xref: /openbsd-src/usr.sbin/vmd/loadfile_elf.c (revision 8652dcf0a20078a0a381bc0c54aae4251c6be96f)
1 /* $NetBSD: loadfile.c,v 1.10 2000/12/03 02:53:04 tsutsui Exp $ */
2 /* $OpenBSD: loadfile_elf.c,v 1.43 2022/11/28 18:24:52 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, 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  *  bootmac: optional PXE boot MAC address
386  *
387  * Return values:
388  *  The size of the bootargs in bytes
389  */
390 static uint32_t
391 push_bootargs(bios_memmap_t *memmap, size_t n, bios_bootmac_t *bootmac)
392 {
393 	uint32_t memmap_sz, consdev_sz, bootmac_sz, i;
394 	bios_consdev_t consdev;
395 	uint32_t ba[1024];
396 
397 	memmap_sz = 3 * sizeof(uint32_t) + n * sizeof(bios_memmap_t);
398 	ba[0] = BOOTARG_MEMMAP;
399 	ba[1] = memmap_sz;
400 	ba[2] = memmap_sz;
401 	memcpy(&ba[3], memmap, n * sizeof(bios_memmap_t));
402 	i = memmap_sz / sizeof(uint32_t);
403 
404 	/* Serial console device, COM1 @ 0x3f8 */
405 	memset(&consdev, 0, sizeof(consdev));
406 	consdev.consdev = makedev(8, 0);
407 	consdev.conspeed = 115200;
408 	consdev.consaddr = 0x3f8;
409 
410 	consdev_sz = 3 * sizeof(uint32_t) + sizeof(bios_consdev_t);
411 	ba[i] = BOOTARG_CONSDEV;
412 	ba[i + 1] = consdev_sz;
413 	ba[i + 2] = consdev_sz;
414 	memcpy(&ba[i + 3], &consdev, sizeof(bios_consdev_t));
415 	i += consdev_sz / sizeof(uint32_t);
416 
417 	if (bootmac) {
418 		bootmac_sz = 3 * sizeof(uint32_t) +
419 		    (sizeof(bios_bootmac_t) + 3) & ~3;
420 		ba[i] = BOOTARG_BOOTMAC;
421 		ba[i + 1] = bootmac_sz;
422 		ba[i + 2] = bootmac_sz;
423 		memcpy(&ba[i + 3], bootmac, sizeof(bios_bootmac_t));
424 		i += bootmac_sz / sizeof(uint32_t);
425 	}
426 
427 	ba[i++] = 0xFFFFFFFF; /* BOOTARG_END */
428 
429 	write_mem(BOOTARGS_PAGE, ba, PAGE_SIZE);
430 
431 	return (i * sizeof(uint32_t));
432 }
433 
434 /*
435  * push_stack
436  *
437  * Creates the boot stack page in the guest address space. When using a real
438  * bootloader, the stack will be prepared using the following format before
439  * transitioning to kernel start, so vmd(8) needs to mimic the same stack
440  * layout. The stack content is pushed to the guest phys RAM at address
441  * STACK_PAGE. The bootloader operates in 32 bit mode; each stack entry is
442  * 4 bytes.
443  *
444  * Stack Layout: (TOS == Top Of Stack)
445  *  TOS		location of boot arguments page
446  *  TOS - 0x4	size of the content in the boot arguments page
447  *  TOS - 0x8	size of low memory (biosbasemem: kernel uses BIOS map only if 0)
448  *  TOS - 0xc	size of high memory (biosextmem, not used by kernel at all)
449  *  TOS - 0x10	kernel 'end' symbol value
450  *  TOS - 0x14	version of bootarg API
451  *
452  * Parameters:
453  *  bootargsz: size of boot arguments
454  *  end: kernel 'end' symbol value
455  *  bootdev: the optional non-default boot device
456  *  howto: optional boot flags for the kernel
457  *
458  * Return values:
459  *  size of the stack
460  */
461 static size_t
462 push_stack(uint32_t bootargsz, uint32_t end)
463 {
464 	uint32_t stack[1024];
465 	uint16_t loc;
466 
467 	memset(&stack, 0, sizeof(stack));
468 	loc = 1024;
469 
470 	stack[--loc] = BOOTARGS_PAGE;
471 	stack[--loc] = bootargsz;
472 	stack[--loc] = 0; /* biosbasemem */
473 	stack[--loc] = 0; /* biosextmem */
474 	stack[--loc] = end;
475 	stack[--loc] = 0x0e;
476 	stack[--loc] = MAKEBOOTDEV(0x4, 0, 0, 0, 0); /* bootdev: sd0a */
477 	stack[--loc] = 0;
478 
479 	write_mem(STACK_PAGE, &stack, PAGE_SIZE);
480 
481 	return (1024 - (loc - 1)) * sizeof(uint32_t);
482 }
483 
484 /*
485  * mread
486  *
487  * Reads 'sz' bytes from the file whose descriptor is provided in 'fd'
488  * into the guest address space at paddr 'addr'.
489  *
490  * Parameters:
491  *  fp: kernel image file to read from.
492  *  addr: guest paddr_t to load to
493  *  sz: number of bytes to load
494  *
495  * Return values:
496  *  returns 'sz' if successful, or 0 otherwise.
497  */
498 size_t
499 mread(gzFile fp, paddr_t addr, size_t sz)
500 {
501 	const char *errstr = NULL;
502 	int errnum = 0;
503 	size_t ct;
504 	size_t i, osz;
505 	char buf[PAGE_SIZE];
506 
507 	/*
508 	 * break up the 'sz' bytes into PAGE_SIZE chunks for use with
509 	 * write_mem
510 	 */
511 	ct = 0;
512 	osz = sz;
513 	if ((addr & PAGE_MASK) != 0) {
514 		memset(buf, 0, sizeof(buf));
515 		if (sz > PAGE_SIZE)
516 			ct = PAGE_SIZE - (addr & PAGE_MASK);
517 		else
518 			ct = sz;
519 
520 		if ((size_t)gzread(fp, buf, ct) != ct) {
521 			errstr = gzerror(fp, &errnum);
522 			if (errnum == Z_ERRNO)
523 				errnum = errno;
524 			log_warnx("%s: error %d in mread, %s", __progname,
525 			    errnum, errstr);
526 			return (0);
527 		}
528 
529 		if (write_mem(addr, buf, ct))
530 			return (0);
531 
532 		addr += ct;
533 	}
534 
535 	sz = sz - ct;
536 
537 	if (sz == 0)
538 		return (osz);
539 
540 	for (i = 0; i < sz; i += PAGE_SIZE, addr += PAGE_SIZE) {
541 		memset(buf, 0, sizeof(buf));
542 		if (i + PAGE_SIZE > sz)
543 			ct = sz - i;
544 		else
545 			ct = PAGE_SIZE;
546 
547 		if ((size_t)gzread(fp, buf, ct) != ct) {
548 			errstr = gzerror(fp, &errnum);
549 			if (errnum == Z_ERRNO)
550 				errnum = errno;
551 			log_warnx("%s: error %d in mread, %s", __progname,
552 			    errnum, errstr);
553 			return (0);
554 		}
555 
556 		if (write_mem(addr, buf, ct))
557 			return (0);
558 	}
559 
560 	return (osz);
561 }
562 
563 /*
564  * marc4random_buf
565  *
566  * load 'sz' bytes of random data into the guest address space at paddr
567  * 'addr'.
568  *
569  * Parameters:
570  *  addr: guest paddr_t to load random bytes into
571  *  sz: number of random bytes to load
572  *
573  * Return values:
574  *  nothing
575  */
576 static void
577 marc4random_buf(paddr_t addr, int sz)
578 {
579 	int i, ct;
580 	char buf[PAGE_SIZE];
581 
582 	/*
583 	 * break up the 'sz' bytes into PAGE_SIZE chunks for use with
584 	 * write_mem
585 	 */
586 	ct = 0;
587 	if (addr % PAGE_SIZE != 0) {
588 		memset(buf, 0, sizeof(buf));
589 		ct = PAGE_SIZE - (addr % PAGE_SIZE);
590 
591 		arc4random_buf(buf, ct);
592 
593 		if (write_mem(addr, buf, ct))
594 			return;
595 
596 		addr += ct;
597 	}
598 
599 	for (i = 0; i < sz; i+= PAGE_SIZE, addr += PAGE_SIZE) {
600 		memset(buf, 0, sizeof(buf));
601 		if (i + PAGE_SIZE > sz)
602 			ct = sz - i;
603 		else
604 			ct = PAGE_SIZE;
605 
606 		arc4random_buf(buf, ct);
607 
608 		if (write_mem(addr, buf, ct))
609 			return;
610 	}
611 }
612 
613 /*
614  * mbzero
615  *
616  * load 'sz' bytes of zeros into the guest address space at paddr
617  * 'addr'.
618  *
619  * Parameters:
620  *  addr: guest paddr_t to zero
621  *  sz: number of zero bytes to store
622  *
623  * Return values:
624  *  nothing
625  */
626 static void
627 mbzero(paddr_t addr, int sz)
628 {
629 	if (write_mem(addr, NULL, sz))
630 		return;
631 }
632 
633 /*
634  * mbcopy
635  *
636  * copies 'sz' bytes from buffer 'src' to guest paddr 'dst'.
637  *
638  * Parameters:
639  *  src: source buffer to copy from
640  *  dst: destination guest paddr_t to copy to
641  *  sz: number of bytes to copy
642  *
643  * Return values:
644  *  nothing
645  */
646 static void
647 mbcopy(void *src, paddr_t dst, int sz)
648 {
649 	write_mem(dst, src, sz);
650 }
651 
652 /*
653  * elf64_exec
654  *
655  * Load the kernel indicated by 'fp' into the guest physical memory
656  * space, at the addresses defined in the ELF header.
657  *
658  * This function is used for 64 bit kernels.
659  *
660  * Parameters:
661  *  fp: kernel image file to load
662  *  elf: ELF header of the kernel
663  *  marks: array to store the offsets of various kernel structures
664  *      (start, bss, etc)
665  *  flags: flag value to indicate which section(s) to load (usually
666  *      LOAD_ALL)
667  *
668  * Return values:
669  *  0 if successful
670  *  1 if unsuccessful
671  */
672 static int
673 elf64_exec(gzFile fp, Elf64_Ehdr *elf, u_long *marks, int flags)
674 {
675 	Elf64_Shdr *shp;
676 	Elf64_Phdr *phdr;
677 	Elf64_Off off;
678 	int i;
679 	size_t sz;
680 	int havesyms;
681 	paddr_t minp = ~0, maxp = 0, pos = 0;
682 	paddr_t offset = marks[MARK_START], shpp, elfp;
683 
684 	sz = elf->e_phnum * sizeof(Elf64_Phdr);
685 	phdr = malloc(sz);
686 
687 	if (gzseek(fp, (off_t)elf->e_phoff, SEEK_SET) == -1)  {
688 		free(phdr);
689 		return 1;
690 	}
691 
692 	if ((size_t)gzread(fp, phdr, sz) != sz) {
693 		free(phdr);
694 		return 1;
695 	}
696 
697 	for (i = 0; i < elf->e_phnum; i++) {
698 		if (phdr[i].p_type == PT_OPENBSD_RANDOMIZE) {
699 			int m;
700 
701 			/* Fill segment if asked for. */
702 			if (flags & LOAD_RANDOM) {
703 				for (pos = 0; pos < phdr[i].p_filesz;
704 				    pos += m) {
705 					m = phdr[i].p_filesz - pos;
706 					marc4random_buf(phdr[i].p_paddr + pos,
707 					    m);
708 				}
709 			}
710 			if (flags & (LOAD_RANDOM | COUNT_RANDOM)) {
711 				marks[MARK_RANDOM] = LOADADDR(phdr[i].p_paddr);
712 				marks[MARK_ERANDOM] =
713 				    marks[MARK_RANDOM] + phdr[i].p_filesz;
714 			}
715 			continue;
716 		}
717 
718 		if (phdr[i].p_type != PT_LOAD ||
719 		    (phdr[i].p_flags & (PF_W|PF_R|PF_X)) == 0)
720 			continue;
721 
722 #define IS_TEXT(p)	(p.p_flags & PF_X)
723 #define IS_DATA(p)	((p.p_flags & PF_X) == 0)
724 #define IS_BSS(p)	(p.p_filesz < p.p_memsz)
725 		/*
726 		 * XXX: Assume first address is lowest
727 		 */
728 		if ((IS_TEXT(phdr[i]) && (flags & LOAD_TEXT)) ||
729 		    (IS_DATA(phdr[i]) && (flags & LOAD_DATA))) {
730 
731 			/* Read in segment. */
732 			if (gzseek(fp, (off_t)phdr[i].p_offset,
733 			    SEEK_SET) == -1) {
734 				free(phdr);
735 				return 1;
736 			}
737 			if (mread(fp, phdr[i].p_paddr, phdr[i].p_filesz) !=
738 			    phdr[i].p_filesz) {
739 				free(phdr);
740 				return 1;
741 			}
742 		}
743 
744 		if ((IS_TEXT(phdr[i]) && (flags & (LOAD_TEXT | COUNT_TEXT))) ||
745 		    (IS_DATA(phdr[i]) && (flags & (LOAD_DATA | COUNT_TEXT)))) {
746 			pos = phdr[i].p_paddr;
747 			if (minp > pos)
748 				minp = pos;
749 			pos += phdr[i].p_filesz;
750 			if (maxp < pos)
751 				maxp = pos;
752 		}
753 
754 		/* Zero out BSS. */
755 		if (IS_BSS(phdr[i]) && (flags & LOAD_BSS)) {
756 			mbzero((phdr[i].p_paddr + phdr[i].p_filesz),
757 			    phdr[i].p_memsz - phdr[i].p_filesz);
758 		}
759 		if (IS_BSS(phdr[i]) && (flags & (LOAD_BSS|COUNT_BSS))) {
760 			pos += phdr[i].p_memsz - phdr[i].p_filesz;
761 			if (maxp < pos)
762 				maxp = pos;
763 		}
764 	}
765 	free(phdr);
766 
767 	/*
768 	 * Copy the ELF and section headers.
769 	 */
770 	elfp = maxp = roundup(maxp, sizeof(Elf64_Addr));
771 	if (flags & (LOAD_HDR | COUNT_HDR))
772 		maxp += sizeof(Elf64_Ehdr);
773 
774 	if (flags & (LOAD_SYM | COUNT_SYM)) {
775 		if (gzseek(fp, (off_t)elf->e_shoff, SEEK_SET) == -1) {
776 			warn("gzseek section headers");
777 			return 1;
778 		}
779 		sz = elf->e_shnum * sizeof(Elf64_Shdr);
780 		shp = malloc(sz);
781 
782 		if ((size_t)gzread(fp, shp, sz) != sz) {
783 			free(shp);
784 			return 1;
785 		}
786 
787 		shpp = maxp;
788 		maxp += roundup(sz, sizeof(Elf64_Addr));
789 
790 		size_t shstrsz = shp[elf->e_shstrndx].sh_size;
791 		char *shstr = malloc(shstrsz);
792 		if (gzseek(fp, (off_t)shp[elf->e_shstrndx].sh_offset,
793 		    SEEK_SET) == -1) {
794 			free(shstr);
795 			free(shp);
796 			return 1;
797 		}
798 		if ((size_t)gzread(fp, shstr, shstrsz) != shstrsz) {
799 			free(shstr);
800 			free(shp);
801 			return 1;
802 		}
803 
804 		/*
805 		 * Now load the symbol sections themselves. Make sure the
806 		 * sections are aligned. Don't bother with string tables if
807 		 * there are no symbol sections.
808 		 */
809 		off = roundup((sizeof(Elf64_Ehdr) + sz), sizeof(Elf64_Addr));
810 
811 		for (havesyms = i = 0; i < elf->e_shnum; i++)
812 			if (shp[i].sh_type == SHT_SYMTAB)
813 				havesyms = 1;
814 
815 		for (i = 0; i < elf->e_shnum; i++) {
816 			if (shp[i].sh_type == SHT_SYMTAB ||
817 			    shp[i].sh_type == SHT_STRTAB ||
818 			    !strcmp(shstr + shp[i].sh_name, ".debug_line") ||
819 			    !strcmp(shstr + shp[i].sh_name, ELF_CTF)) {
820 				if (havesyms && (flags & LOAD_SYM)) {
821 					if (gzseek(fp, (off_t)shp[i].sh_offset,
822 					    SEEK_SET) == -1) {
823 						free(shstr);
824 						free(shp);
825 						return 1;
826 					}
827 					if (mread(fp, maxp,
828 					    shp[i].sh_size) != shp[i].sh_size) {
829 						free(shstr);
830 						free(shp);
831 						return 1;
832 					}
833 				}
834 				maxp += roundup(shp[i].sh_size,
835 				    sizeof(Elf64_Addr));
836 				shp[i].sh_offset = off;
837 				shp[i].sh_flags |= SHF_ALLOC;
838 				off += roundup(shp[i].sh_size,
839 				    sizeof(Elf64_Addr));
840 			}
841 		}
842 		if (flags & LOAD_SYM) {
843 			mbcopy(shp, shpp, sz);
844 		}
845 		free(shstr);
846 		free(shp);
847 	}
848 
849 	/*
850 	 * Frob the copied ELF header to give information relative
851 	 * to elfp.
852 	 */
853 	if (flags & LOAD_HDR) {
854 		elf->e_phoff = 0;
855 		elf->e_shoff = sizeof(Elf64_Ehdr);
856 		elf->e_phentsize = 0;
857 		elf->e_phnum = 0;
858 		mbcopy(elf, elfp, sizeof(*elf));
859 	}
860 
861 	marks[MARK_START] = LOADADDR(minp);
862 	marks[MARK_ENTRY] = LOADADDR(elf->e_entry);
863 	marks[MARK_NSYM] = 1;	/* XXX: Kernel needs >= 0 */
864 	marks[MARK_SYM] = LOADADDR(elfp);
865 	marks[MARK_END] = LOADADDR(maxp);
866 
867 	return 0;
868 }
869 
870 /*
871  * elf32_exec
872  *
873  * Load the kernel indicated by 'fp' into the guest physical memory
874  * space, at the addresses defined in the ELF header.
875  *
876  * This function is used for 32 bit kernels.
877  *
878  * Parameters:
879  *  fp: kernel image file to load
880  *  elf: ELF header of the kernel
881  *  marks: array to store the offsets of various kernel structures
882  *      (start, bss, etc)
883  *  flags: flag value to indicate which section(s) to load (usually
884  *      LOAD_ALL)
885  *
886  * Return values:
887  *  0 if successful
888  *  1 if unsuccessful
889  */
890 static int
891 elf32_exec(gzFile fp, Elf32_Ehdr *elf, u_long *marks, int flags)
892 {
893 	Elf32_Shdr *shp;
894 	Elf32_Phdr *phdr;
895 	Elf32_Off off;
896 	int i;
897 	size_t sz;
898 	int havesyms;
899 	paddr_t minp = ~0, maxp = 0, pos = 0;
900 	paddr_t offset = marks[MARK_START], shpp, elfp;
901 
902 	sz = elf->e_phnum * sizeof(Elf32_Phdr);
903 	phdr = malloc(sz);
904 
905 	if (gzseek(fp, (off_t)elf->e_phoff, SEEK_SET) == -1)  {
906 		free(phdr);
907 		return 1;
908 	}
909 
910 	if ((size_t)gzread(fp, phdr, sz) != sz) {
911 		free(phdr);
912 		return 1;
913 	}
914 
915 	for (i = 0; i < elf->e_phnum; i++) {
916 		if (phdr[i].p_type == PT_OPENBSD_RANDOMIZE) {
917 			int m;
918 
919 			/* Fill segment if asked for. */
920 			if (flags & LOAD_RANDOM) {
921 				for (pos = 0; pos < phdr[i].p_filesz;
922 				    pos += m) {
923 					m = phdr[i].p_filesz - pos;
924 					marc4random_buf(phdr[i].p_paddr + pos,
925 					    m);
926 				}
927 			}
928 			if (flags & (LOAD_RANDOM | COUNT_RANDOM)) {
929 				marks[MARK_RANDOM] = LOADADDR(phdr[i].p_paddr);
930 				marks[MARK_ERANDOM] =
931 				    marks[MARK_RANDOM] + phdr[i].p_filesz;
932 			}
933 			continue;
934 		}
935 
936 		if (phdr[i].p_type != PT_LOAD ||
937 		    (phdr[i].p_flags & (PF_W|PF_R|PF_X)) == 0)
938 			continue;
939 
940 #define IS_TEXT(p)	(p.p_flags & PF_X)
941 #define IS_DATA(p)	((p.p_flags & PF_X) == 0)
942 #define IS_BSS(p)	(p.p_filesz < p.p_memsz)
943 		/*
944 		 * XXX: Assume first address is lowest
945 		 */
946 		if ((IS_TEXT(phdr[i]) && (flags & LOAD_TEXT)) ||
947 		    (IS_DATA(phdr[i]) && (flags & LOAD_DATA))) {
948 
949 			/* Read in segment. */
950 			if (gzseek(fp, (off_t)phdr[i].p_offset,
951 			    SEEK_SET) == -1) {
952 				free(phdr);
953 				return 1;
954 			}
955 			if (mread(fp, phdr[i].p_paddr, phdr[i].p_filesz) !=
956 			    phdr[i].p_filesz) {
957 				free(phdr);
958 				return 1;
959 			}
960 		}
961 
962 		if ((IS_TEXT(phdr[i]) && (flags & (LOAD_TEXT | COUNT_TEXT))) ||
963 		    (IS_DATA(phdr[i]) && (flags & (LOAD_DATA | COUNT_TEXT)))) {
964 			pos = phdr[i].p_paddr;
965 			if (minp > pos)
966 				minp = pos;
967 			pos += phdr[i].p_filesz;
968 			if (maxp < pos)
969 				maxp = pos;
970 		}
971 
972 		/* Zero out BSS. */
973 		if (IS_BSS(phdr[i]) && (flags & LOAD_BSS)) {
974 			mbzero((phdr[i].p_paddr + phdr[i].p_filesz),
975 			    phdr[i].p_memsz - phdr[i].p_filesz);
976 		}
977 		if (IS_BSS(phdr[i]) && (flags & (LOAD_BSS|COUNT_BSS))) {
978 			pos += phdr[i].p_memsz - phdr[i].p_filesz;
979 			if (maxp < pos)
980 				maxp = pos;
981 		}
982 	}
983 	free(phdr);
984 
985 	/*
986 	 * Copy the ELF and section headers.
987 	 */
988 	elfp = maxp = roundup(maxp, sizeof(Elf32_Addr));
989 	if (flags & (LOAD_HDR | COUNT_HDR))
990 		maxp += sizeof(Elf32_Ehdr);
991 
992 	if (flags & (LOAD_SYM | COUNT_SYM)) {
993 		if (gzseek(fp, (off_t)elf->e_shoff, SEEK_SET) == -1) {
994 			warn("lseek section headers");
995 			return 1;
996 		}
997 		sz = elf->e_shnum * sizeof(Elf32_Shdr);
998 		shp = malloc(sz);
999 
1000 		if ((size_t)gzread(fp, shp, sz) != sz) {
1001 			free(shp);
1002 			return 1;
1003 		}
1004 
1005 		shpp = maxp;
1006 		maxp += roundup(sz, sizeof(Elf32_Addr));
1007 
1008 		size_t shstrsz = shp[elf->e_shstrndx].sh_size;
1009 		char *shstr = malloc(shstrsz);
1010 		if (gzseek(fp, (off_t)shp[elf->e_shstrndx].sh_offset,
1011 		    SEEK_SET) == -1) {
1012 			free(shstr);
1013 			free(shp);
1014 			return 1;
1015 		}
1016 		if ((size_t)gzread(fp, shstr, shstrsz) != shstrsz) {
1017 			free(shstr);
1018 			free(shp);
1019 			return 1;
1020 		}
1021 
1022 		/*
1023 		 * Now load the symbol sections themselves. Make sure the
1024 		 * sections are aligned. Don't bother with string tables if
1025 		 * there are no symbol sections.
1026 		 */
1027 		off = roundup((sizeof(Elf32_Ehdr) + sz), sizeof(Elf32_Addr));
1028 
1029 		for (havesyms = i = 0; i < elf->e_shnum; i++)
1030 			if (shp[i].sh_type == SHT_SYMTAB)
1031 				havesyms = 1;
1032 
1033 		for (i = 0; i < elf->e_shnum; i++) {
1034 			if (shp[i].sh_type == SHT_SYMTAB ||
1035 			    shp[i].sh_type == SHT_STRTAB ||
1036 			    !strcmp(shstr + shp[i].sh_name, ".debug_line")) {
1037 				if (havesyms && (flags & LOAD_SYM)) {
1038 					if (gzseek(fp, (off_t)shp[i].sh_offset,
1039 					    SEEK_SET) == -1) {
1040 						free(shstr);
1041 						free(shp);
1042 						return 1;
1043 					}
1044 					if (mread(fp, maxp,
1045 					    shp[i].sh_size) != shp[i].sh_size) {
1046 						free(shstr);
1047 						free(shp);
1048 						return 1;
1049 					}
1050 				}
1051 				maxp += roundup(shp[i].sh_size,
1052 				    sizeof(Elf32_Addr));
1053 				shp[i].sh_offset = off;
1054 				shp[i].sh_flags |= SHF_ALLOC;
1055 				off += roundup(shp[i].sh_size,
1056 				    sizeof(Elf32_Addr));
1057 			}
1058 		}
1059 		if (flags & LOAD_SYM) {
1060 			mbcopy(shp, shpp, sz);
1061 		}
1062 		free(shstr);
1063 		free(shp);
1064 	}
1065 
1066 	/*
1067 	 * Frob the copied ELF header to give information relative
1068 	 * to elfp.
1069 	 */
1070 	if (flags & LOAD_HDR) {
1071 		elf->e_phoff = 0;
1072 		elf->e_shoff = sizeof(Elf32_Ehdr);
1073 		elf->e_phentsize = 0;
1074 		elf->e_phnum = 0;
1075 		mbcopy(elf, elfp, sizeof(*elf));
1076 	}
1077 
1078 	marks[MARK_START] = LOADADDR(minp);
1079 	marks[MARK_ENTRY] = LOADADDR(elf->e_entry);
1080 	marks[MARK_NSYM] = 1;	/* XXX: Kernel needs >= 0 */
1081 	marks[MARK_SYM] = LOADADDR(elfp);
1082 	marks[MARK_END] = LOADADDR(maxp);
1083 
1084 	return 0;
1085 }
1086