment 145 * limit: limit of the segment 146 * type: type of the segment 147 * dpl: privilege level of the egment 148 * def32: default 16/32 bit size of the segment 149 * gran: granularity of the segment (byte/page) 150 */ 151 static void 152 setsegment(struct mem_segment_descriptor *sd, uint32_t base, size_t limit, 153 int type, int dpl, int def32, int gran) 154 { 155 sd->sd_lolimit = (int)limit; 156 sd->sd_lobase = (int)base; 157 sd->sd_type = type; 158 sd->sd_dpl = dpl; 159 sd->sd_p = 1; 160 sd->sd_hilimit = (int)limit >> 16; 161 sd->sd_avl = 0; 162 sd->sd_long = 0; 163 sd->sd_def32 = def32; 164 sd->sd_gran = gran; 165 sd->sd_hibase = (int)base >> 24; 166 } 167 168 /* 169 * push_gdt 170 * 171 * Allocates and populates a page in the guest phys memory space to hold 172 * the boot-time GDT. Since vmd(8) is acting as the bootloader, we need to 173 * create the same GDT that a real bootloader would have created. 174 * This is loaded into the guest phys RAM space at address GDT_PAGE. 175 */ 176 static void 177 push_gdt(void) 178 { 179 uint8_t gdtpage[PAGE_SIZE]; 180 struct mem_segment_descriptor *sd; 181 182 memset(&gdtpage, 0, sizeof(gdtpage)); 183 184 sd = (struct mem_segment_descriptor *)&gdtpage; 185 186 /* 187 * Create three segment descriptors: 188 * 189 * GDT[0] : null desriptor. "Created" via memset above. 190 * GDT[1] (selector @ 0x8): Executable segment, for CS 191 * GDT[2] (selector @ 0x10): RW Data segment, for DS/ES/SS 192 */ 193 setsegment(&sd[1], 0, 0xffffffff, SDT_MEMERA, SEL_KPL, 1, 1); 194 setsegment(&sd[2], 0, 0xffffffff, SDT_MEMRWA, SEL_KPL, 1, 1); 195 196 write_mem(GDT_PAGE, gdtpage, PAGE_SIZE); 197 } 198 199 /* 200 * push_pt_32 201 * 202 * Create an identity-mapped page directory hierarchy mapping the first 203 * 4GB of physical memory. This is used during bootstrapping i386 VMs on 204 * CPUs without unrestricted guest capability. 205 */ 206 static void 207 push_pt_32(void) 208 { 209 uint32_t ptes[1024], i; 210 211 memset(ptes, 0, sizeof(ptes)); 212 for (i = 0 ; i < 1024; i++) { 213 ptes[i] = PG_V | PG_RW | PG_u | PG_PS | ((4096 * 1024) * i); 214 } 215 write_mem(PML3_PAGE, ptes, PAGE_SIZE); 216 } 217 218 /* 219 * push_pt_64 220 * 221 * Create an identity-mapped page directory hierarchy mapping the first 222 * 1GB of physical memory. This is used during bootstrapping 64 bit VMs on 223 * CPUs without unrestricted guest capability. 224 */ 225 static void 226 push_pt_64(void) 227 { 228 uint64_t ptes[512], i; 229 230 /* PDPDE0 - first 1GB */ 231 memset(ptes, 0, sizeof(ptes)); 232 ptes[0] = PG_V | PML3_PAGE; 233 write_mem(PML4_PAGE, ptes, PAGE_SIZE); 234 235 /* PDE0 - first 1GB */ 236 memset(ptes, 0, sizeof(ptes)); 237 ptes[0] = PG_V | PG_RW | PG_u | PML2_PAGE; 238 write_mem(PML3_PAGE, ptes, PAGE_SIZE); 239 240 /* First 1GB (in 2MB pages) */ 241 memset(ptes, 0, sizeof(ptes)); 242 for (i = 0 ; i < 512; i++) { 243 ptes[i] = PG_V | PG_RW | PG_u | PG_PS | ((2048 * 1024) * i); 244 } 245 write_mem(PML2_PAGE, ptes, PAGE_SIZE); 246 } 247 248 /* 249 * loadfile_elf 250 * 251 * Loads an ELF kernel to it's defined load address in the guest VM. 252 * The kernel is loaded to its defined start point as set in the ELF header. 253 * 254 * Parameters: 255 * fp: file of a kernel file to load 256 * vcp: the VM create parameters, holding the exact memory map 257 * (out) vrs: register state to set on init for this kernel 258 * bootdev: the optional non-default boot device 259 * howto: optional boot flags for the kernel 260 * 261 * Return values: 262 * 0 if successful 263 * various error codes returned from gzread(3) or loadelf functions 264 */ 265 int 266 loadfile_elf(gzFile fp, struct vm_create_params *vcp, 267 struct vcpu_reg_state *vrs) 268 { 269 int r, is_i386 = 0; 270 uint32_t bootargsz; 271 size_t n, stacksize; 272 u_long marks[MARK_MAX]; 273 bios_memmap_t memmap[VMM_MAX_MEM_RANGES + 1]; 274 275 if ((r = gzread(fp, &hdr, sizeof(hdr))) != sizeof(hdr)) 276 return 1; 277 278 memset(&marks, 0, sizeof(marks)); 279 if (memcmp(hdr.elf32.e_ident, ELFMAG, SELFMAG) == 0 && 280 hdr.elf32.e_ident[EI_CLASS] == ELFCLASS32) { 281 r = elf32_exec(fp, &hdr.elf32, marks, LOAD_ALL); 282 is_i386 = 1; 283 } else if (memcmp(hdr.elf64.e_ident, ELFMAG, SELFMAG) == 0 && 284 hdr.elf64.e_ident[EI_CLASS] == ELFCLASS64) { 285 r = elf64_exec(fp, &hdr.elf64, marks, LOAD_ALL); 286 } else 287 errno = ENOEXEC; 288 289 if (r) 290 return (r); 291 292 push_gdt(); 293 294 if (is_i386) { 295 push_pt_32(); 296 /* Reconfigure the default flat-64 register set for 32 bit */ 297 vrs->vrs_crs[VCPU_REGS_CR3] = PML3_PAGE; 298 vrs->vrs_crs[VCPU_REGS_CR4] = CR4_PSE; 299 vrs->vrs_msrs[VCPU_REGS_EFER] = 0ULL; 300 } 301 else 302 push_pt_64(); 303 304 n = create_bios_memmap(vcp, memmap); 305 bootargsz = push_bootargs(memmap, n); 306 stacksize = push_stack(bootargsz, marks[MARK_END]); 307 308 vrs->vrs_gprs[VCPU_REGS_RIP] = (uint64_t)marks[MARK_ENTRY]; 309 vrs->vrs_gprs[VCPU_REGS_RSP] = (uint64_t)(STACK_PAGE + PAGE_SIZE) - stacksize; 310 vrs->vrs_gdtr.vsi_base = GDT_PAGE; 311 312 log_debug("%s: loaded ELF kernel", __func__); 313 314 return (0); 315 } 316 317 /* 318 * create_bios_memmap 319 * 320 * Construct a memory map as returned by the BIOS INT 0x15, e820 routine. 321 * 322 * Parameters: 323 * vcp: the VM create parameters, containing the memory map passed to vmm(4) 324 * memmap (out): the BIOS memory map 325 * 326 * Return values: 327 * Number of bios_memmap_t entries, including the terminating nul-entry. 328 */ 329 static size_t 330 create_bios_memmap(struct vm_create_params *vcp, bios_memmap_t *memmap) 331 { 332 size_t i, n = 0, sz; 333 paddr_t gpa; 334 struct vm_mem_range *vmr; 335 336 for (i = 0; i < vcp->vcp_nmemranges; i++) { 337 vmr = &vcp->vcp_memranges[i]; 338 gpa = vmr->vmr_gpa; 339 sz = vmr->vmr_size; 340 341 /* 342 * Make sure that we do not mark the ROM/video RAM area in the 343 * low memory as physcal memory available to the kernel. 344 */ 345 if (gpa < 0x100000 && gpa + sz > LOWMEM_KB * 1024) { 346 if (gpa >= LOWMEM_KB * 1024) 347 sz = 0; 348 else 349 sz = LOWMEM_KB * 1024 - gpa; 350 } 351 352 if (sz != 0) { 353 memmap[n].addr = gpa; 354 memmap[n].size = sz; 355 memmap[n].type = 0x1; /* Type 1 : Normal memory */ 356 n++; 357 } 358 } 359 360 /* Null mem map entry to denote the end of the ranges */ 361 memmap[n].addr = 0x0; 362 memmap[n].size = 0x0; 363 memmap[n].type = 0x0; 364 n++; 365 366 return (n); 367 } 368 369 /* 370 * push_bootargs 371 * 372 * Creates the boot arguments page in the guest address space. 373 * Since vmd(8) is acting as the bootloader, we need to create the same boot 374 * arguments page that a real bootloader would have created. This is loaded 375 * into the guest phys RAM space at address BOOTARGS_PAGE. 376 * 377 * Parameters: 378 * memmap: the BIOS memory map 379 * n: number of entries in memmap 380 * 381 * Return values: 382 * The size of the bootargs 383 */ 384 static uint32_t 385 push_bootargs(bios_memmap_t *memmap, size_t n) 386 { 387 uint32_t memmap_sz, consdev_sz, i; 388 bios_consdev_t consdev; 389 uint32_t ba[1024]; 390 391 memmap_sz = 3 * sizeof(int) + n * sizeof(bios_memmap_t); 392 ba[0] = 0x0; /* memory map */ 393 ba[1] = memmap_sz; 394 ba[2] = memmap_sz; /* next */ 395 memcpy(&ba[3], memmap, n * sizeof(bios_memmap_t)); 396 i = memmap_sz / sizeof(int); 397 398 /* Serial console device, COM1 @ 0x3f8 */ 399 consdev.consdev = makedev(8, 0); /* com1 @ 0x3f8 */ 400 consdev.conspeed = 115200; 401 consdev.consaddr = 0x3f8; 402 consdev.consfreq = 0; 403 404 consdev_sz = 3 * sizeof(int) + sizeof(bios_consdev_t); 405 ba[i] = 0x5; /* consdev */ 406 ba[i + 1] = consdev_sz; 407 ba[i + 2] = consdev_sz; 408 memcpy(&ba[i + 3], &consdev, sizeof(bios_consdev_t)); 409 i += consdev_sz / sizeof(int); 410 411 ba[i++] = 0xFFFFFFFF; /* BOOTARG_END */ 412 413 write_mem(BOOTARGS_PAGE, ba, PAGE_SIZE); 414 415 return (i * sizeof(int)); 416 } 417 418 /* 419 * push_stack 420 * 421 * Creates the boot stack page in the guest address space. When using a real 422 * bootloader, the stack will be prepared using the following format before 423 * transitioning to kernel start, so vmd(8) needs to mimic the same stack 424 * layout. The stack content is pushed to the guest phys RAM at address 425 * STACK_PAGE. The bootloader operates in 32 bit mode; each stack entry is 426 * 4 bytes. 427 * 428 * Stack Layout: (TOS == Top Of Stack) 429 * TOS location of boot arguments page 430 * TOS - 0x4 size of the content in the boot arguments page 431 * TOS - 0x8 size of low memory (biosbasemem: kernel uses BIOS map only if 0) 432 * TOS - 0xc size of high memory (biosextmem, not used by kernel at all) 433 * TOS - 0x10 kernel 'end' symbol value 434 * TOS - 0x14 version of bootarg API 435 * 436 * Parameters: 437 * bootargsz: size of boot arguments 438 * end: kernel 'end' symbol value 439 * bootdev: the optional non-default boot device 440 * howto: optional boot flags for the kernel 441 * 442 * Return values: 443 * size of the stack 444 */ 445 static size_t 446 push_stack(uint32_t bootargsz, uint32_t end) 447 { 448 uint32_t stack[1024]; 449 uint16_t loc; 450 451 memset(&stack, 0, sizeof(stack)); 452 loc = 1024; 453 454 stack[--loc] = BOOTARGS_PAGE; 455 stack[--loc] = bootargsz; 456 stack[--loc] = 0; /* biosbasemem */ 457 stack[--loc] = 0; /* biosextmem */ 458 stack[--loc] = end; 459 stack[--loc] = 0x0e; 460 stack[--loc] = MAKEBOOTDEV(0x4, 0, 0, 0, 0); /* bootdev: sd0a */ 461 stack[--loc] = 0; 462 463 write_mem(STACK_PAGE, &stack, PAGE_SIZE); 464 465 return (1024 - (loc - 1)) * sizeof(uint32_t); 466 } 467 468 /* 469 * mread 470 * 471 * Reads 'sz' bytes from the file whose descriptor is provided in 'fd' 472 * into the guest address space at paddr 'addr'. 473 * 474 * Parameters: 475 * fp: kernel image file to read from. 476 * addr: guest paddr_t to load to 477 * sz: number of bytes to load 478 * 479 * Return values: 480 * returns 'sz' if successful, or 0 otherwise. 481 */ 482 size_t 483 mread(gzFile fp, paddr_t addr, size_t sz) 484 { 485 size_t ct; 486 size_t i, rd, osz; 487 char buf[PAGE_SIZE]; 488 489 /* 490 * break up the 'sz' bytes into PAGE_SIZE chunks for use with 491 * write_mem 492 */ 493 ct = 0; 494 rd = 0; 495 osz = sz; 496 if ((addr & PAGE_MASK) != 0) { 497 memset(buf, 0, sizeof(buf)); 498 if (sz > PAGE_SIZE) 499 ct = PAGE_SIZE - (addr & PAGE_MASK); 500 else 501 ct = sz; 502 503 if ((size_t)gzread(fp, buf, ct) != ct) { 504 log_warn("%s: error %d in mread", __progname, errno); 505 return (0); 506 } 507 rd += ct; 508 509 if (write_mem(addr, buf, ct)) 510 return (0); 511 512 addr += ct; 513 } 514 515 sz = sz - ct; 516 517 if (sz == 0) 518 return (osz); 519 520 for (i = 0; i < sz; i += PAGE_SIZE, addr += PAGE_SIZE) { 521 memset(buf, 0, sizeof(buf)); 522 if (i + PAGE_SIZE > sz) 523 ct = sz - i; 524 else 525 ct = PAGE_SIZE; 526 527 if ((size_t)gzread(fp, buf, ct) != ct) { 528 log_warn("%s: error %d in mread", __progname, errno); 529 return (0); 530 } 531 rd += ct; 532 533 if (write_mem(addr, buf, ct)) 534 return (0); 535 } 536 537 return (osz); 538 } 539 540 /* 541 * marc4random_buf 542 * 543 * load 'sz' bytes of random data into the guest address space at paddr 544 * 'addr'. 545 * 546 * Parameters: 547 * addr: guest paddr_t to load random bytes into 548 * sz: number of random bytes to load 549 * 550 * Return values: 551 * nothing 552 */ 553 static void 554 marc4random_buf(paddr_t addr, int sz) 555 { 556 int i, ct; 557 char buf[PAGE_SIZE]; 558 559 /* 560 * break up the 'sz' bytes into PAGE_SIZE chunks for use with 561 * write_mem 562 */ 563 ct = 0; 564 if (addr % PAGE_SIZE != 0) { 565 memset(buf, 0, sizeof(buf)); 566 ct = PAGE_SIZE - (addr % PAGE_SIZE); 567 568 arc4random_buf(buf, ct); 569 570 if (write_mem(addr, buf, ct)) 571 return; 572 573 addr += ct; 574 } 575 576 for (i = 0; i < sz; i+= PAGE_SIZE, addr += PAGE_SIZE) { 577 memset(buf, 0, sizeof(buf)); 578 if (i + PAGE_SIZE > sz) 579 ct = sz - i; 580 else 581 ct = PAGE_SIZE; 582 583 arc4random_buf(buf, ct); 584 585 if (write_mem(addr, buf, ct)) 586 return; 587 } 588 } 589 590 /* 591 * mbzero 592 * 593 * load 'sz' bytes of zeros into the guest address space at paddr 594 * 'addr'. 595 * 596 * Parameters: 597 * addr: guest paddr_t to zero 598 * sz: number of zero bytes to store 599 * 600 * Return values: 601 * nothing 602 */ 603 static void 604 mbzero(paddr_t addr, int sz) 605 { 606 if (write_mem(addr, NULL, sz)) 607 return; 608 } 609 610 /* 611 * mbcopy 612 * 613 * copies 'sz' bytes from buffer 'src' to guest paddr 'dst'. 614 * 615 * Parameters: 616 * src: source buffer to copy from 617 * dst: destination guest paddr_t to copy to 618 * sz: number of bytes to copy 619 * 620 * Return values: 621 * nothing 622 */ 623 static void 624 mbcopy(void *src, paddr_t dst, int sz) 625 { 626 write_mem(dst, src, sz); 627 } 628 629 /* 630 * elf64_exec 631 * 632 * Load the kernel indicated by 'fp' into the guest physical memory 633 * space, at the addresses defined in the ELF header. 634 * 635 * This function is used for 64 bit kernels. 636 * 637 * Parameters: 638 * fp: kernel image file to load 639 * elf: ELF header of the kernel 640 * marks: array to store the offsets of various kernel structures 641 * (start, bss, etc) 642 * flags: flag value to indicate which section(s) to load (usually 643 * LOAD_ALL) 644 * 645 * Return values: 646 * 0 if successful 647 * 1 if unsuccessful 648 */ 649 static int 650 elf64_exec(gzFile fp, Elf64_Ehdr *elf, u_long *marks, int flags) 651 { 652 Elf64_Shdr *shp; 653 Elf64_Phdr *phdr; 654 Elf64_Off off; 655 int i; 656 size_t sz; 657 int first; 658 int havesyms; 659 paddr_t minp = ~0, maxp = 0, pos = 0; 660 paddr_t offset = marks[MARK_START], shpp, elfp; 661 662 sz = elf->e_phnum * sizeof(Elf64_Phdr); 663 phdr = malloc(sz); 664 665 if (gzseek(fp, (off_t)elf->e_phoff, SEEK_SET) == -1) { 666 free(phdr); 667 return 1; 668 } 669 670 if ((size_t)gzread(fp, phdr, sz) != sz) { 671 free(phdr); 672 return 1; 673 } 674 675 for (first = 1, i = 0; i < elf->e_phnum; i++) { 676 if (phdr[i].p_type == PT_OPENBSD_RANDOMIZE) { 677 int m; 678 679 /* Fill segment if asked for. */ 680 if (flags & LOAD_RANDOM) { 681 for (pos = 0; pos < phdr[i].p_filesz; 682 pos += m) { 683 m = phdr[i].p_filesz - pos; 684 marc4random_buf(phdr[i].p_paddr + pos, 685 m); 686 } 687 } 688 if (flags & (LOAD_RANDOM | COUNT_RANDOM)) { 689 marks[MARK_RANDOM] = LOADADDR(phdr[i].p_paddr); 690 marks[MARK_ERANDOM] = 691 marks[MARK_RANDOM] + phdr[i].p_filesz; 692 } 693 continue; 694 } 695 696 if (phdr[i].p_type != PT_LOAD || 697 (phdr[i].p_flags & (PF_W|PF_R|PF_X)) == 0) 698 continue; 699 700 #define IS_TEXT(p) (p.p_flags & PF_X) 701 #define IS_DATA(p) ((p.p_flags & PF_X) == 0) 702 #define IS_BSS(p) (p.p_filesz < p.p_memsz) 703 /* 704 * XXX: Assume first address is lowest 705 */ 706 if ((IS_TEXT(phdr[i]) && (flags & LOAD_TEXT)) || 707 (IS_DATA(phdr[i]) && (flags & LOAD_DATA))) { 708 709 /* Read in segment. */ 710 if (gzseek(fp, (off_t)phdr[i].p_offset, 711 SEEK_SET) == -1) { 712 free(phdr); 713 return 1; 714 } 715 if (mread(fp, phdr[i].p_paddr, phdr[i].p_filesz) != 716 phdr[i].p_filesz) { 717 free(phdr); 718 return 1; 719 } 720 721 first = 0; 722 } 723 724 if ((IS_TEXT(phdr[i]) && (flags & (LOAD_TEXT | COUNT_TEXT))) || 725 (IS_DATA(phdr[i]) && (flags & (LOAD_DATA | COUNT_TEXT)))) { 726 pos = phdr[i].p_paddr; 727 if (minp > pos) 728 minp = pos; 729 pos += phdr[i].p_filesz; 730 if (maxp < pos) 731 maxp = pos; 732 } 733 734 /* Zero out BSS. */ 735 if (IS_BSS(phdr[i]) && (flags & LOAD_BSS)) { 736 mbzero((phdr[i].p_paddr + phdr[i].p_filesz), 737 phdr[i].p_memsz - phdr[i].p_filesz); 738 } 739 if (IS_BSS(phdr[i]) && (flags & (LOAD_BSS|COUNT_BSS))) { 740 pos += phdr[i].p_memsz - phdr[i].p_filesz; 741 if (maxp < pos) 742 maxp = pos; 743 } 744 } 745 free(phdr); 746 747 /* 748 * Copy the ELF and section headers. 749 */ 750 elfp = maxp = roundup(maxp, sizeof(Elf64_Addr)); 751 if (flags & (LOAD_HDR | COUNT_HDR)) 752 maxp += sizeof(Elf64_Ehdr); 753 754 if (flags & (LOAD_SYM | COUNT_SYM)) { 755 if (gzseek(fp, (off_t)elf->e_shoff, SEEK_SET) == -1) { 756 warn("gzseek section headers"); 757 return 1; 758 } 759 sz = elf->e_shnum * sizeof(Elf64_Shdr); 760 shp = malloc(sz); 761 762 if ((size_t)gzread(fp, shp, sz) != sz) { 763 free(shp); 764 return 1; 765 } 766 767 shpp = maxp; 768 maxp += roundup(sz, sizeof(Elf64_Addr)); 769 770 size_t shstrsz = shp[elf->e_shstrndx].sh_size; 771 char *shstr = malloc(shstrsz); 772 if (gzseek(fp, (off_t)shp[elf->e_shstrndx].sh_offset, 773 SEEK_SET) == -1) { 774 free(shstr); 775 free(shp); 776 return 1; 777 } 778 if ((size_t)gzread(fp, shstr, shstrsz) != shstrsz) { 779 free(shstr); 780 free(shp); 781 return 1; 782 } 783 784 /* 785 * Now load the symbol sections themselves. Make sure the 786 * sections are aligned. Don't bother with string tables if 787 * there are no symbol sections. 788 */ 789 off = roundup((sizeof(Elf64_Ehdr) + sz), sizeof(Elf64_Addr)); 790 791 for (havesyms = i = 0; i < elf->e_shnum; i++) 792 if (shp[i].sh_type == SHT_SYMTAB) 793 havesyms = 1; 794 795 for (first = 1, i = 0; i < elf->e_shnum; i++) { 796 if (shp[i].sh_type == SHT_SYMTAB || 797 shp[i].sh_type == SHT_STRTAB || 798 !strcmp(shstr + shp[i].sh_name, ".debug_line") || 799 !strcmp(shstr + shp[i].sh_name, ELF_CTF)) { 800 if (havesyms && (flags & LOAD_SYM)) { 801 if (gzseek(fp, (off_t)shp[i].sh_offset, 802 SEEK_SET) == -1) { 803 free(shstr); 804 free(shp); 805 return 1; 806 } 807 if (mread(fp, maxp, 808 shp[i].sh_size) != shp[i].sh_size) { 809 free(shstr); 810 free(shp); 811 return 1; 812 } 813 } 814 maxp += roundup(shp[i].sh_size, 815 sizeof(Elf64_Addr)); 816 shp[i].sh_offset = off; 817 shp[i].sh_flags |= SHF_ALLOC; 818 off += roundup(shp[i].sh_size, 819 sizeof(Elf64_Addr)); 820 first = 0; 821 } 822 } 823 if (flags & LOAD_SYM) { 824 mbcopy(shp, shpp, sz); 825 } 826 free(shstr); 827 free(shp); 828 } 829 830 /* 831 * Frob the copied ELF header to give information relative 832 * to elfp. 833 */ 834 if (flags & LOAD_HDR) { 835 elf->e_phoff = 0; 836 elf->e_shoff = sizeof(Elf64_Ehdr); 837 elf->e_phentsize = 0; 838 elf->e_phnum = 0; 839 mbcopy(elf, elfp, sizeof(*elf)); 840 } 841 842 marks[MARK_START] = LOADADDR(minp); 843 marks[MARK_ENTRY] = LOADADDR(elf->e_entry); 844 marks[MARK_NSYM] = 1; /* XXX: Kernel needs >= 0 */ 845 marks[MARK_SYM] = LOADADDR(elfp); 846 marks[MARK_END] = LOADADDR(maxp); 847 848 return 0; 849 } 850 851 /* 852 * elf32_exec 853 * 854 * Load the kernel indicated by 'fp' into the guest physical memory 855 * space, at the addresses defined in the ELF header. 856 * 857 * This function is used for 32 bit kernels. 858 * 859 * Parameters: 860 * fp: kernel image file to load 861 * elf: ELF header of the kernel 862 * marks: array to store the offsets of various kernel structures 863 * (start, bss, etc) 864 * flags: flag value to indicate which section(s) to load (usually 865 * LOAD_ALL) 866 * 867 * Return values: 868 * 0 if successful 869 * 1 if unsuccessful 870 */ 871 static int 872 elf32_exec(gzFile fp, Elf32_Ehdr *elf, u_long *marks, int flags) 873 { 874 Elf32_Shdr *shp; 875 Elf32_Phdr *phdr; 876 Elf32_Off off; 877 int i; 878 size_t sz; 879 int first; 880 int havesyms; 881 paddr_t minp = ~0, maxp = 0, pos = 0; 882 paddr_t offset = marks[MARK_START], shpp, elfp; 883 884 sz = elf->e_phnum * sizeof(Elf32_Phdr); 885 phdr = malloc(sz); 886 887 if (gzseek(fp, (off_t)elf->e_phoff, SEEK_SET) == -1) { 888 free(phdr); 889 return 1; 890 } 891 892 if ((size_t)gzread(fp, phdr, sz) != sz) { 893 free(phdr); 894 return 1; 895 } 896 897 for (first = 1, i = 0; i < elf->e_phnum; i++) { 898 if (phdr[i].p_type == PT_OPENBSD_RANDOMIZE) { 899 int m; 900 901 /* Fill segment if asked for. */ 902 if (flags & LOAD_RANDOM) { 903 for (pos = 0; pos < phdr[i].p_filesz; 904 pos += m) { 905 m = phdr[i].p_filesz - pos; 906 marc4random_buf(phdr[i].p_paddr + pos, 907 m); 908 } 909 } 910 if (flags & (LOAD_RANDOM | COUNT_RANDOM)) { 911 marks[MARK_RANDOM] = LOADADDR(phdr[i].p_paddr); 912 marks[MARK_ERANDOM] = 913 marks[MARK_RANDOM] + phdr[i].p_filesz; 914 } 915 continue; 916 } 917 918 if (phdr[i].p_type != PT_LOAD || 919 (phdr[i].p_flags & (PF_W|PF_R|PF_X)) == 0) 920 continue; 921 922 #define IS_TEXT(p) (p.p_flags & PF_X) 923 #define IS_DATA(p) ((p.p_flags & PF_X) == 0) 924 #define IS_BSS(p) (p.p_filesz < p.p_memsz) 925 /* 926 * XXX: Assume first address is lowest 927 */ 928 if ((IS_TEXT(phdr[i]) && (flags & LOAD_TEXT)) || 929 (IS_DATA(phdr[i]) && (flags & LOAD_DATA))) { 930 931 /* Read in segment. */ 932 if (gzseek(fp, (off_t)phdr[i].p_offset, 933 SEEK_SET) == -1) { 934 free(phdr); 935 return 1; 936 } 937 if (mread(fp, phdr[i].p_paddr, phdr[i].p_filesz) != 938 phdr[i].p_filesz) { 939 free(phdr); 940 return 1; 941 } 942 943 first = 0; 944 } 945 946 if ((IS_TEXT(phdr[i]) && (flags & (LOAD_TEXT | COUNT_TEXT))) || 947 (IS_DATA(phdr[i]) && (flags & (LOAD_DATA | COUNT_TEXT)))) { 948 pos = phdr[i].p_paddr; 949 if (minp > pos) 950 minp = pos; 951 pos += phdr[i].p_filesz; 952 if (maxp < pos) 953 maxp = pos; 954 } 955 956 /* Zero out BSS. */ 957 if (IS_BSS(phdr[i]) && (flags & LOAD_BSS)) { 958 mbzero((phdr[i].p_paddr + phdr[i].p_filesz), 959 phdr[i].p_memsz - phdr[i].p_filesz); 960 } 961 if (IS_BSS(phdr[i]) && (flags & (LOAD_BSS|COUNT_BSS))) { 962 pos += phdr[i].p_memsz - phdr[i].p_filesz; 963 if (maxp < pos) 964 maxp = pos; 965 } 966 } 967 free(phdr); 968 969 /* 970 * Copy the ELF and section headers. 971 */ 972 elfp = maxp = roundup(maxp, sizeof(Elf32_Addr)); 973 if (flags & (LOAD_HDR | COUNT_HDR)) 974 maxp += sizeof(Elf32_Ehdr); 975 976 if (flags & (LOAD_SYM | COUNT_SYM)) { 977 if (gzseek(fp, (off_t)elf->e_shoff, SEEK_SET) == -1) { 978 warn("lseek section headers"); 979 return 1; 980 } 981 sz = elf->e_shnum * sizeof(Elf32_Shdr); 982 shp = malloc(sz); 983 984 if ((size_t)gzread(fp, shp, sz) != sz) { 985 free(shp); 986 return 1; 987 } 988 989 shpp = maxp; 990 maxp += roundup(sz, sizeof(Elf32_Addr)); 991 992 size_t shstrsz = shp[elf->e_shstrndx].sh_size; 993 char *shstr = malloc(shstrsz); 994 if (gzseek(fp, (off_t)shp[elf->e_shstrndx].sh_offset, 995 SEEK_SET) == -1) { 996 free(shstr); 997 free(shp); 998 return 1; 999 } 1000 if ((size_t)gzread(fp, shstr, shstrsz) != shstrsz) { 1001 free(shstr); 1002 free(shp); 1003 return 1; 1004 } 1005 1006 /* 1007 * Now load the symbol sections themselves. Make sure the 1008 * sections are aligned. Don't bother with string tables if 1009 * there are no symbol sections. 1010 */ 1011 off = roundup((sizeof(Elf32_Ehdr) + sz), sizeof(Elf32_Addr)); 1012 1013 for (havesyms = i = 0; i < elf->e_shnum; i++) 1014 if (shp[i].sh_type == SHT_SYMTAB) 1015 havesyms = 1; 1016 1017 for (first = 1, i = 0; i < elf->e_shnum; i++) { 1018 if (shp[i].sh_type == SHT_SYMTAB || 1019 shp[i].sh_type == SHT_STRTAB || 1020 !strcmp(shstr + shp[i].sh_name, ".debug_line")) { 1021 if (havesyms && (flags & LOAD_SYM)) { 1022 if (gzseek(fp, (off_t)shp[i].sh_offset, 1023 SEEK_SET) == -1) { 1024 free(shstr); 1025 free(shp); 1026 return 1; 1027 } 1028 if (mread(fp, maxp, 1029 shp[i].sh_size) != shp[i].sh_size) { 1030 free(shstr); 1031 free(shp); 1032 return 1; 1033 } 1034 } 1035 maxp += roundup(shp[i].sh_size, 1036 sizeof(Elf32_Addr)); 1037 shp[i].sh_offset = off; 1038 shp[i].sh_flags |= SHF_ALLOC; 1039 off += roundup(shp[i].sh_size, 1040 sizeof(Elf32_Addr)); 1041 first = 0; 1042 } 1043 } 1044 if (flags & LOAD_SYM) { 1045 mbcopy(shp, shpp, sz); 1046 } 1047 free(shstr); 1048 free(shp); 1049 } 1050 1051 /* 1052 * Frob the copied ELF header to give information relative 1053 * to elfp. 1054 */ 1055 if (flags & LOAD_HDR) { 1056 elf->e_phoff = 0; 1057 elf->e_shoff = sizeof(Elf32_Ehdr); 1058 elf->e_phentsize = 0; 1059 elf->e_phnum = 0; 1060 mbcopy(elf, elfp, sizeof(*elf)); 1061 } 1062 1063 marks[MARK_START] = LOADADDR(minp); 1064 marks[MARK_ENTRY] = LOADADDR(elf->e_entry); 1065 marks[MARK_NSYM] = 1; /* XXX: Kernel needs >= 0 */ 1066 marks[MARK_SYM] = LOADADDR(elfp); 1067 marks[MARK_END] = LOADADDR(maxp); 1068 1069 return 0; 1070 } 1071