xref: /openbsd-src/usr.sbin/vmd/loadfile_elf.c (revision c1a45aed656e7d5627c30c92421893a76f370ccb)
1 /* $NetBSD: loadfile.c,v 1.10 2000/12/03 02:53:04 tsutsui Exp $ */
2 /* $OpenBSD: loadfile_elf.c,v 1.42 2022/01/28 06:33:27 guenther 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, bios_bootmac_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 its 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, unsigned int bootdevice)
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 	bios_bootmac_t bm, *bootmac = NULL;
275 
276 	if ((r = gzread(fp, &hdr, sizeof(hdr))) != sizeof(hdr))
277 		return 1;
278 
279 	memset(&marks, 0, sizeof(marks));
280 	if (memcmp(hdr.elf32.e_ident, ELFMAG, SELFMAG) == 0 &&
281 	    hdr.elf32.e_ident[EI_CLASS] == ELFCLASS32) {
282 		r = elf32_exec(fp, &hdr.elf32, marks, LOAD_ALL);
283 		is_i386 = 1;
284 	} else if (memcmp(hdr.elf64.e_ident, ELFMAG, SELFMAG) == 0 &&
285 	    hdr.elf64.e_ident[EI_CLASS] == ELFCLASS64) {
286 		r = elf64_exec(fp, &hdr.elf64, marks, LOAD_ALL);
287 	} else
288 		errno = ENOEXEC;
289 
290 	if (r)
291 		return (r);
292 
293 	push_gdt();
294 
295 	if (is_i386) {
296 		push_pt_32();
297 		/* Reconfigure the default flat-64 register set for 32 bit */
298 		vrs->vrs_crs[VCPU_REGS_CR3] = PML3_PAGE;
299 		vrs->vrs_crs[VCPU_REGS_CR4] = CR4_PSE;
300 		vrs->vrs_msrs[VCPU_REGS_EFER] = 0ULL;
301 	}
302 	else
303 		push_pt_64();
304 
305 	if (bootdevice == VMBOOTDEV_NET) {
306 		bootmac = &bm;
307 		memcpy(bootmac, vcp->vcp_macs[0], ETHER_ADDR_LEN);
308 	}
309 	n = create_bios_memmap(vcp, memmap);
310 	bootargsz = push_bootargs(memmap, n, bootmac);
311 	stacksize = push_stack(bootargsz, marks[MARK_END]);
312 
313 	vrs->vrs_gprs[VCPU_REGS_RIP] = (uint64_t)marks[MARK_ENTRY];
314 	vrs->vrs_gprs[VCPU_REGS_RSP] = (uint64_t)(STACK_PAGE + PAGE_SIZE) - stacksize;
315 	vrs->vrs_gdtr.vsi_base = GDT_PAGE;
316 
317 	log_debug("%s: loaded ELF kernel", __func__);
318 
319 	return (0);
320 }
321 
322 /*
323  * create_bios_memmap
324  *
325  * Construct a memory map as returned by the BIOS INT 0x15, e820 routine.
326  *
327  * Parameters:
328  *  vcp: the VM create parameters, containing the memory map passed to vmm(4)
329  *   memmap (out): the BIOS memory map
330  *
331  * Return values:
332  * Number of bios_memmap_t entries, including the terminating nul-entry.
333  */
334 static size_t
335 create_bios_memmap(struct vm_create_params *vcp, bios_memmap_t *memmap)
336 {
337 	size_t i, n = 0, sz;
338 	paddr_t gpa;
339 	struct vm_mem_range *vmr;
340 
341 	for (i = 0; i < vcp->vcp_nmemranges; i++) {
342 		vmr = &vcp->vcp_memranges[i];
343 		gpa = vmr->vmr_gpa;
344 		sz = vmr->vmr_size;
345 
346 		/*
347 		 * Make sure that we do not mark the ROM/video RAM area in the
348 		 * low memory as physcal memory available to the kernel.
349 		 */
350 		if (gpa < 0x100000 && gpa + sz > LOWMEM_KB * 1024) {
351 			if (gpa >= LOWMEM_KB * 1024)
352 				sz = 0;
353 			else
354 				sz = LOWMEM_KB * 1024 - gpa;
355 		}
356 
357 		if (sz != 0) {
358 			memmap[n].addr = gpa;
359 			memmap[n].size = sz;
360 			memmap[n].type = 0x1;	/* Type 1 : Normal memory */
361 			n++;
362 		}
363 	}
364 
365 	/* Null mem map entry to denote the end of the ranges */
366 	memmap[n].addr = 0x0;
367 	memmap[n].size = 0x0;
368 	memmap[n].type = 0x0;
369 	n++;
370 
371 	return (n);
372 }
373 
374 /*
375  * push_bootargs
376  *
377  * Creates the boot arguments page in the guest address space.
378  * Since vmd(8) is acting as the bootloader, we need to create the same boot
379  * arguments page that a real bootloader would have created. This is loaded
380  * into the guest phys RAM space at address BOOTARGS_PAGE.
381  *
382  * Parameters:
383  *  memmap: the BIOS memory map
384  *  n: number of entries in memmap
385  *
386  * Return values:
387  *  The size of the bootargs
388  */
389 static uint32_t
390 push_bootargs(bios_memmap_t *memmap, size_t n, bios_bootmac_t *bootmac)
391 {
392 	uint32_t memmap_sz, consdev_sz, bootmac_sz, i;
393 	bios_consdev_t consdev;
394 	uint32_t ba[1024];
395 
396 	memmap_sz = 3 * sizeof(int) + n * sizeof(bios_memmap_t);
397 	ba[0] = 0x0;    /* memory map */
398 	ba[1] = memmap_sz;
399 	ba[2] = memmap_sz;	/* next */
400 	memcpy(&ba[3], memmap, n * sizeof(bios_memmap_t));
401 	i = memmap_sz / sizeof(int);
402 
403 	/* Serial console device, COM1 @ 0x3f8 */
404 	consdev.consdev = makedev(8, 0);	/* com1 @ 0x3f8 */
405 	consdev.conspeed = 115200;
406 	consdev.consaddr = 0x3f8;
407 	consdev.consfreq = 0;
408 
409 	consdev_sz = 3 * sizeof(int) + sizeof(bios_consdev_t);
410 	ba[i] = 0x5;   /* consdev */
411 	ba[i + 1] = consdev_sz;
412 	ba[i + 2] = consdev_sz;
413 	memcpy(&ba[i + 3], &consdev, sizeof(bios_consdev_t));
414 	i += consdev_sz / sizeof(int);
415 
416 	if (bootmac) {
417 		bootmac_sz = 3 * sizeof(int) + (sizeof(bios_bootmac_t) + 3) & ~3;
418 		ba[i] = 0x7;   /* bootmac */
419 		ba[i + 1] = bootmac_sz;
420 		ba[i + 2] = bootmac_sz;
421 		memcpy(&ba[i + 3], bootmac, sizeof(bios_bootmac_t));
422 		i += bootmac_sz / sizeof(int);
423 	}
424 
425 	ba[i++] = 0xFFFFFFFF; /* BOOTARG_END */
426 
427 	write_mem(BOOTARGS_PAGE, ba, PAGE_SIZE);
428 
429 	return (i * sizeof(int));
430 }
431 
432 /*
433  * push_stack
434  *
435  * Creates the boot stack page in the guest address space. When using a real
436  * bootloader, the stack will be prepared using the following format before
437  * transitioning to kernel start, so vmd(8) needs to mimic the same stack
438  * layout. The stack content is pushed to the guest phys RAM at address
439  * STACK_PAGE. The bootloader operates in 32 bit mode; each stack entry is
440  * 4 bytes.
441  *
442  * Stack Layout: (TOS == Top Of Stack)
443  *  TOS		location of boot arguments page
444  *  TOS - 0x4	size of the content in the boot arguments page
445  *  TOS - 0x8	size of low memory (biosbasemem: kernel uses BIOS map only if 0)
446  *  TOS - 0xc	size of high memory (biosextmem, not used by kernel at all)
447  *  TOS - 0x10	kernel 'end' symbol value
448  *  TOS - 0x14	version of bootarg API
449  *
450  * Parameters:
451  *  bootargsz: size of boot arguments
452  *  end: kernel 'end' symbol value
453  *  bootdev: the optional non-default boot device
454  *  howto: optional boot flags for the kernel
455  *
456  * Return values:
457  *  size of the stack
458  */
459 static size_t
460 push_stack(uint32_t bootargsz, uint32_t end)
461 {
462 	uint32_t stack[1024];
463 	uint16_t loc;
464 
465 	memset(&stack, 0, sizeof(stack));
466 	loc = 1024;
467 
468 	stack[--loc] = BOOTARGS_PAGE;
469 	stack[--loc] = bootargsz;
470 	stack[--loc] = 0; /* biosbasemem */
471 	stack[--loc] = 0; /* biosextmem */
472 	stack[--loc] = end;
473 	stack[--loc] = 0x0e;
474 	stack[--loc] = MAKEBOOTDEV(0x4, 0, 0, 0, 0); /* bootdev: sd0a */
475 	stack[--loc] = 0;
476 
477 	write_mem(STACK_PAGE, &stack, PAGE_SIZE);
478 
479 	return (1024 - (loc - 1)) * sizeof(uint32_t);
480 }
481 
482 /*
483  * mread
484  *
485  * Reads 'sz' bytes from the file whose descriptor is provided in 'fd'
486  * into the guest address space at paddr 'addr'.
487  *
488  * Parameters:
489  *  fp: kernel image file to read from.
490  *  addr: guest paddr_t to load to
491  *  sz: number of bytes to load
492  *
493  * Return values:
494  *  returns 'sz' if successful, or 0 otherwise.
495  */
496 size_t
497 mread(gzFile fp, paddr_t addr, size_t sz)
498 {
499 	const char *errstr = NULL;
500 	int errnum = 0;
501 	size_t ct;
502 	size_t i, osz;
503 	char buf[PAGE_SIZE];
504 
505 	/*
506 	 * break up the 'sz' bytes into PAGE_SIZE chunks for use with
507 	 * write_mem
508 	 */
509 	ct = 0;
510 	osz = sz;
511 	if ((addr & PAGE_MASK) != 0) {
512 		memset(buf, 0, sizeof(buf));
513 		if (sz > PAGE_SIZE)
514 			ct = PAGE_SIZE - (addr & PAGE_MASK);
515 		else
516 			ct = sz;
517 
518 		if ((size_t)gzread(fp, buf, ct) != ct) {
519 			errstr = gzerror(fp, &errnum);
520 			if (errnum == Z_ERRNO)
521 				errnum = errno;
522 			log_warnx("%s: error %d in mread, %s", __progname,
523 			    errnum, errstr);
524 			return (0);
525 		}
526 
527 		if (write_mem(addr, buf, ct))
528 			return (0);
529 
530 		addr += ct;
531 	}
532 
533 	sz = sz - ct;
534 
535 	if (sz == 0)
536 		return (osz);
537 
538 	for (i = 0; i < sz; i += PAGE_SIZE, addr += PAGE_SIZE) {
539 		memset(buf, 0, sizeof(buf));
540 		if (i + PAGE_SIZE > sz)
541 			ct = sz - i;
542 		else
543 			ct = PAGE_SIZE;
544 
545 		if ((size_t)gzread(fp, buf, ct) != ct) {
546 			errstr = gzerror(fp, &errnum);
547 			if (errnum == Z_ERRNO)
548 				errnum = errno;
549 			log_warnx("%s: error %d in mread, %s", __progname,
550 			    errnum, errstr);
551 			return (0);
552 		}
553 
554 		if (write_mem(addr, buf, ct))
555 			return (0);
556 	}
557 
558 	return (osz);
559 }
560 
561 /*
562  * marc4random_buf
563  *
564  * load 'sz' bytes of random data into the guest address space at paddr
565  * 'addr'.
566  *
567  * Parameters:
568  *  addr: guest paddr_t to load random bytes into
569  *  sz: number of random bytes to load
570  *
571  * Return values:
572  *  nothing
573  */
574 static void
575 marc4random_buf(paddr_t addr, int sz)
576 {
577 	int i, ct;
578 	char buf[PAGE_SIZE];
579 
580 	/*
581 	 * break up the 'sz' bytes into PAGE_SIZE chunks for use with
582 	 * write_mem
583 	 */
584 	ct = 0;
585 	if (addr % PAGE_SIZE != 0) {
586 		memset(buf, 0, sizeof(buf));
587 		ct = PAGE_SIZE - (addr % PAGE_SIZE);
588 
589 		arc4random_buf(buf, ct);
590 
591 		if (write_mem(addr, buf, ct))
592 			return;
593 
594 		addr += ct;
595 	}
596 
597 	for (i = 0; i < sz; i+= PAGE_SIZE, addr += PAGE_SIZE) {
598 		memset(buf, 0, sizeof(buf));
599 		if (i + PAGE_SIZE > sz)
600 			ct = sz - i;
601 		else
602 			ct = PAGE_SIZE;
603 
604 		arc4random_buf(buf, ct);
605 
606 		if (write_mem(addr, buf, ct))
607 			return;
608 	}
609 }
610 
611 /*
612  * mbzero
613  *
614  * load 'sz' bytes of zeros into the guest address space at paddr
615  * 'addr'.
616  *
617  * Parameters:
618  *  addr: guest paddr_t to zero
619  *  sz: number of zero bytes to store
620  *
621  * Return values:
622  *  nothing
623  */
624 static void
625 mbzero(paddr_t addr, int sz)
626 {
627 	if (write_mem(addr, NULL, sz))
628 		return;
629 }
630 
631 /*
632  * mbcopy
633  *
634  * copies 'sz' bytes from buffer 'src' to guest paddr 'dst'.
635  *
636  * Parameters:
637  *  src: source buffer to copy from
638  *  dst: destination guest paddr_t to copy to
639  *  sz: number of bytes to copy
640  *
641  * Return values:
642  *  nothing
643  */
644 static void
645 mbcopy(void *src, paddr_t dst, int sz)
646 {
647 	write_mem(dst, src, sz);
648 }
649 
650 /*
651  * elf64_exec
652  *
653  * Load the kernel indicated by 'fp' into the guest physical memory
654  * space, at the addresses defined in the ELF header.
655  *
656  * This function is used for 64 bit kernels.
657  *
658  * Parameters:
659  *  fp: kernel image file to load
660  *  elf: ELF header of the kernel
661  *  marks: array to store the offsets of various kernel structures
662  *      (start, bss, etc)
663  *  flags: flag value to indicate which section(s) to load (usually
664  *      LOAD_ALL)
665  *
666  * Return values:
667  *  0 if successful
668  *  1 if unsuccessful
669  */
670 static int
671 elf64_exec(gzFile fp, Elf64_Ehdr *elf, u_long *marks, int flags)
672 {
673 	Elf64_Shdr *shp;
674 	Elf64_Phdr *phdr;
675 	Elf64_Off off;
676 	int i;
677 	size_t sz;
678 	int havesyms;
679 	paddr_t minp = ~0, maxp = 0, pos = 0;
680 	paddr_t offset = marks[MARK_START], shpp, elfp;
681 
682 	sz = elf->e_phnum * sizeof(Elf64_Phdr);
683 	phdr = malloc(sz);
684 
685 	if (gzseek(fp, (off_t)elf->e_phoff, SEEK_SET) == -1)  {
686 		free(phdr);
687 		return 1;
688 	}
689 
690 	if ((size_t)gzread(fp, phdr, sz) != sz) {
691 		free(phdr);
692 		return 1;
693 	}
694 
695 	for (i = 0; i < elf->e_phnum; i++) {
696 		if (phdr[i].p_type == PT_OPENBSD_RANDOMIZE) {
697 			int m;
698 
699 			/* Fill segment if asked for. */
700 			if (flags & LOAD_RANDOM) {
701 				for (pos = 0; pos < phdr[i].p_filesz;
702 				    pos += m) {
703 					m = phdr[i].p_filesz - pos;
704 					marc4random_buf(phdr[i].p_paddr + pos,
705 					    m);
706 				}
707 			}
708 			if (flags & (LOAD_RANDOM | COUNT_RANDOM)) {
709 				marks[MARK_RANDOM] = LOADADDR(phdr[i].p_paddr);
710 				marks[MARK_ERANDOM] =
711 				    marks[MARK_RANDOM] + phdr[i].p_filesz;
712 			}
713 			continue;
714 		}
715 
716 		if (phdr[i].p_type != PT_LOAD ||
717 		    (phdr[i].p_flags & (PF_W|PF_R|PF_X)) == 0)
718 			continue;
719 
720 #define IS_TEXT(p)	(p.p_flags & PF_X)
721 #define IS_DATA(p)	((p.p_flags & PF_X) == 0)
722 #define IS_BSS(p)	(p.p_filesz < p.p_memsz)
723 		/*
724 		 * XXX: Assume first address is lowest
725 		 */
726 		if ((IS_TEXT(phdr[i]) && (flags & LOAD_TEXT)) ||
727 		    (IS_DATA(phdr[i]) && (flags & LOAD_DATA))) {
728 
729 			/* Read in segment. */
730 			if (gzseek(fp, (off_t)phdr[i].p_offset,
731 			    SEEK_SET) == -1) {
732 				free(phdr);
733 				return 1;
734 			}
735 			if (mread(fp, phdr[i].p_paddr, phdr[i].p_filesz) !=
736 			    phdr[i].p_filesz) {
737 				free(phdr);
738 				return 1;
739 			}
740 		}
741 
742 		if ((IS_TEXT(phdr[i]) && (flags & (LOAD_TEXT | COUNT_TEXT))) ||
743 		    (IS_DATA(phdr[i]) && (flags & (LOAD_DATA | COUNT_TEXT)))) {
744 			pos = phdr[i].p_paddr;
745 			if (minp > pos)
746 				minp = pos;
747 			pos += phdr[i].p_filesz;
748 			if (maxp < pos)
749 				maxp = pos;
750 		}
751 
752 		/* Zero out BSS. */
753 		if (IS_BSS(phdr[i]) && (flags & LOAD_BSS)) {
754 			mbzero((phdr[i].p_paddr + phdr[i].p_filesz),
755 			    phdr[i].p_memsz - phdr[i].p_filesz);
756 		}
757 		if (IS_BSS(phdr[i]) && (flags & (LOAD_BSS|COUNT_BSS))) {
758 			pos += phdr[i].p_memsz - phdr[i].p_filesz;
759 			if (maxp < pos)
760 				maxp = pos;
761 		}
762 	}
763 	free(phdr);
764 
765 	/*
766 	 * Copy the ELF and section headers.
767 	 */
768 	elfp = maxp = roundup(maxp, sizeof(Elf64_Addr));
769 	if (flags & (LOAD_HDR | COUNT_HDR))
770 		maxp += sizeof(Elf64_Ehdr);
771 
772 	if (flags & (LOAD_SYM | COUNT_SYM)) {
773 		if (gzseek(fp, (off_t)elf->e_shoff, SEEK_SET) == -1) {
774 			warn("gzseek section headers");
775 			return 1;
776 		}
777 		sz = elf->e_shnum * sizeof(Elf64_Shdr);
778 		shp = malloc(sz);
779 
780 		if ((size_t)gzread(fp, shp, sz) != sz) {
781 			free(shp);
782 			return 1;
783 		}
784 
785 		shpp = maxp;
786 		maxp += roundup(sz, sizeof(Elf64_Addr));
787 
788 		size_t shstrsz = shp[elf->e_shstrndx].sh_size;
789 		char *shstr = malloc(shstrsz);
790 		if (gzseek(fp, (off_t)shp[elf->e_shstrndx].sh_offset,
791 		    SEEK_SET) == -1) {
792 			free(shstr);
793 			free(shp);
794 			return 1;
795 		}
796 		if ((size_t)gzread(fp, shstr, shstrsz) != shstrsz) {
797 			free(shstr);
798 			free(shp);
799 			return 1;
800 		}
801 
802 		/*
803 		 * Now load the symbol sections themselves. Make sure the
804 		 * sections are aligned. Don't bother with string tables if
805 		 * there are no symbol sections.
806 		 */
807 		off = roundup((sizeof(Elf64_Ehdr) + sz), sizeof(Elf64_Addr));
808 
809 		for (havesyms = i = 0; i < elf->e_shnum; i++)
810 			if (shp[i].sh_type == SHT_SYMTAB)
811 				havesyms = 1;
812 
813 		for (i = 0; i < elf->e_shnum; i++) {
814 			if (shp[i].sh_type == SHT_SYMTAB ||
815 			    shp[i].sh_type == SHT_STRTAB ||
816 			    !strcmp(shstr + shp[i].sh_name, ".debug_line") ||
817 			    !strcmp(shstr + shp[i].sh_name, ELF_CTF)) {
818 				if (havesyms && (flags & LOAD_SYM)) {
819 					if (gzseek(fp, (off_t)shp[i].sh_offset,
820 					    SEEK_SET) == -1) {
821 						free(shstr);
822 						free(shp);
823 						return 1;
824 					}
825 					if (mread(fp, maxp,
826 					    shp[i].sh_size) != shp[i].sh_size) {
827 						free(shstr);
828 						free(shp);
829 						return 1;
830 					}
831 				}
832 				maxp += roundup(shp[i].sh_size,
833 				    sizeof(Elf64_Addr));
834 				shp[i].sh_offset = off;
835 				shp[i].sh_flags |= SHF_ALLOC;
836 				off += roundup(shp[i].sh_size,
837 				    sizeof(Elf64_Addr));
838 			}
839 		}
840 		if (flags & LOAD_SYM) {
841 			mbcopy(shp, shpp, sz);
842 		}
843 		free(shstr);
844 		free(shp);
845 	}
846 
847 	/*
848 	 * Frob the copied ELF header to give information relative
849 	 * to elfp.
850 	 */
851 	if (flags & LOAD_HDR) {
852 		elf->e_phoff = 0;
853 		elf->e_shoff = sizeof(Elf64_Ehdr);
854 		elf->e_phentsize = 0;
855 		elf->e_phnum = 0;
856 		mbcopy(elf, elfp, sizeof(*elf));
857 	}
858 
859 	marks[MARK_START] = LOADADDR(minp);
860 	marks[MARK_ENTRY] = LOADADDR(elf->e_entry);
861 	marks[MARK_NSYM] = 1;	/* XXX: Kernel needs >= 0 */
862 	marks[MARK_SYM] = LOADADDR(elfp);
863 	marks[MARK_END] = LOADADDR(maxp);
864 
865 	return 0;
866 }
867 
868 /*
869  * elf32_exec
870  *
871  * Load the kernel indicated by 'fp' into the guest physical memory
872  * space, at the addresses defined in the ELF header.
873  *
874  * This function is used for 32 bit kernels.
875  *
876  * Parameters:
877  *  fp: kernel image file to load
878  *  elf: ELF header of the kernel
879  *  marks: array to store the offsets of various kernel structures
880  *      (start, bss, etc)
881  *  flags: flag value to indicate which section(s) to load (usually
882  *      LOAD_ALL)
883  *
884  * Return values:
885  *  0 if successful
886  *  1 if unsuccessful
887  */
888 static int
889 elf32_exec(gzFile fp, Elf32_Ehdr *elf, u_long *marks, int flags)
890 {
891 	Elf32_Shdr *shp;
892 	Elf32_Phdr *phdr;
893 	Elf32_Off off;
894 	int i;
895 	size_t sz;
896 	int havesyms;
897 	paddr_t minp = ~0, maxp = 0, pos = 0;
898 	paddr_t offset = marks[MARK_START], shpp, elfp;
899 
900 	sz = elf->e_phnum * sizeof(Elf32_Phdr);
901 	phdr = malloc(sz);
902 
903 	if (gzseek(fp, (off_t)elf->e_phoff, SEEK_SET) == -1)  {
904 		free(phdr);
905 		return 1;
906 	}
907 
908 	if ((size_t)gzread(fp, phdr, sz) != sz) {
909 		free(phdr);
910 		return 1;
911 	}
912 
913 	for (i = 0; i < elf->e_phnum; i++) {
914 		if (phdr[i].p_type == PT_OPENBSD_RANDOMIZE) {
915 			int m;
916 
917 			/* Fill segment if asked for. */
918 			if (flags & LOAD_RANDOM) {
919 				for (pos = 0; pos < phdr[i].p_filesz;
920 				    pos += m) {
921 					m = phdr[i].p_filesz - pos;
922 					marc4random_buf(phdr[i].p_paddr + pos,
923 					    m);
924 				}
925 			}
926 			if (flags & (LOAD_RANDOM | COUNT_RANDOM)) {
927 				marks[MARK_RANDOM] = LOADADDR(phdr[i].p_paddr);
928 				marks[MARK_ERANDOM] =
929 				    marks[MARK_RANDOM] + phdr[i].p_filesz;
930 			}
931 			continue;
932 		}
933 
934 		if (phdr[i].p_type != PT_LOAD ||
935 		    (phdr[i].p_flags & (PF_W|PF_R|PF_X)) == 0)
936 			continue;
937 
938 #define IS_TEXT(p)	(p.p_flags & PF_X)
939 #define IS_DATA(p)	((p.p_flags & PF_X) == 0)
940 #define IS_BSS(p)	(p.p_filesz < p.p_memsz)
941 		/*
942 		 * XXX: Assume first address is lowest
943 		 */
944 		if ((IS_TEXT(phdr[i]) && (flags & LOAD_TEXT)) ||
945 		    (IS_DATA(phdr[i]) && (flags & LOAD_DATA))) {
946 
947 			/* Read in segment. */
948 			if (gzseek(fp, (off_t)phdr[i].p_offset,
949 			    SEEK_SET) == -1) {
950 				free(phdr);
951 				return 1;
952 			}
953 			if (mread(fp, phdr[i].p_paddr, phdr[i].p_filesz) !=
954 			    phdr[i].p_filesz) {
955 				free(phdr);
956 				return 1;
957 			}
958 		}
959 
960 		if ((IS_TEXT(phdr[i]) && (flags & (LOAD_TEXT | COUNT_TEXT))) ||
961 		    (IS_DATA(phdr[i]) && (flags & (LOAD_DATA | COUNT_TEXT)))) {
962 			pos = phdr[i].p_paddr;
963 			if (minp > pos)
964 				minp = pos;
965 			pos += phdr[i].p_filesz;
966 			if (maxp < pos)
967 				maxp = pos;
968 		}
969 
970 		/* Zero out BSS. */
971 		if (IS_BSS(phdr[i]) && (flags & LOAD_BSS)) {
972 			mbzero((phdr[i].p_paddr + phdr[i].p_filesz),
973 			    phdr[i].p_memsz - phdr[i].p_filesz);
974 		}
975 		if (IS_BSS(phdr[i]) && (flags & (LOAD_BSS|COUNT_BSS))) {
976 			pos += phdr[i].p_memsz - phdr[i].p_filesz;
977 			if (maxp < pos)
978 				maxp = pos;
979 		}
980 	}
981 	free(phdr);
982 
983 	/*
984 	 * Copy the ELF and section headers.
985 	 */
986 	elfp = maxp = roundup(maxp, sizeof(Elf32_Addr));
987 	if (flags & (LOAD_HDR | COUNT_HDR))
988 		maxp += sizeof(Elf32_Ehdr);
989 
990 	if (flags & (LOAD_SYM | COUNT_SYM)) {
991 		if (gzseek(fp, (off_t)elf->e_shoff, SEEK_SET) == -1) {
992 			warn("lseek section headers");
993 			return 1;
994 		}
995 		sz = elf->e_shnum * sizeof(Elf32_Shdr);
996 		shp = malloc(sz);
997 
998 		if ((size_t)gzread(fp, shp, sz) != sz) {
999 			free(shp);
1000 			return 1;
1001 		}
1002 
1003 		shpp = maxp;
1004 		maxp += roundup(sz, sizeof(Elf32_Addr));
1005 
1006 		size_t shstrsz = shp[elf->e_shstrndx].sh_size;
1007 		char *shstr = malloc(shstrsz);
1008 		if (gzseek(fp, (off_t)shp[elf->e_shstrndx].sh_offset,
1009 		    SEEK_SET) == -1) {
1010 			free(shstr);
1011 			free(shp);
1012 			return 1;
1013 		}
1014 		if ((size_t)gzread(fp, shstr, shstrsz) != shstrsz) {
1015 			free(shstr);
1016 			free(shp);
1017 			return 1;
1018 		}
1019 
1020 		/*
1021 		 * Now load the symbol sections themselves. Make sure the
1022 		 * sections are aligned. Don't bother with string tables if
1023 		 * there are no symbol sections.
1024 		 */
1025 		off = roundup((sizeof(Elf32_Ehdr) + sz), sizeof(Elf32_Addr));
1026 
1027 		for (havesyms = i = 0; i < elf->e_shnum; i++)
1028 			if (shp[i].sh_type == SHT_SYMTAB)
1029 				havesyms = 1;
1030 
1031 		for (i = 0; i < elf->e_shnum; i++) {
1032 			if (shp[i].sh_type == SHT_SYMTAB ||
1033 			    shp[i].sh_type == SHT_STRTAB ||
1034 			    !strcmp(shstr + shp[i].sh_name, ".debug_line")) {
1035 				if (havesyms && (flags & LOAD_SYM)) {
1036 					if (gzseek(fp, (off_t)shp[i].sh_offset,
1037 					    SEEK_SET) == -1) {
1038 						free(shstr);
1039 						free(shp);
1040 						return 1;
1041 					}
1042 					if (mread(fp, maxp,
1043 					    shp[i].sh_size) != shp[i].sh_size) {
1044 						free(shstr);
1045 						free(shp);
1046 						return 1;
1047 					}
1048 				}
1049 				maxp += roundup(shp[i].sh_size,
1050 				    sizeof(Elf32_Addr));
1051 				shp[i].sh_offset = off;
1052 				shp[i].sh_flags |= SHF_ALLOC;
1053 				off += roundup(shp[i].sh_size,
1054 				    sizeof(Elf32_Addr));
1055 			}
1056 		}
1057 		if (flags & LOAD_SYM) {
1058 			mbcopy(shp, shpp, sz);
1059 		}
1060 		free(shstr);
1061 		free(shp);
1062 	}
1063 
1064 	/*
1065 	 * Frob the copied ELF header to give information relative
1066 	 * to elfp.
1067 	 */
1068 	if (flags & LOAD_HDR) {
1069 		elf->e_phoff = 0;
1070 		elf->e_shoff = sizeof(Elf32_Ehdr);
1071 		elf->e_phentsize = 0;
1072 		elf->e_phnum = 0;
1073 		mbcopy(elf, elfp, sizeof(*elf));
1074 	}
1075 
1076 	marks[MARK_START] = LOADADDR(minp);
1077 	marks[MARK_ENTRY] = LOADADDR(elf->e_entry);
1078 	marks[MARK_NSYM] = 1;	/* XXX: Kernel needs >= 0 */
1079 	marks[MARK_SYM] = LOADADDR(elfp);
1080 	marks[MARK_END] = LOADADDR(maxp);
1081 
1082 	return 0;
1083 }
1084