xref: /freebsd/stand/common/load_elf.c (revision 681ce946)

/*-
 * Copyright (c) 1998 Michael Smith <msmith@freebsd.org>
 * Copyright (c) 1998 Peter Wemm <peter@freebsd.org>
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 */

#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");

#include <sys/param.h>
#include <sys/endian.h>
#include <sys/exec.h>
#include <sys/linker.h>
#include <sys/module.h>
#include <sys/stdint.h>
#include <string.h>
#include <machine/elf.h>
#include <stand.h>
#define FREEBSD_ELF
#include <sys/link_elf.h>

#include "bootstrap.h"

#define COPYOUT(s,d,l)	archsw.arch_copyout((vm_offset_t)(s), d, l)

#if defined(__i386__) && __ELF_WORD_SIZE == 64
#undef ELF_TARG_CLASS
#undef ELF_TARG_MACH
#define ELF_TARG_CLASS  ELFCLASS64
#define ELF_TARG_MACH   EM_X86_64
#endif

typedef struct elf_file {
	Elf_Phdr	*ph;
	Elf_Ehdr	*ehdr;
	Elf_Sym		*symtab;
	Elf_Hashelt	*hashtab;
	Elf_Hashelt	nbuckets;
	Elf_Hashelt	nchains;
	Elf_Hashelt	*buckets;
	Elf_Hashelt	*chains;
	Elf_Rel	*rel;
	size_t	relsz;
	Elf_Rela	*rela;
	size_t	relasz;
	char	*strtab;
	size_t	strsz;
	int		fd;
	caddr_t	firstpage;
	size_t	firstlen;
	int		kernel;
	uint64_t	off;
#ifdef LOADER_VERIEXEC_VECTX
	struct vectx	*vctx;
#endif
} *elf_file_t;

#ifdef LOADER_VERIEXEC_VECTX
#define VECTX_HANDLE(ef) (ef)->vctx
#else
#define VECTX_HANDLE(ef) (ef)->fd
#endif

static int __elfN(loadimage)(struct preloaded_file *mp, elf_file_t ef,
    uint64_t loadaddr);
static int __elfN(lookup_symbol)(elf_file_t ef, const char* name,
    Elf_Sym *sym, unsigned char type);
static int __elfN(reloc_ptr)(struct preloaded_file *mp, elf_file_t ef,
    Elf_Addr p, void *val, size_t len);
static int __elfN(parse_modmetadata)(struct preloaded_file *mp, elf_file_t ef,
    Elf_Addr p_start, Elf_Addr p_end);
static symaddr_fn __elfN(symaddr);
static char	*fake_modname(const char *name);

const char	*__elfN(kerneltype) = "elf kernel";
const char	*__elfN(moduletype) = "elf module";

uint64_t	__elfN(relocation_offset) = 0;

extern void elf_wrong_field_size(void);
#define CONVERT_FIELD(b, f, e)			\
	switch (sizeof((b)->f)) {		\
	case 2:					\
		(b)->f = e ## 16toh((b)->f);	\
		break;				\
	case 4:					\
		(b)->f = e ## 32toh((b)->f);	\
		break;				\
	case 8:					\
		(b)->f = e ## 64toh((b)->f);	\
		break;				\
	default:				\
		/* Force a link time error. */	\
		elf_wrong_field_size();		\
		break;				\
	}

#define CONVERT_SWITCH(h, d, f)			\
	switch ((h)->e_ident[EI_DATA]) {	\
	case ELFDATA2MSB:			\
		f(d, be);			\
		break;				\
	case ELFDATA2LSB:			\
		f(d, le);			\
		break;				\
	default:				\
		return (EINVAL);		\
	}


static int elf_header_convert(Elf_Ehdr *ehdr)
{
	/*
	 * Fixup ELF header endianness.
	 *
	 * The Xhdr structure was loaded using block read call to optimize file
	 * accesses. It might happen, that the endianness of the system memory
	 * is different that endianness of the ELF header.  Swap fields here to
	 * guarantee that Xhdr always contain valid data regardless of
	 * architecture.
	 */
#define HEADER_FIELDS(b, e)			\
	CONVERT_FIELD(b, e_type, e);		\
	CONVERT_FIELD(b, e_machine, e);		\
	CONVERT_FIELD(b, e_version, e);		\
	CONVERT_FIELD(b, e_entry, e);		\
	CONVERT_FIELD(b, e_phoff, e);		\
	CONVERT_FIELD(b, e_shoff, e);		\
	CONVERT_FIELD(b, e_flags, e);		\
	CONVERT_FIELD(b, e_ehsize, e);		\
	CONVERT_FIELD(b, e_phentsize, e);	\
	CONVERT_FIELD(b, e_phnum, e);		\
	CONVERT_FIELD(b, e_shentsize, e);	\
	CONVERT_FIELD(b, e_shnum, e);		\
	CONVERT_FIELD(b, e_shstrndx, e)

	CONVERT_SWITCH(ehdr, ehdr, HEADER_FIELDS);

#undef HEADER_FIELDS

	return (0);
}

static int elf_program_header_convert(const Elf_Ehdr *ehdr, Elf_Phdr *phdr)
{
#define PROGRAM_HEADER_FIELDS(b, e)		\
	CONVERT_FIELD(b, p_type, e);		\
	CONVERT_FIELD(b, p_flags, e);		\
	CONVERT_FIELD(b, p_offset, e);		\
	CONVERT_FIELD(b, p_vaddr, e);		\
	CONVERT_FIELD(b, p_paddr, e);		\
	CONVERT_FIELD(b, p_filesz, e);		\
	CONVERT_FIELD(b, p_memsz, e);		\
	CONVERT_FIELD(b, p_align, e)

	CONVERT_SWITCH(ehdr, phdr, PROGRAM_HEADER_FIELDS);

#undef PROGRAM_HEADER_FIELDS

	return (0);
}

static int elf_section_header_convert(const Elf_Ehdr *ehdr, Elf_Shdr *shdr)
{
#define SECTION_HEADER_FIELDS(b, e)		\
	CONVERT_FIELD(b, sh_name, e);		\
	CONVERT_FIELD(b, sh_type, e);		\
	CONVERT_FIELD(b, sh_link, e);		\
	CONVERT_FIELD(b, sh_info, e);		\
	CONVERT_FIELD(b, sh_flags, e);		\
	CONVERT_FIELD(b, sh_addr, e);		\
	CONVERT_FIELD(b, sh_offset, e);		\
	CONVERT_FIELD(b, sh_size, e);		\
	CONVERT_FIELD(b, sh_addralign, e);	\
	CONVERT_FIELD(b, sh_entsize, e)

	CONVERT_SWITCH(ehdr, shdr, SECTION_HEADER_FIELDS);

#undef SECTION_HEADER_FIELDS

	return (0);
}
#undef CONVERT_SWITCH
#undef CONVERT_FIELD


#ifdef __amd64__
static bool
is_kernphys_relocatable(elf_file_t ef)
{
	Elf_Sym sym;

	return (__elfN(lookup_symbol)(ef, "kernphys", &sym, STT_OBJECT) == 0 &&
	    sym.st_size == 8);
}
#endif

#ifdef __i386__
static bool
is_tg_kernel_support(struct preloaded_file *fp, elf_file_t ef)
{
	Elf_Sym		sym;
	Elf_Addr	p_start, p_end, v, p;
	char		vd_name[16];
	int		error;

	if (__elfN(lookup_symbol)(ef, "__start_set_vt_drv_set", &sym, STT_NOTYPE) != 0)
		return (false);
	p_start = sym.st_value + ef->off;
	if (__elfN(lookup_symbol)(ef, "__stop_set_vt_drv_set", &sym, STT_NOTYPE) != 0)
		return (false);
	p_end = sym.st_value + ef->off;

	/*
	 * Walk through vt_drv_set, each vt driver structure starts with
	 * static 16 chars for driver name. If we have "vbefb", return true.
	 */
	for (p = p_start; p < p_end; p += sizeof(Elf_Addr)) {
		COPYOUT(p, &v, sizeof(v));

		error = __elfN(reloc_ptr)(fp, ef, p, &v, sizeof(v));
		if (error == EOPNOTSUPP)
			v += ef->off;
		else if (error != 0)
			return (false);
		COPYOUT(v, &vd_name, sizeof(vd_name));
		if (strncmp(vd_name, "vbefb", sizeof(vd_name)) == 0)
			return (true);
	}

	return (false);
}
#endif

static int
__elfN(load_elf_header)(char *filename, elf_file_t ef)
{
	ssize_t			 bytes_read;
	Elf_Ehdr		*ehdr;
	int			 err;

	/*
	 * Open the image, read and validate the ELF header
	 */
	if (filename == NULL)	/* can't handle nameless */
		return (EFTYPE);
	if ((ef->fd = open(filename, O_RDONLY)) == -1)
		return (errno);
	ef->firstpage = malloc(PAGE_SIZE);
	if (ef->firstpage == NULL) {
		close(ef->fd);
		return (ENOMEM);
	}
	preload(ef->fd);
#ifdef LOADER_VERIEXEC_VECTX
	{
		int verror;

		ef->vctx = vectx_open(ef->fd, filename, 0L, NULL, &verror, __func__);
		if (verror) {
			printf("Unverified %s: %s\n", filename, ve_error_get());
			close(ef->fd);
			free(ef->vctx);
			return (EAUTH);
		}
	}
#endif
	bytes_read = VECTX_READ(VECTX_HANDLE(ef), ef->firstpage, PAGE_SIZE);
	ef->firstlen = (size_t)bytes_read;
	if (bytes_read < 0 || ef->firstlen <= sizeof(Elf_Ehdr)) {
		err = EFTYPE; /* could be EIO, but may be small file */
		goto error;
	}
	ehdr = ef->ehdr = (Elf_Ehdr *)ef->firstpage;

	/* Is it ELF? */
	if (!IS_ELF(*ehdr)) {
		err = EFTYPE;
		goto error;
	}

	if (ehdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || /* Layout ? */
	    ehdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
	    ehdr->e_ident[EI_VERSION] != EV_CURRENT) /* Version ? */ {
		err = EFTYPE;
		goto error;
	}

	err = elf_header_convert(ehdr);
	if (err)
		goto error;

	if (ehdr->e_version != EV_CURRENT || ehdr->e_machine != ELF_TARG_MACH) {
		/* Machine ? */
		err = EFTYPE;
		goto error;
	}

#if defined(LOADER_VERIEXEC) && !defined(LOADER_VERIEXEC_VECTX)
	if (verify_file(ef->fd, filename, bytes_read, VE_MUST, __func__) < 0) {
		err = EAUTH;
		goto error;
	}
#endif
	return (0);

error:
	if (ef->firstpage != NULL) {
		free(ef->firstpage);
		ef->firstpage = NULL;
	}
	if (ef->fd != -1) {
#ifdef LOADER_VERIEXEC_VECTX
		free(ef->vctx);
#endif
		close(ef->fd);
		ef->fd = -1;
	}
	return (err);
}

/*
 * Attempt to load the file (file) as an ELF module.  It will be stored at
 * (dest), and a pointer to a module structure describing the loaded object
 * will be saved in (result).
 */
int
__elfN(loadfile)(char *filename, uint64_t dest, struct preloaded_file **result)
{
	return (__elfN(loadfile_raw)(filename, dest, result, 0));
}

int
__elfN(loadfile_raw)(char *filename, uint64_t dest,
    struct preloaded_file **result, int multiboot)
{
	struct preloaded_file	*fp, *kfp;
	struct elf_file		ef;
	Elf_Ehdr		*ehdr;
	int			err;

	fp = NULL;
	bzero(&ef, sizeof(struct elf_file));
	ef.fd = -1;

	err = __elfN(load_elf_header)(filename, &ef);
	if (err != 0)
		return (err);

	ehdr = ef.ehdr;

	/*
	 * Check to see what sort of module we are.
	 */
	kfp = file_findfile(NULL, __elfN(kerneltype));
#ifdef __powerpc__
	/*
	 * Kernels can be ET_DYN, so just assume the first loaded object is the
	 * kernel. This assumption will be checked later.
	 */
	if (kfp == NULL)
		ef.kernel = 1;
#endif
	if (ef.kernel || ehdr->e_type == ET_EXEC) {
		/* Looks like a kernel */
		if (kfp != NULL) {
			printf("elf" __XSTRING(__ELF_WORD_SIZE)
			    "_loadfile: kernel already loaded\n");
			err = EPERM;
			goto oerr;
		}
		/*
		 * Calculate destination address based on kernel entrypoint.
		 *
		 * For ARM, the destination address is independent of any values
		 * in the elf header (an ARM kernel can be loaded at any 2MB
		 * boundary), so we leave dest set to the value calculated by
		 * archsw.arch_loadaddr() and passed in to this function.
		 */
#ifndef __arm__
		if (ehdr->e_type == ET_EXEC)
			dest = (ehdr->e_entry & ~PAGE_MASK);
#endif
		if ((ehdr->e_entry & ~PAGE_MASK) == 0) {
			printf("elf" __XSTRING(__ELF_WORD_SIZE)
			    "_loadfile: not a kernel (maybe static binary?)\n");
			err = EPERM;
			goto oerr;
		}
		ef.kernel = 1;

	} else if (ehdr->e_type == ET_DYN) {
		/* Looks like a kld module */
		if (multiboot != 0) {
			printf("elf" __XSTRING(__ELF_WORD_SIZE)
			    "_loadfile: can't load module as multiboot\n");
			err = EPERM;
			goto oerr;
		}
		if (kfp == NULL) {
			printf("elf" __XSTRING(__ELF_WORD_SIZE)
			    "_loadfile: can't load module before kernel\n");
			err = EPERM;
			goto oerr;
		}
		if (strcmp(__elfN(kerneltype), kfp->f_type)) {
			printf("elf" __XSTRING(__ELF_WORD_SIZE)
			 "_loadfile: can't load module with kernel type '%s'\n",
			    kfp->f_type);
			err = EPERM;
			goto oerr;
		}
		/* Looks OK, got ahead */
		ef.kernel = 0;

	} else {
		err = EFTYPE;
		goto oerr;
	}

	if (archsw.arch_loadaddr != NULL)
		dest = archsw.arch_loadaddr(LOAD_ELF, ehdr, dest);
	else
		dest = roundup(dest, PAGE_SIZE);

	/*
	 * Ok, we think we should handle this.
	 */
	fp = file_alloc();
	if (fp == NULL) {
		printf("elf" __XSTRING(__ELF_WORD_SIZE)
		    "_loadfile: cannot allocate module info\n");
		err = EPERM;
		goto out;
	}
	if (ef.kernel == 1 && multiboot == 0)
		setenv("kernelname", filename, 1);
	fp->f_name = strdup(filename);
	if (multiboot == 0)
		fp->f_type = strdup(ef.kernel ?
		    __elfN(kerneltype) : __elfN(moduletype));
	else
		fp->f_type = strdup("elf multiboot kernel");

#ifdef ELF_VERBOSE
	if (ef.kernel)
		printf("%s entry at 0x%jx\n", filename,
		    (uintmax_t)ehdr->e_entry);
#else
	printf("%s ", filename);
#endif

	fp->f_size = __elfN(loadimage)(fp, &ef, dest);
	if (fp->f_size == 0 || fp->f_addr == 0)
		goto ioerr;

	/* save exec header as metadata */
	file_addmetadata(fp, MODINFOMD_ELFHDR, sizeof(*ehdr), ehdr);

	/* Load OK, return module pointer */
	*result = (struct preloaded_file *)fp;
	err = 0;
#ifdef __amd64__
	fp->f_kernphys_relocatable = multiboot || is_kernphys_relocatable(&ef);
#endif
#ifdef __i386__
	fp->f_tg_kernel_support = is_tg_kernel_support(fp, &ef);
#endif
	goto out;

ioerr:
	err = EIO;
oerr:
	file_discard(fp);
out:
	if (ef.firstpage)
		free(ef.firstpage);
	if (ef.fd != -1) {
#ifdef LOADER_VERIEXEC_VECTX
		if (!err && ef.vctx) {
			int verror;

			verror = vectx_close(ef.vctx, VE_MUST, __func__);
			if (verror) {
				err = EAUTH;
				file_discard(fp);
			}
		}
#endif
		close(ef.fd);
	}
	return (err);
}

/*
 * With the file (fd) open on the image, and (ehdr) containing
 * the Elf header, load the image at (off)
 */
static int
__elfN(loadimage)(struct preloaded_file *fp, elf_file_t ef, uint64_t off)
{
	int		i;
	u_int		j;
	Elf_Ehdr	*ehdr;
	Elf_Phdr	*phdr, *php;
	Elf_Shdr	*shdr;
	char		*shstr;
	int		ret;
	vm_offset_t	firstaddr;
	vm_offset_t	lastaddr;
	size_t		chunk;
	ssize_t		result;
	Elf_Addr	ssym, esym;
	Elf_Dyn		*dp;
	Elf_Addr	adp;
	Elf_Addr	ctors;
	int		ndp;
	int		symstrindex;
	int		symtabindex;
	Elf_Size	size;
	u_int		fpcopy;
	Elf_Sym		sym;
	Elf_Addr	p_start, p_end;

	dp = NULL;
	shdr = NULL;
	ret = 0;
	firstaddr = lastaddr = 0;
	ehdr = ef->ehdr;
#ifdef __powerpc__
	if (ef->kernel) {
#else
	if (ehdr->e_type == ET_EXEC) {
#endif
#if defined(__i386__) || defined(__amd64__)
#if __ELF_WORD_SIZE == 64
		/* x86_64 relocates after locore */
		off = - (off & 0xffffffffff000000ull);
#else
		/* i386 relocates after locore */
		off = - (off & 0xff000000u);
#endif
#elif defined(__powerpc__)
		/*
		 * On the purely virtual memory machines like e500, the kernel
		 * is linked against its final VA range, which is most often
		 * not available at the loader stage, but only after kernel
		 * initializes and completes its VM settings. In such cases we
		 * cannot use p_vaddr field directly to load ELF segments, but
		 * put them at some 'load-time' locations.
		 */
		if (off & 0xf0000000u) {
			off = -(off & 0xf0000000u);
			/*
			 * XXX the physical load address should not be
			 * hardcoded. Note that the Book-E kernel assumes that
			 * it's loaded at a 16MB boundary for now...
			 */
			off += 0x01000000;
		}
		ehdr->e_entry += off;
#ifdef ELF_VERBOSE
		printf("Converted entry 0x%jx\n", (uintmax_t)ehdr->e_entry);
#endif
#elif defined(__arm__) && !defined(EFI)
		/*
		 * The elf headers in arm kernels specify virtual addresses in
		 * all header fields, even the ones that should be physical
		 * addresses.  We assume the entry point is in the first page,
		 * and masking the page offset will leave us with the virtual
		 * address the kernel was linked at.  We subtract that from the
		 * load offset, making 'off' into the value which, when added
		 * to a virtual address in an elf header, translates it to a
		 * physical address.  We do the va->pa conversion on the entry
		 * point address in the header now, so that later we can launch
		 * the kernel by just jumping to that address.
		 *
		 * When booting from UEFI the copyin and copyout functions
		 * handle adjusting the location relative to the first virtual
		 * address.  Because of this there is no need to adjust the
		 * offset or entry point address as these will both be handled
		 * by the efi code.
		 */
		off -= ehdr->e_entry & ~PAGE_MASK;
		ehdr->e_entry += off;
#ifdef ELF_VERBOSE
		printf("ehdr->e_entry 0x%jx, va<->pa off %llx\n",
		    (uintmax_t)ehdr->e_entry, off);
#endif
#else
		off = 0;	/* other archs use direct mapped kernels */
#endif
	}
	ef->off = off;

	if (ef->kernel)
		__elfN(relocation_offset) = off;

	if ((ehdr->e_phoff + ehdr->e_phnum * sizeof(*phdr)) > ef->firstlen) {
		printf("elf" __XSTRING(__ELF_WORD_SIZE)
		    "_loadimage: program header not within first page\n");
		goto out;
	}
	phdr = (Elf_Phdr *)(ef->firstpage + ehdr->e_phoff);

	for (i = 0; i < ehdr->e_phnum; i++) {
		if (elf_program_header_convert(ehdr, phdr))
			continue;

		/* We want to load PT_LOAD segments only.. */
		if (phdr[i].p_type != PT_LOAD)
			continue;

#ifdef ELF_VERBOSE
		printf("Segment: 0x%lx@0x%lx -> 0x%lx-0x%lx",
		    (long)phdr[i].p_filesz, (long)phdr[i].p_offset,
		    (long)(phdr[i].p_vaddr + off),
		    (long)(phdr[i].p_vaddr + off + phdr[i].p_memsz - 1));
#else
		if ((phdr[i].p_flags & PF_W) == 0) {
			printf("text=0x%lx ", (long)phdr[i].p_filesz);
		} else {
			printf("data=0x%lx", (long)phdr[i].p_filesz);
			if (phdr[i].p_filesz < phdr[i].p_memsz)
				printf("+0x%lx", (long)(phdr[i].p_memsz -
				    phdr[i].p_filesz));
			printf(" ");
		}
#endif
		fpcopy = 0;
		if (ef->firstlen > phdr[i].p_offset) {
			fpcopy = ef->firstlen - phdr[i].p_offset;
			archsw.arch_copyin(ef->firstpage + phdr[i].p_offset,
			    phdr[i].p_vaddr + off, fpcopy);
		}
		if (phdr[i].p_filesz > fpcopy) {
			if (kern_pread(VECTX_HANDLE(ef),
			    phdr[i].p_vaddr + off + fpcopy,
			    phdr[i].p_filesz - fpcopy,
			    phdr[i].p_offset + fpcopy) != 0) {
				printf("\nelf" __XSTRING(__ELF_WORD_SIZE)
				    "_loadimage: read failed\n");
				goto out;
			}
		}
		/* clear space from oversized segments; eg: bss */
		if (phdr[i].p_filesz < phdr[i].p_memsz) {
#ifdef ELF_VERBOSE
			printf(" (bss: 0x%lx-0x%lx)",
			    (long)(phdr[i].p_vaddr + off + phdr[i].p_filesz),
			    (long)(phdr[i].p_vaddr + off + phdr[i].p_memsz -1));
#endif

			kern_bzero(phdr[i].p_vaddr + off + phdr[i].p_filesz,
			    phdr[i].p_memsz - phdr[i].p_filesz);
		}
#ifdef ELF_VERBOSE
		printf("\n");
#endif

		if (archsw.arch_loadseg != NULL)
			archsw.arch_loadseg(ehdr, phdr + i, off);

		if (firstaddr == 0 || firstaddr > (phdr[i].p_vaddr + off))
			firstaddr = phdr[i].p_vaddr + off;
		if (lastaddr == 0 || lastaddr <
		    (phdr[i].p_vaddr + off + phdr[i].p_memsz))
			lastaddr = phdr[i].p_vaddr + off + phdr[i].p_memsz;
	}
	lastaddr = roundup(lastaddr, sizeof(long));

	/*
	 * Get the section headers.  We need this for finding the .ctors
	 * section as well as for loading any symbols.  Both may be hard
	 * to do if reading from a .gz file as it involves seeking.  I
	 * think the rule is going to have to be that you must strip a
	 * file to remove symbols before gzipping it.
	 */
	chunk = (size_t)ehdr->e_shnum * (size_t)ehdr->e_shentsize;
	if (chunk == 0 || ehdr->e_shoff == 0)
		goto nosyms;
	shdr = alloc_pread(VECTX_HANDLE(ef), ehdr->e_shoff, chunk);
	if (shdr == NULL) {
		printf("\nelf" __XSTRING(__ELF_WORD_SIZE)
		    "_loadimage: failed to read section headers");
		goto nosyms;
	}

	for (i = 0; i < ehdr->e_shnum; i++)
		elf_section_header_convert(ehdr, &shdr[i]);

	file_addmetadata(fp, MODINFOMD_SHDR, chunk, shdr);

	/*
	 * Read the section string table and look for the .ctors section.
	 * We need to tell the kernel where it is so that it can call the
	 * ctors.
	 */
	chunk = shdr[ehdr->e_shstrndx].sh_size;
	if (chunk) {
		shstr = alloc_pread(VECTX_HANDLE(ef),
		    shdr[ehdr->e_shstrndx].sh_offset, chunk);
		if (shstr) {
			for (i = 0; i < ehdr->e_shnum; i++) {
				if (strcmp(shstr + shdr[i].sh_name,
				    ".ctors") != 0)
					continue;
				ctors = shdr[i].sh_addr;
				file_addmetadata(fp, MODINFOMD_CTORS_ADDR,
				    sizeof(ctors), &ctors);
				size = shdr[i].sh_size;
				file_addmetadata(fp, MODINFOMD_CTORS_SIZE,
				    sizeof(size), &size);
				break;
			}
			free(shstr);
		}
	}

	/*
	 * Now load any symbols.
	 */
	symtabindex = -1;
	symstrindex = -1;
	for (i = 0; i < ehdr->e_shnum; i++) {
		if (shdr[i].sh_type != SHT_SYMTAB)
			continue;
		for (j = 0; j < ehdr->e_phnum; j++) {
			if (phdr[j].p_type != PT_LOAD)
				continue;
			if (shdr[i].sh_offset >= phdr[j].p_offset &&
			    (shdr[i].sh_offset + shdr[i].sh_size <=
			    phdr[j].p_offset + phdr[j].p_filesz)) {
				shdr[i].sh_offset = 0;
				shdr[i].sh_size = 0;
				break;
			}
		}
		if (shdr[i].sh_offset == 0 || shdr[i].sh_size == 0)
			continue;	/* alread loaded in a PT_LOAD above */
		/* Save it for loading below */
		symtabindex = i;
		symstrindex = shdr[i].sh_link;
	}
	if (symtabindex < 0 || symstrindex < 0)
		goto nosyms;

	/* Ok, committed to a load. */
#ifndef ELF_VERBOSE
	printf("syms=[");
#endif
	ssym = lastaddr;
	for (i = symtabindex; i >= 0; i = symstrindex) {
#ifdef ELF_VERBOSE
		char	*secname;

		switch(shdr[i].sh_type) {
		case SHT_SYMTAB:		/* Symbol table */
			secname = "symtab";
			break;
		case SHT_STRTAB:		/* String table */
			secname = "strtab";
			break;
		default:
			secname = "WHOA!!";
			break;
		}
#endif
		size = shdr[i].sh_size;

		archsw.arch_copyin(&size, lastaddr, sizeof(size));
		lastaddr += sizeof(size);

#ifdef ELF_VERBOSE
		printf("\n%s: 0x%jx@0x%jx -> 0x%jx-0x%jx", secname,
		    (uintmax_t)shdr[i].sh_size, (uintmax_t)shdr[i].sh_offset,
		    (uintmax_t)lastaddr,
		    (uintmax_t)(lastaddr + shdr[i].sh_size));
#else
		if (i == symstrindex)
			printf("+");
		printf("0x%lx+0x%lx", (long)sizeof(size), (long)size);
#endif

		if (VECTX_LSEEK(VECTX_HANDLE(ef), (off_t)shdr[i].sh_offset, SEEK_SET) == -1) {
			printf("\nelf" __XSTRING(__ELF_WORD_SIZE)
			   "_loadimage: could not seek for symbols - skipped!");
			lastaddr = ssym;
			ssym = 0;
			goto nosyms;
		}
		result = archsw.arch_readin(VECTX_HANDLE(ef), lastaddr, shdr[i].sh_size);
		if (result < 0 || (size_t)result != shdr[i].sh_size) {
			printf("\nelf" __XSTRING(__ELF_WORD_SIZE)
			    "_loadimage: could not read symbols - skipped! "
			    "(%ju != %ju)", (uintmax_t)result,
			    (uintmax_t)shdr[i].sh_size);
			lastaddr = ssym;
			ssym = 0;
			goto nosyms;
		}
		/* Reset offsets relative to ssym */
		lastaddr += shdr[i].sh_size;
		lastaddr = roundup(lastaddr, sizeof(size));
		if (i == symtabindex)
			symtabindex = -1;
		else if (i == symstrindex)
			symstrindex = -1;
	}
	esym = lastaddr;
#ifndef ELF_VERBOSE
	printf("]");
#endif

	file_addmetadata(fp, MODINFOMD_SSYM, sizeof(ssym), &ssym);
	file_addmetadata(fp, MODINFOMD_ESYM, sizeof(esym), &esym);

nosyms:
	printf("\n");

	ret = lastaddr - firstaddr;
	fp->f_addr = firstaddr;

	php = NULL;
	for (i = 0; i < ehdr->e_phnum; i++) {
		if (phdr[i].p_type == PT_DYNAMIC) {
			php = phdr + i;
			adp = php->p_vaddr;
			file_addmetadata(fp, MODINFOMD_DYNAMIC, sizeof(adp),
			    &adp);
			break;
		}
	}

	if (php == NULL) /* this is bad, we cannot get to symbols or _DYNAMIC */
		goto out;

	ndp = php->p_filesz / sizeof(Elf_Dyn);
	if (ndp == 0)
		goto out;
	dp = malloc(php->p_filesz);
	if (dp == NULL)
		goto out;
	archsw.arch_copyout(php->p_vaddr + off, dp, php->p_filesz);

	ef->strsz = 0;
	for (i = 0; i < ndp; i++) {
		if (dp[i].d_tag == 0)
			break;
		switch (dp[i].d_tag) {
		case DT_HASH:
			ef->hashtab =
			    (Elf_Hashelt*)(uintptr_t)(dp[i].d_un.d_ptr + off);
			break;
		case DT_STRTAB:
			ef->strtab =
			    (char *)(uintptr_t)(dp[i].d_un.d_ptr + off);
			break;
		case DT_STRSZ:
			ef->strsz = dp[i].d_un.d_val;
			break;
		case DT_SYMTAB:
			ef->symtab =
			    (Elf_Sym *)(uintptr_t)(dp[i].d_un.d_ptr + off);
			break;
		case DT_REL:
			ef->rel =
			    (Elf_Rel *)(uintptr_t)(dp[i].d_un.d_ptr + off);
			break;
		case DT_RELSZ:
			ef->relsz = dp[i].d_un.d_val;
			break;
		case DT_RELA:
			ef->rela =
			    (Elf_Rela *)(uintptr_t)(dp[i].d_un.d_ptr + off);
			break;
		case DT_RELASZ:
			ef->relasz = dp[i].d_un.d_val;
			break;
		default:
			break;
		}
	}
	if (ef->hashtab == NULL || ef->symtab == NULL ||
	    ef->strtab == NULL || ef->strsz == 0)
		goto out;
	COPYOUT(ef->hashtab, &ef->nbuckets, sizeof(ef->nbuckets));
	COPYOUT(ef->hashtab + 1, &ef->nchains, sizeof(ef->nchains));
	ef->buckets = ef->hashtab + 2;
	ef->chains = ef->buckets + ef->nbuckets;

	if (__elfN(lookup_symbol)(ef, "__start_set_modmetadata_set", &sym,
	    STT_NOTYPE) != 0)
		return 0;
	p_start = sym.st_value + ef->off;
	if (__elfN(lookup_symbol)(ef, "__stop_set_modmetadata_set", &sym,
	    STT_NOTYPE) != 0)
		return 0;
	p_end = sym.st_value + ef->off;

	if (__elfN(parse_modmetadata)(fp, ef, p_start, p_end) == 0)
		goto out;

	if (ef->kernel)		/* kernel must not depend on anything */
		goto out;

out:
	if (dp)
		free(dp);
	if (shdr)
		free(shdr);
	return ret;
}

static char invalid_name[] = "bad";

char *
fake_modname(const char *name)
{
	const char *sp, *ep;
	char *fp;
	size_t len;

	sp = strrchr(name, '/');
	if (sp)
		sp++;
	else
		sp = name;

	ep = strrchr(sp, '.');
	if (ep == NULL) {
		ep = sp + strlen(sp);
	}
	if (ep == sp) {
		sp = invalid_name;
		ep = invalid_name + sizeof(invalid_name) - 1;
	}

	len = ep - sp;
	fp = malloc(len + 1);
	if (fp == NULL)
		return NULL;
	memcpy(fp, sp, len);
	fp[len] = '\0';
	return fp;
}

#if (defined(__i386__) || defined(__powerpc__)) && __ELF_WORD_SIZE == 64
struct mod_metadata64 {
	int		md_version;	/* structure version MDTV_* */
	int		md_type;	/* type of entry MDT_* */
	uint64_t	md_data;	/* specific data */
	uint64_t	md_cval;	/* common string label */
};
#endif
#if defined(__amd64__) && __ELF_WORD_SIZE == 32
struct mod_metadata32 {
	int		md_version;	/* structure version MDTV_* */
	int		md_type;	/* type of entry MDT_* */
	uint32_t	md_data;	/* specific data */
	uint32_t	md_cval;	/* common string label */
};
#endif

int
__elfN(load_modmetadata)(struct preloaded_file *fp, uint64_t dest)
{
	struct elf_file		 ef;
	int			 err, i, j;
	Elf_Shdr		*sh_meta, *shdr = NULL;
	Elf_Shdr		*sh_data[2];
	char			*shstrtab = NULL;
	size_t			 size;
	Elf_Addr		 p_start, p_end;

	bzero(&ef, sizeof(struct elf_file));
	ef.fd = -1;

	err = __elfN(load_elf_header)(fp->f_name, &ef);
	if (err != 0)
		goto out;

	if (ef.kernel == 1 || ef.ehdr->e_type == ET_EXEC) {
		ef.kernel = 1;
	} else if (ef.ehdr->e_type != ET_DYN) {
		err = EFTYPE;
		goto out;
	}

	size = (size_t)ef.ehdr->e_shnum * (size_t)ef.ehdr->e_shentsize;
	shdr = alloc_pread(VECTX_HANDLE(&ef), ef.ehdr->e_shoff, size);
	if (shdr == NULL) {
		err = ENOMEM;
		goto out;
	}

	/* Load shstrtab. */
	shstrtab = alloc_pread(VECTX_HANDLE(&ef), shdr[ef.ehdr->e_shstrndx].sh_offset,
	    shdr[ef.ehdr->e_shstrndx].sh_size);
	if (shstrtab == NULL) {
		printf("\nelf" __XSTRING(__ELF_WORD_SIZE)
		    "load_modmetadata: unable to load shstrtab\n");
		err = EFTYPE;
		goto out;
	}

	/* Find set_modmetadata_set and data sections. */
	sh_data[0] = sh_data[1] = sh_meta = NULL;
	for (i = 0, j = 0; i < ef.ehdr->e_shnum; i++) {
		if (strcmp(&shstrtab[shdr[i].sh_name],
		    "set_modmetadata_set") == 0) {
			sh_meta = &shdr[i];
		}
		if ((strcmp(&shstrtab[shdr[i].sh_name], ".data") == 0) ||
		    (strcmp(&shstrtab[shdr[i].sh_name], ".rodata") == 0)) {
			sh_data[j++] = &shdr[i];
		}
	}
	if (sh_meta == NULL || sh_data[0] == NULL || sh_data[1] == NULL) {
		printf("\nelf" __XSTRING(__ELF_WORD_SIZE)
    "load_modmetadata: unable to find set_modmetadata_set or data sections\n");
		err = EFTYPE;
		goto out;
	}

	/* Load set_modmetadata_set into memory */
	err = kern_pread(VECTX_HANDLE(&ef), dest, sh_meta->sh_size, sh_meta->sh_offset);
	if (err != 0) {
		printf("\nelf" __XSTRING(__ELF_WORD_SIZE)
    "load_modmetadata: unable to load set_modmetadata_set: %d\n", err);
		goto out;
	}
	p_start = dest;
	p_end = dest + sh_meta->sh_size;
	dest += sh_meta->sh_size;

	/* Load data sections into memory. */
	err = kern_pread(VECTX_HANDLE(&ef), dest, sh_data[0]->sh_size,
	    sh_data[0]->sh_offset);
	if (err != 0) {
		printf("\nelf" __XSTRING(__ELF_WORD_SIZE)
		    "load_modmetadata: unable to load data: %d\n", err);
		goto out;
	}

	/*
	 * We have to increment the dest, so that the offset is the same into
	 * both the .rodata and .data sections.
	 */
	ef.off = -(sh_data[0]->sh_addr - dest);
	dest +=	(sh_data[1]->sh_addr - sh_data[0]->sh_addr);

	err = kern_pread(VECTX_HANDLE(&ef), dest, sh_data[1]->sh_size,
	    sh_data[1]->sh_offset);
	if (err != 0) {
		printf("\nelf" __XSTRING(__ELF_WORD_SIZE)
		    "load_modmetadata: unable to load data: %d\n", err);
		goto out;
	}

	err = __elfN(parse_modmetadata)(fp, &ef, p_start, p_end);
	if (err != 0) {
		printf("\nelf" __XSTRING(__ELF_WORD_SIZE)
		    "load_modmetadata: unable to parse metadata: %d\n", err);
		goto out;
	}

out:
	if (shstrtab != NULL)
		free(shstrtab);
	if (shdr != NULL)
		free(shdr);
	if (ef.firstpage != NULL)
		free(ef.firstpage);
	if (ef.fd != -1) {
#ifdef LOADER_VERIEXEC_VECTX
		if (!err && ef.vctx) {
			int verror;

			verror = vectx_close(ef.vctx, VE_MUST, __func__);
			if (verror) {
				err = EAUTH;
				file_discard(fp);
			}
		}
#endif
		close(ef.fd);
	}
	return (err);
}

int
__elfN(parse_modmetadata)(struct preloaded_file *fp, elf_file_t ef,
    Elf_Addr p_start, Elf_Addr p_end)
{
	struct mod_metadata md;
#if (defined(__i386__) || defined(__powerpc__)) && __ELF_WORD_SIZE == 64
	struct mod_metadata64 md64;
#elif defined(__amd64__) && __ELF_WORD_SIZE == 32
	struct mod_metadata32 md32;
#endif
	struct mod_depend *mdepend;
	struct mod_version mver;
	char *s;
	int error, modcnt, minfolen;
	Elf_Addr v, p;

	modcnt = 0;
	p = p_start;
	while (p < p_end) {
		COPYOUT(p, &v, sizeof(v));
		error = __elfN(reloc_ptr)(fp, ef, p, &v, sizeof(v));
		if (error == EOPNOTSUPP)
			v += ef->off;
		else if (error != 0)
			return (error);
#if (defined(__i386__) || defined(__powerpc__)) && __ELF_WORD_SIZE == 64
		COPYOUT(v, &md64, sizeof(md64));
		error = __elfN(reloc_ptr)(fp, ef, v, &md64, sizeof(md64));
		if (error == EOPNOTSUPP) {
			md64.md_cval += ef->off;
			md64.md_data += ef->off;
		} else if (error != 0)
			return (error);
		md.md_version = md64.md_version;
		md.md_type = md64.md_type;
		md.md_cval = (const char *)(uintptr_t)md64.md_cval;
		md.md_data = (void *)(uintptr_t)md64.md_data;
#elif defined(__amd64__) && __ELF_WORD_SIZE == 32
		COPYOUT(v, &md32, sizeof(md32));
		error = __elfN(reloc_ptr)(fp, ef, v, &md32, sizeof(md32));
		if (error == EOPNOTSUPP) {
			md32.md_cval += ef->off;
			md32.md_data += ef->off;
		} else if (error != 0)
			return (error);
		md.md_version = md32.md_version;
		md.md_type = md32.md_type;
		md.md_cval = (const char *)(uintptr_t)md32.md_cval;
		md.md_data = (void *)(uintptr_t)md32.md_data;
#else
		COPYOUT(v, &md, sizeof(md));
		error = __elfN(reloc_ptr)(fp, ef, v, &md, sizeof(md));
		if (error == EOPNOTSUPP) {
			md.md_cval += ef->off;
			md.md_data = (void *)((uintptr_t)md.md_data +
			    (uintptr_t)ef->off);
		} else if (error != 0)
			return (error);
#endif
		p += sizeof(Elf_Addr);
		switch(md.md_type) {
		case MDT_DEPEND:
			if (ef->kernel) /* kernel must not depend on anything */
				break;
			s = strdupout((vm_offset_t)md.md_cval);
			minfolen = sizeof(*mdepend) + strlen(s) + 1;
			mdepend = malloc(minfolen);
			if (mdepend == NULL)
				return ENOMEM;
			COPYOUT((vm_offset_t)md.md_data, mdepend,
			    sizeof(*mdepend));
			strcpy((char*)(mdepend + 1), s);
			free(s);
			file_addmetadata(fp, MODINFOMD_DEPLIST, minfolen,
			    mdepend);
			free(mdepend);
			break;
		case MDT_VERSION:
			s = strdupout((vm_offset_t)md.md_cval);
			COPYOUT((vm_offset_t)md.md_data, &mver, sizeof(mver));
			file_addmodule(fp, s, mver.mv_version, NULL);
			free(s);
			modcnt++;
			break;
		}
	}
	if (modcnt == 0) {
		s = fake_modname(fp->f_name);
		file_addmodule(fp, s, 1, NULL);
		free(s);
	}
	return 0;
}

static unsigned long
elf_hash(const char *name)
{
	const unsigned char *p = (const unsigned char *) name;
	unsigned long h = 0;
	unsigned long g;

	while (*p != '\0') {
		h = (h << 4) + *p++;
		if ((g = h & 0xf0000000) != 0)
			h ^= g >> 24;
		h &= ~g;
	}
	return h;
}

static const char __elfN(bad_symtable)[] = "elf" __XSTRING(__ELF_WORD_SIZE)
    "_lookup_symbol: corrupt symbol table\n";
int
__elfN(lookup_symbol)(elf_file_t ef, const char* name, Elf_Sym *symp,
    unsigned char type)
{
	Elf_Hashelt symnum;
	Elf_Sym sym;
	char *strp;
	unsigned long hash;

	if (ef->nbuckets == 0) {
		printf(__elfN(bad_symtable));
		return ENOENT;
	}

	hash = elf_hash(name);
	COPYOUT(&ef->buckets[hash % ef->nbuckets], &symnum, sizeof(symnum));

	while (symnum != STN_UNDEF) {
		if (symnum >= ef->nchains) {
			printf(__elfN(bad_symtable));
			return ENOENT;
		}

		COPYOUT(ef->symtab + symnum, &sym, sizeof(sym));
		if (sym.st_name == 0) {
			printf(__elfN(bad_symtable));
			return ENOENT;
		}

		strp = strdupout((vm_offset_t)(ef->strtab + sym.st_name));
		if (strcmp(name, strp) == 0) {
			free(strp);
			if (sym.st_shndx != SHN_UNDEF ||
			    (sym.st_value != 0 &&
			    ELF_ST_TYPE(sym.st_info) == type)) {
				*symp = sym;
				return 0;
			}
			return ENOENT;
		}
		free(strp);
		COPYOUT(&ef->chains[symnum], &symnum, sizeof(symnum));
	}
	return ENOENT;
}

/*
 * Apply any intra-module relocations to the value. p is the load address
 * of the value and val/len is the value to be modified. This does NOT modify
 * the image in-place, because this is done by kern_linker later on.
 *
 * Returns EOPNOTSUPP if no relocation method is supplied.
 */
static int
__elfN(reloc_ptr)(struct preloaded_file *mp, elf_file_t ef,
    Elf_Addr p, void *val, size_t len)
{
	size_t n;
	Elf_Rela a;
	Elf_Rel r;
	int error;

	/*
	 * The kernel is already relocated, but we still want to apply
	 * offset adjustments.
	 */
	if (ef->kernel)
		return (EOPNOTSUPP);

	for (n = 0; n < ef->relsz / sizeof(r); n++) {
		COPYOUT(ef->rel + n, &r, sizeof(r));

		error = __elfN(reloc)(ef, __elfN(symaddr), &r, ELF_RELOC_REL,
		    ef->off, p, val, len);
		if (error != 0)
			return (error);
	}
	for (n = 0; n < ef->relasz / sizeof(a); n++) {
		COPYOUT(ef->rela + n, &a, sizeof(a));

		error = __elfN(reloc)(ef, __elfN(symaddr), &a, ELF_RELOC_RELA,
		    ef->off, p, val, len);
		if (error != 0)
			return (error);
	}

	return (0);
}

static Elf_Addr
__elfN(symaddr)(struct elf_file *ef, Elf_Size symidx)
{

	/* Symbol lookup by index not required here. */
	return (0);
}