dis - zygaena/hammerhead

root / base / usr / src / cmd / dis
dis Plain Text 2521 lines 61.0 KB
#
# CDDL HEADER START
#
# The contents of this file are subject to the terms of the
# Common Development and Distribution License (the "License").
# You may not use this file except in compliance with the License.
#
# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
# or http://www.opensolaris.org/os/licensing.
# See the License for the specific language governing permissions
# and limitations under the License.
#
# When distributing Covered Code, include this CDDL HEADER in each
# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
# If applicable, add the following below this CDDL HEADER, with the
# fields enclosed by brackets "[]" replaced with your own identifying
# information: Portions Copyright [yyyy] [name of copyright owner]
#
# CDDL HEADER END
#
#
# Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
# Use is subject to license terms.
#
# Copyright 2018 Jason King
# Copyright 2024 Oxide Computer Company
#

PROG=	dis
OBJS=	dis_target.o dis_main.o dis_util.o dis_list.o $(HEXDUMP_OBJS)

include	../Makefile.cmd
include $(SRC)/common/hexdump/Makefile.com

LDLIBS += -ldisasm -luutil -lelf -ldemangle-sys
CERRWARN += $(CNOWARN_UNINIT)

.KEEP_STATE:

all: $(PROG)

$(PROG): $(OBJS)
	$(LINK.c) -o $@ $(OBJS) $(LDLIBS)
	$(POST_PROCESS)

install: all $(ROOTPROG) $(ROOTCCSBINLINK)

clean:
	$(RM) $(OBJS) $(PROG)

include $(SRC)/common/hexdump/Makefile.targ
include ../Makefile.targ
include ../Makefile.ctf
/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */

/*
 * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */

#include <stddef.h>
#include <stdlib.h>
#include <string.h>

#include "dis_target.h"
#include "dis_list.h"
#include "dis_util.h"

/*
 * List support functions.
 *
 * Support routines for managing lists of sections and functions.  We first
 * process the command line arguments into lists of strings.  For each target,
 * we resolve these strings against the set of available sections and/or
 * functions to arrive at the set of objects to disassemble.
 *
 * We export two types of lists, namelists and resolvelists.  The first is used
 * to record names given as command line options.  The latter is used to
 * maintain the data objects specific to a given target.
 */

typedef struct unresolved_name {
	const char	*un_name;	/* name of function or object */
	int		un_value;	/* user-supplied data */
	int		un_mark;	/* internal counter */
	uu_list_node_t	un_node;	/* uulist node */
} unresolved_name_t;

typedef struct resolved_name {
	void		*rn_data;	/* section or function data */
	int		rn_value;	/* user-supplied data */
	uu_list_node_t	rn_node;	/* uulist node */
} resolved_name_t;

static uu_list_pool_t *unresolved_pool;
static uu_list_pool_t *resolved_pool;
static int current_mark = 0;

static void
initialize_pools(void)
{
	unresolved_pool = uu_list_pool_create(
	    "unresolved_pool", sizeof (unresolved_name_t),
	    offsetof(unresolved_name_t, un_node), NULL, 0);
	resolved_pool = uu_list_pool_create(
	    "resolved_pool", sizeof (resolved_name_t),
	    offsetof(resolved_name_t, rn_node), NULL, 0);

	if (unresolved_pool == NULL ||
	    resolved_pool == NULL)
		die("out of memory");
}

/*
 * Returns an empty list of unresolved names.
 */
dis_namelist_t *
dis_namelist_create(void)
{
	uu_list_t *listp;

	/*
	 * If this is the first request to create a list, initialize the list
	 * pools.
	 */
	if (unresolved_pool == NULL)
		initialize_pools();

	if ((listp = uu_list_create(unresolved_pool, NULL, 0)) == NULL)
		die("out of memory");

	return (listp);
}

/*
 * Adds the given name to the unresolved list.  'value' is an arbitrary value
 * which is preserved for this entry, even when resolved against a target.  This
 * allows the caller to associate similar behavior (such as the difference
 * between -d, -D, and -s) without having to create multiple lists.
 */
void
dis_namelist_add(dis_namelist_t *list, const char *name, int value)
{
	unresolved_name_t *node;

	node = safe_malloc(sizeof (unresolved_name_t));

	node->un_name = name;
	node->un_value = value;
	node->un_mark = 0;

	(void) uu_list_insert_before(list, NULL, node);
}

/*
 * Internal callback structure used
 */
typedef struct cb_data {
	int		cb_mark;
	uu_list_t	*cb_source;
	uu_list_t	*cb_resolved;
} cb_data_t;

/*
 * For each section, walk the list of unresolved names and resolve those that
 * correspond to real functions.  We mark functions as we see them, and re-walk
 * the list a second time to warn about functions we didn't find.
 *
 * This is an O(n * m) algorithm, but we typically search for only a single
 * function.
 */
/* ARGSUSED */
static void
walk_sections(dis_tgt_t *tgt, dis_scn_t *scn, void *data)
{
	cb_data_t *cb = data;
	unresolved_name_t *unp;
	uu_list_walk_t *walk;

	if ((walk = uu_list_walk_start(cb->cb_source, UU_DEFAULT)) == NULL)
		die("out of memory");

	while ((unp = uu_list_walk_next(walk)) != NULL) {
		if (strcmp(unp->un_name, dis_section_name(scn)) == 0) {
			resolved_name_t *resolved;

			/*
			 * Mark the current node as seen
			 */
			unp->un_mark = cb->cb_mark;

			/*
			 * Add the data to the resolved list
			 */
			resolved = safe_malloc(sizeof (resolved_name_t));

			resolved->rn_data = dis_section_copy(scn);
			resolved->rn_value = unp->un_value;

			(void) uu_list_insert_before(cb->cb_resolved, NULL,
			    resolved);
		}
	}

	uu_list_walk_end(walk);
}

/*
 * Take a list of unresolved names and create a resolved list of sections.  We
 * rely on walk_sections() to do the dirty work.  After resolving the sections,
 * we check for any unmarked names and warn the user about missing sections.
 */
dis_scnlist_t *
dis_namelist_resolve_sections(dis_namelist_t *namelist, dis_tgt_t *tgt)
{
	uu_list_t *listp;
	cb_data_t cb;
	unresolved_name_t *unp;
	uu_list_walk_t *walk;

	/*
	 * Walk all sections in the target, calling walk_sections() for each
	 * one.
	 */
	if ((listp = uu_list_create(resolved_pool, NULL, UU_DEFAULT)) == NULL)
		die("out of memory");

	cb.cb_mark = ++current_mark;
	cb.cb_source = namelist;
	cb.cb_resolved = listp;

	dis_tgt_section_iter(tgt, walk_sections, &cb);

	/*
	 * Walk all elements of the unresolved list, and report any that we
	 * didn't mark in the process.
	 */
	if ((walk = uu_list_walk_start(namelist, UU_DEFAULT)) == NULL)
		die("out of memory");

	while ((unp = uu_list_walk_next(walk)) != NULL) {
		if (unp->un_mark != current_mark)
			warn("failed to find section '%s' in '%s'",
			    unp->un_name, dis_tgt_name(tgt));
	}

	uu_list_walk_end(walk);

	return (listp);
}

/*
 * Similar to walk_sections(), but for functions.
 */
/* ARGSUSED */
static void
walk_functions(dis_tgt_t *tgt, dis_func_t *func, void *data)
{
	cb_data_t *cb = data;
	unresolved_name_t *unp;
	uu_list_walk_t *walk;

	if ((walk = uu_list_walk_start(cb->cb_source, UU_DEFAULT)) == NULL)
		die("out of memory");

	while ((unp = uu_list_walk_next(walk)) != NULL) {
		if (strcmp(unp->un_name, dis_function_name(func)) == 0) {
			resolved_name_t *resolved;

			unp->un_mark = cb->cb_mark;

			resolved = safe_malloc(sizeof (resolved_name_t));

			resolved->rn_data = dis_function_copy(func);
			resolved->rn_value = unp->un_value;

			(void) uu_list_insert_before(cb->cb_resolved, NULL,
			    resolved);
		}
	}

	uu_list_walk_end(walk);
}

/*
 * Take a list of unresolved names and create a resolved list of functions.  We
 * rely on walk_functions() to do the dirty work.  After resolving the
 * functions, * we check for any unmarked names and warn the user about missing
 * functions.
 */
dis_funclist_t *
dis_namelist_resolve_functions(dis_namelist_t *namelist, dis_tgt_t *tgt)
{
	uu_list_t *listp;
	uu_list_walk_t *walk;
	unresolved_name_t *unp;
	cb_data_t cb;

	if ((listp = uu_list_create(resolved_pool, NULL, UU_DEFAULT)) == NULL)
		die("out of memory");

	cb.cb_mark = ++current_mark;
	cb.cb_source = namelist;
	cb.cb_resolved = listp;

	dis_tgt_function_iter(tgt, walk_functions, &cb);

	/*
	 * Walk unresolved list and report any missing functions.
	 */
	if ((walk = uu_list_walk_start(namelist, UU_DEFAULT)) == NULL)
		die("out of memory");

	while ((unp = uu_list_walk_next(walk)) != NULL) {
		if (unp->un_mark != current_mark)
			warn("failed to find function '%s' in '%s'",
			    unp->un_name, dis_tgt_name(tgt));
	}

	uu_list_walk_end(walk);

	return (listp);
}

/*
 * Returns true if the given list is empty.
 */
int
dis_namelist_empty(dis_namelist_t *list)
{
	return (uu_list_numnodes(list) == 0);
}

static void
free_list(uu_list_t *list)
{
	uu_list_walk_t *walk;
	void *data;

	if ((walk = uu_list_walk_start(list, UU_WALK_ROBUST)) == NULL)
		die("out of memory");

	while ((data = uu_list_walk_next(walk)) != NULL) {
		uu_list_remove(list, data);
		free(data);
	}

	uu_list_walk_end(walk);

	uu_list_destroy(list);
}

/*
 * Destroy a list of sections.  First, walk the list and free the associated
 * section data.  Pass the list onto to free_list() to clean up the rest of the
 * list.
 */
void
dis_scnlist_destroy(dis_scnlist_t *list)
{
	uu_list_walk_t *walk;
	resolved_name_t *data;

	if ((walk = uu_list_walk_start(list, UU_DEFAULT)) == NULL)
		die("out of memory");

	while ((data = uu_list_walk_next(walk)) != NULL)
		dis_section_free(data->rn_data);

	uu_list_walk_end(walk);

	free_list(list);
}

/*
 * Destroy a list of functions.  First, walk the list and free the associated
 * function data.  Pass the list onto to free_list() to clean up the rest of the
 * list.
 */
void
dis_funclist_destroy(dis_funclist_t *list)
{
	uu_list_walk_t *walk;
	resolved_name_t *data;

	if ((walk = uu_list_walk_start(list, UU_DEFAULT)) == NULL)
		die("out of memory");

	while ((data = uu_list_walk_next(walk)) != NULL)
		dis_function_free(data->rn_data);

	uu_list_walk_end(walk);

	free_list(list);
}

/*
 * Destroy a lis tof unresolved names.
 */
void
dis_namelist_destroy(dis_namelist_t *list)
{
	free_list(list);
}

/*
 * Iterate over a resolved list of sections.
 */
void
dis_scnlist_iter(uu_list_t *list, void (*func)(dis_scn_t *, int, void *),
    void *arg)
{
	uu_list_walk_t *walk;
	resolved_name_t *data;

	if ((walk = uu_list_walk_start(list, UU_DEFAULT)) == NULL)
		die("out of memory");

	while ((data = uu_list_walk_next(walk)) != NULL)
		func(data->rn_data, data->rn_value, arg);

	uu_list_walk_end(walk);
}

/*
 * Iterate over a resolved list of functions.
 */
void
dis_funclist_iter(uu_list_t *list, void (*func)(dis_func_t *, int, void *),
    void *arg)
{
	uu_list_walk_t *walk;
	resolved_name_t *data;

	if ((walk = uu_list_walk_start(list, UU_DEFAULT)) == NULL)
		die("out of memory");

	while ((data = uu_list_walk_next(walk)) != NULL)
		func(data->rn_data, data->rn_value, arg);

	uu_list_walk_end(walk);
}
/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */

/*
 * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */

#ifndef	_DIS_LIST_H
#define	_DIS_LIST_H

#include <libuutil.h>

/* #include "dis_target.h" */

#ifdef	__cplusplus
extern "C" {
#endif

typedef uu_list_t dis_namelist_t;
typedef uu_list_t dis_scnlist_t;
typedef uu_list_t dis_funclist_t;

dis_namelist_t *dis_namelist_create(void);
void dis_namelist_add(dis_namelist_t *, const char *, int);
dis_funclist_t *dis_namelist_resolve_functions(dis_namelist_t *, dis_tgt_t *);
dis_scnlist_t *dis_namelist_resolve_sections(dis_namelist_t *, dis_tgt_t *);
void dis_scnlist_iter(dis_scnlist_t *, void (*)(dis_scn_t *, int, void *),
    void *);
void dis_funclist_iter(dis_funclist_t *, void (*)(dis_func_t *, int, void *),
    void *);
int dis_namelist_empty(dis_namelist_t *);
void dis_scnlist_destroy(dis_scnlist_t *);
void dis_funclist_destroy(dis_funclist_t *);
void dis_namelist_destroy(dis_namelist_t *);

#ifdef	__cplusplus
}
#endif

#endif	/* _DIS_LIST_H */
/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */

/*
 * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 *
 * Copyright 2011 Jason King.  All rights reserved.
 * Copyright 2012 Joshua M. Clulow <josh@sysmgr.org>
 * Copyright 2015 Josef 'Jeff' Sipek <jeffpc@josefsipek.net>
 * Copyright 2018, Joyent, Inc.
 * Copyright 2024 Oxide Computer Company
 */

#include <ctype.h>
#include <getopt.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/hexdump.h>
#include <sys/sysmacros.h>
#include <sys/elf_SPARC.h>

#include <libdisasm.h>

#include "dis_target.h"
#include "dis_util.h"
#include "dis_list.h"

int g_demangle;		/* Demangle C++ names */
int g_quiet;		/* Quiet mode */
int g_numeric;		/* Numeric mode */
int g_flags;		/* libdisasm language flags */
int g_doall;		/* true if no functions or sections were given */

dis_namelist_t *g_funclist;	/* list of functions to disassemble, if any */
dis_namelist_t *g_seclist;	/* list of sections to disassemble, if any */

/*
 * Section options for -d, -D, and -s
 */
#define	DIS_DATA_RELATIVE	1
#define	DIS_DATA_ABSOLUTE	2
#define	DIS_TEXT		3

/*
 * libdisasm callback data.  Keeps track of current data (function or section)
 * and offset within that data.
 */
typedef struct dis_buffer {
	dis_tgt_t	*db_tgt;	/* current dis target */
	void		*db_data;	/* function or section data */
	uint64_t	db_addr;	/* address of function start */
	size_t		db_size;	/* size of data */
	uint64_t	db_nextaddr;	/* next address to be read */
} dis_buffer_t;

#define	MINSYMWIDTH	22	/* Minimum width of symbol portion of line */

/*
 * Given a symbol+offset as returned by dis_tgt_lookup(), print an appropriately
 * formatted symbol, based on the offset and current setttings.
 */
void
getsymname(uint64_t addr, const char *symbol, off_t offset, char *buf,
    size_t buflen)
{
	if (symbol == NULL || g_numeric) {
		if (g_flags & DIS_OCTAL)
			(void) snprintf(buf, buflen, "0%llo", addr);
		else
			(void) snprintf(buf, buflen, "0x%llx", addr);
	} else {
		if (g_demangle)
			symbol = dis_demangle(symbol);

		if (offset == 0)
			(void) snprintf(buf, buflen, "%s", symbol);
		else if (g_flags & DIS_OCTAL)
			(void) snprintf(buf, buflen, "%s+0%o", symbol, offset);
		else
			(void) snprintf(buf, buflen, "%s+0x%x", symbol, offset);
	}
}

/*
 * Determine if we are on an architecture with fixed-size instructions,
 * and if so, what size they are.
 */
static int
insn_size(dis_handle_t *dhp)
{
	int min = dis_min_instrlen(dhp);
	int max = dis_max_instrlen(dhp);

	if (min == max)
		return (min);

	return (0);
}

/*
 * The main disassembly routine.  Given a fixed-sized buffer and starting
 * address, disassemble the data using the supplied target and libdisasm handle.
 */
void
dis_data(dis_tgt_t *tgt, dis_handle_t *dhp, uint64_t addr, void *data,
    size_t datalen)
{
	dis_buffer_t db = { 0 };
	char buf[BUFSIZE];
	char symbuf[BUFSIZE];
	const char *symbol;
	const char *last_symbol;
	off_t symoffset;
	int i;
	int bytesperline;
	size_t symsize;
	int isfunc;
	size_t symwidth = 0;
	int ret;
	int insz = insn_size(dhp);

	db.db_tgt = tgt;
	db.db_data = data;
	db.db_addr = addr;
	db.db_size = datalen;

	dis_set_data(dhp, &db);

	if ((bytesperline = dis_max_instrlen(dhp)) > 6)
		bytesperline = 6;

	symbol = NULL;

	while (addr < db.db_addr + db.db_size) {

		ret = dis_disassemble(dhp, addr, buf, BUFSIZE);
		if (ret != 0 && insz > 0) {
			/*
			 * Since we know instructions are fixed size, we
			 * always know the address of the next instruction
			 */
			(void) snprintf(buf, sizeof (buf),
			    "*** invalid opcode ***");
			db.db_nextaddr = addr + insz;

		} else if (ret != 0) {
			off_t next;

			(void) snprintf(buf, sizeof (buf),
			    "*** invalid opcode ***");

			/*
			 * On architectures with variable sized instructions
			 * we have no way to figure out where the next
			 * instruction starts if we encounter an invalid
			 * instruction.  Instead we print the rest of the
			 * instruction stream as hex until we reach the
			 * next valid symbol in the section.
			 */
			if ((next = dis_tgt_next_symbol(tgt, addr)) == 0) {
				db.db_nextaddr = db.db_addr + db.db_size;
			} else {
				if (next > db.db_size)
					db.db_nextaddr = db.db_addr +
					    db.db_size;
				else
					db.db_nextaddr = addr + next;
			}
		}

		/*
		 * Print out the line as:
		 *
		 *	address:	bytes	text
		 *
		 * If there are more than 6 bytes in any given instruction,
		 * spread the bytes across two lines.  We try to get symbolic
		 * information for the address, but if that fails we print out
		 * the numeric address instead.
		 *
		 * We try to keep the address portion of the text aligned at
		 * MINSYMWIDTH characters.  If we are disassembling a function
		 * with a long name, this can be annoying.  So we pick a width
		 * based on the maximum width that the current symbol can be.
		 * This at least produces text aligned within each function.
		 */
		last_symbol = symbol;
		symbol = dis_tgt_lookup(tgt, addr, &symoffset, 1, &symsize,
		    &isfunc);
		if (symbol == NULL) {
			symbol = dis_find_section(tgt, addr, &symoffset);
			symsize = symoffset;
		}

		if (symbol != last_symbol)
			getsymname(addr, symbol, symsize, symbuf,
			    sizeof (symbuf));

		symwidth = MAX(symwidth, strlen(symbuf));
		getsymname(addr, symbol, symoffset, symbuf, sizeof (symbuf));

		/*
		 * If we've crossed a new function boundary, print out the
		 * function name on a blank line.
		 */
		if (!g_quiet && symoffset == 0 && symbol != NULL && isfunc)
			(void) printf("%s()\n", symbol);

		(void) printf("    %s:%*s ", symbuf,
		    symwidth - strlen(symbuf), "");

		/* print bytes */
		for (i = 0; i < MIN(bytesperline, (db.db_nextaddr - addr));
		    i++) {
			int byte = *((uchar_t *)data + (addr - db.db_addr) + i);
			if (g_flags & DIS_OCTAL)
				(void) printf("%03o ", byte);
			else
				(void) printf("%02x ", byte);
		}

		/* trailing spaces for missing bytes */
		for (; i < bytesperline; i++) {
			if (g_flags & DIS_OCTAL)
				(void) printf("    ");
			else
				(void) printf("   ");
		}

		/* contents of disassembly */
		(void) printf(" %s", buf);

		/* excess bytes that spill over onto subsequent lines */
		for (; i < db.db_nextaddr - addr; i++) {
			int byte = *((uchar_t *)data + (addr - db.db_addr) + i);
			if (i % bytesperline == 0)
				(void) printf("\n    %*s  ", symwidth, "");
			if (g_flags & DIS_OCTAL)
				(void) printf("%03o ", byte);
			else
				(void) printf("%02x ", byte);
		}

		(void) printf("\n");

		addr = db.db_nextaddr;
	}
}

/*
 * libdisasm wrapper around symbol lookup.  Invoke the target-specific lookup
 * function, and convert the result using getsymname().
 */
int
do_lookup(void *data, uint64_t addr, char *buf, size_t buflen, uint64_t *start,
    size_t *symlen)
{
	dis_buffer_t *db = data;
	const char *symbol;
	off_t offset;
	size_t size;

	/*
	 * If NULL symbol is returned, getsymname takes care of
	 * printing appropriate address in buf instead of symbol.
	 */
	symbol = dis_tgt_lookup(db->db_tgt, addr, &offset, 0, &size, NULL);

	if (buf != NULL)
		getsymname(addr, symbol, offset, buf, buflen);

	if (start != NULL)
		*start = addr - offset;
	if (symlen != NULL)
		*symlen = size;

	if (symbol == NULL)
		return (-1);

	return (0);
}

/*
 * libdisasm wrapper around target reading.  libdisasm will always read data
 * in order, so update our current offset within the buffer appropriately.
 * We only support reading from within the current object; libdisasm should
 * never ask us to do otherwise.
 */
int
do_read(void *data, uint64_t addr, void *buf, size_t len)
{
	dis_buffer_t *db = data;
	size_t offset;

	if (addr < db->db_addr || addr >= db->db_addr + db->db_size)
		return (-1);

	offset = addr - db->db_addr;
	len = MIN(len, db->db_size - offset);

	(void) memcpy(buf, (char *)db->db_data + offset, len);

	db->db_nextaddr = addr + len;

	return (len);
}

/*
 * Routine to dump raw data in a human-readable format.  Used by the -d and -D
 * options.
 */
void
dump_data(uint64_t addr, void *data, size_t datalen)
{
	hexdump_t h;

	hexdump_init(&h);
	/* Print out data in two-byte chunks. */
	hexdump_set_grouping(&h, 2);
	hexdump_set_addr(&h, addr);

	/*
	 * Determine if the address given to us fits in 32-bit range, in which
	 * case use a 4-byte width.
	 */
	if (((addr + datalen) & 0xffffffff00000000ULL) == 0ULL)
		hexdump_set_addrwidth(&h, 8);
	else
		hexdump_set_addrwidth(&h, 16);


	(void) hexdump_fileh(&h, data, datalen, HDF_DEFAULT | HDF_ALIGN,
	    stdout);

	hexdump_fini(&h);
}

/*
 * Disassemble a section implicitly specified as part of a file.  This function
 * is called for all sections when no other flags are specified.  We ignore any
 * data sections, and print out only those sections containing text.
 */
void
dis_text_section(dis_tgt_t *tgt, dis_scn_t *scn, void *data)
{
	dis_handle_t *dhp = data;

	/* ignore data sections */
	if (!dis_section_istext(scn))
		return;

	if (!g_quiet)
		(void) printf("\nsection %s\n", dis_section_name(scn));

	dis_data(tgt, dhp, dis_section_addr(scn), dis_section_data(scn),
	    dis_section_size(scn));
}

/*
 * Structure passed to dis_named_{section,function} which keeps track of both
 * the target and the libdisasm handle.
 */
typedef struct callback_arg {
	dis_tgt_t	*ca_tgt;
	dis_handle_t	*ca_handle;
} callback_arg_t;

/*
 * Disassemble a section explicitly named with -s, -d, or -D.  The 'type'
 * argument contains the type of argument given.  Pass the data onto the
 * appropriate helper routine.
 */
void
dis_named_section(dis_scn_t *scn, int type, void *data)
{
	callback_arg_t *ca = data;

	if (!g_quiet)
		(void) printf("\nsection %s\n", dis_section_name(scn));

	switch (type) {
	case DIS_DATA_RELATIVE:
		dump_data(0, dis_section_data(scn), dis_section_size(scn));
		break;
	case DIS_DATA_ABSOLUTE:
		dump_data(dis_section_addr(scn), dis_section_data(scn),
		    dis_section_size(scn));
		break;
	case DIS_TEXT:
		dis_data(ca->ca_tgt, ca->ca_handle, dis_section_addr(scn),
		    dis_section_data(scn), dis_section_size(scn));
		break;
	}
}

/*
 * Disassemble a function explicitly specified with '-F'.  The 'type' argument
 * is unused.
 */
/* ARGSUSED */
void
dis_named_function(dis_func_t *func, int type, void *data)
{
	callback_arg_t *ca = data;

	dis_data(ca->ca_tgt, ca->ca_handle, dis_function_addr(func),
	    dis_function_data(func), dis_function_size(func));
}

/*
 * Disassemble a complete file.  First, we determine the type of the file based
 * on the ELF machine type, and instantiate a version of the disassembler
 * appropriate for the file.  We then resolve any named sections or functions
 * against the file, and iterate over the results (or all sections if no flags
 * were specified).
 */
void
dis_file(const char *filename)
{
	dis_tgt_t *tgt, *current;
	dis_scnlist_t *sections;
	dis_funclist_t *functions;
	dis_handle_t *dhp;
	GElf_Ehdr ehdr;

	/*
	 * First, initialize the target
	 */
	if ((tgt = dis_tgt_create(filename)) == NULL)
		return;

	if (!g_quiet)
		(void) printf("disassembly for %s\n\n",  filename);

	/*
	 * A given file may contain multiple targets (if it is an archive, for
	 * example).  We iterate over all possible targets if this is the case.
	 */
	for (current = tgt; current != NULL; current = dis_tgt_next(current)) {
		dis_tgt_ehdr(current, &ehdr);

		/*
		 * Eventually, this should probably live within libdisasm, and
		 * we should be able to disassemble targets from different
		 * architectures.  For now, we only support objects as the
		 * native machine type.
		 */
		switch (ehdr.e_machine) {
		case EM_SPARC:
			if (ehdr.e_ident[EI_CLASS] != ELFCLASS32 ||
			    ehdr.e_ident[EI_DATA] != ELFDATA2MSB) {
				warn("invalid E_IDENT field for SPARC object");
				return;
			}
			g_flags |= DIS_SPARC_V8;
			break;

		case EM_SPARC32PLUS:
		{
			uint64_t flags = ehdr.e_flags & EF_SPARC_32PLUS_MASK;

			if (ehdr.e_ident[EI_CLASS] != ELFCLASS32 ||
			    ehdr.e_ident[EI_DATA] != ELFDATA2MSB) {
				warn("invalid E_IDENT field for SPARC object");
				return;
			}

			if (flags != 0 &&
			    (flags & (EF_SPARC_32PLUS | EF_SPARC_SUN_US1 |
			    EF_SPARC_SUN_US3)) != EF_SPARC_32PLUS)
				g_flags |= DIS_SPARC_V9 | DIS_SPARC_V9_SGI;
			else
				g_flags |= DIS_SPARC_V9;
			break;
		}

		case EM_SPARCV9:
			if (ehdr.e_ident[EI_CLASS] != ELFCLASS64 ||
			    ehdr.e_ident[EI_DATA] != ELFDATA2MSB) {
				warn("invalid E_IDENT field for SPARC object");
				return;
			}

			g_flags |= DIS_SPARC_V9 | DIS_SPARC_V9_SGI;
			break;

		case EM_386:
			g_flags |= DIS_X86_SIZE32;
			break;

		case EM_AMD64:
			g_flags |= DIS_X86_SIZE64;
			break;

		case EM_S370:
			g_flags |= DIS_S370;

			if (ehdr.e_ident[EI_CLASS] != ELFCLASS32 ||
			    ehdr.e_ident[EI_DATA] != ELFDATA2MSB) {
				warn("invalid E_IDENT field for S370 object");
				return;
			}
			break;

		case EM_S390:
			/*
			 * Both 390 and z/Architecture use EM_S390, the only
			 * differences is the class: ELFCLASS32 for plain
			 * old s390 and ELFCLASS64 for z/Architecture (aka.
			 * s390x).
			 */
			if (ehdr.e_ident[EI_CLASS] == ELFCLASS32) {
				g_flags |= DIS_S390_31;
			} else if (ehdr.e_ident[EI_CLASS] == ELFCLASS64) {
				g_flags |= DIS_S390_64;
			} else {
				warn("invalid E_IDENT field for S390 object");
				return;
			}

			if (ehdr.e_ident[EI_DATA] != ELFDATA2MSB) {
				warn("invalid E_IDENT field for S390 object");
				return;
			}
			break;

		case EM_RISCV:
			/*
			 * RISC-V is defined to be litle endian. The current ISA
			 * makes it clear that the 64-bit instructions can
			 * co-exist with the 32-bit ones and therefore we don't
			 * need a separate elf class at this time.
			 */
			if (ehdr.e_ident[EI_DATA] != ELFDATA2LSB) {
				warn("invalid EI_DATA field for RISC-V object");
				return;
			}

			if (ehdr.e_ident[EI_CLASS] == ELFCLASS32) {
				g_flags |= DIS_RISCV_32;
			} else if (ehdr.e_ident[EI_CLASS] == ELFCLASS64) {
				g_flags |= DIS_RISCV_64;
			} else {
				warn("invalid EI_CLASS field for RISC-V "
				    "object");
				return;
			}
			break;

		default:
			die("%s: unsupported ELF machine 0x%x", filename,
			    ehdr.e_machine);
		}

		/*
		 * If ET_REL (.o), printing immediate symbols is likely to
		 * result in garbage, as symbol lookups on unrelocated
		 * immediates find false and useless matches.
		 */

		if (ehdr.e_type == ET_REL)
			g_flags |= DIS_NOIMMSYM;

		if (!g_quiet && dis_tgt_member(current) != NULL)
			(void) printf("\narchive member %s\n",
			    dis_tgt_member(current));

		/*
		 * Instantiate a libdisasm handle based on the file type.
		 */
		if ((dhp = dis_handle_create(g_flags, current, do_lookup,
		    do_read)) == NULL)
			die("%s: failed to initialize disassembler: %s",
			    filename, dis_strerror(dis_errno()));

		if (g_doall) {
			/*
			 * With no arguments, iterate over all sections and
			 * disassemble only those that contain text.
			 */
			dis_tgt_section_iter(current, dis_text_section, dhp);
		} else {
			callback_arg_t ca;

			ca.ca_tgt = current;
			ca.ca_handle = dhp;

			/*
			 * If sections or functions were explicitly specified,
			 * resolve those names against the object, and iterate
			 * over just the resulting data.
			 */
			sections = dis_namelist_resolve_sections(g_seclist,
			    current);
			functions = dis_namelist_resolve_functions(g_funclist,
			    current);

			dis_scnlist_iter(sections, dis_named_section, &ca);
			dis_funclist_iter(functions, dis_named_function, &ca);

			dis_scnlist_destroy(sections);
			dis_funclist_destroy(functions);
		}

		dis_handle_destroy(dhp);
	}

	dis_tgt_destroy(tgt);
}

void
usage(void)
{
	(void) fprintf(stderr, "usage: dis [-CVoqn] [-d sec] \n");
	(void) fprintf(stderr, "\t[-D sec] [-F function] [-t sec] file ..\n");
	exit(2);
}

typedef struct lib_node {
	char *path;
	struct lib_node *next;
} lib_node_t;

int
main(int argc, char **argv)
{
	int optchar;
	int i;
	lib_node_t *libs = NULL;

	g_funclist = dis_namelist_create();
	g_seclist = dis_namelist_create();

	while ((optchar = getopt(argc, argv, "Cd:D:F:l:Lot:Vqn")) != -1) {
		switch (optchar) {
		case 'C':
			g_demangle = 1;
			break;
		case 'd':
			dis_namelist_add(g_seclist, optarg, DIS_DATA_RELATIVE);
			break;
		case 'D':
			dis_namelist_add(g_seclist, optarg, DIS_DATA_ABSOLUTE);
			break;
		case 'F':
			dis_namelist_add(g_funclist, optarg, 0);
			break;
		case 'l': {
			/*
			 * The '-l foo' option historically would attempt to
			 * disassemble '$LIBDIR/libfoo.a'.  The $LIBDIR
			 * environment variable has never been supported or
			 * documented for our linker.  However, until this
			 * option is formally EOLed, we have to support it.
			 */
			char *dir;
			lib_node_t *node;
			size_t len;

			if ((dir = getenv("LIBDIR")) == NULL ||
			    dir[0] == '\0')
				dir = "/usr/lib";
			node = safe_malloc(sizeof (lib_node_t));
			len = strlen(optarg) + strlen(dir) + sizeof ("/lib.a");
			node->path = safe_malloc(len);

			(void) snprintf(node->path, len, "%s/lib%s.a", dir,
			    optarg);
			node->next = libs;
			libs = node;
			break;
		}
		case 'L':
			/*
			 * The '-L' option historically would attempt to read
			 * the .debug section of the target to determine source
			 * line information in order to annotate the output.
			 * No compiler has emitted these sections in many years,
			 * and the option has never done what it purported to
			 * do.  We silently consume the option for
			 * compatibility.
			 */
			break;
		case 'n':
			g_numeric = 1;
			break;
		case 'o':
			g_flags |= DIS_OCTAL;
			break;
		case 'q':
			g_quiet = 1;
			break;
		case 't':
			dis_namelist_add(g_seclist, optarg, DIS_TEXT);
			break;
		case 'V':
			(void) printf("Solaris disassembler version 1.0\n");
			return (0);
		default:
			usage();
			break;
		}
	}

	argc -= optind;
	argv += optind;

	if (argc == 0 && libs == NULL) {
		warn("no objects specified");
		usage();
	}

	if (dis_namelist_empty(g_funclist) && dis_namelist_empty(g_seclist))
		g_doall = 1;

	/*
	 * See comment for 'l' option, above.
	 */
	while (libs != NULL) {
		lib_node_t *node = libs->next;

		dis_file(libs->path);
		free(libs->path);
		free(libs);
		libs = node;
	}

	for (i = 0; i < argc; i++)
		dis_file(argv[i]);

	dis_namelist_destroy(g_funclist);
	dis_namelist_destroy(g_seclist);

	return (g_error);
}
/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */

/*
 * Copyright (c) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
 *
 * Copyright 2011 Jason King.  All rights reserved.
 */

#include <assert.h>
#include <errno.h>
#include <fcntl.h>
#include <gelf.h>
#include <libelf.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>

#include <sys/fcntl.h>
#include <sys/stat.h>
#include <sys/sysmacros.h>
#include <sys/types.h>

#include "dis_target.h"
#include "dis_util.h"

/*
 * Standard ELF disassembler target.
 *
 * We only support disassembly of ELF files, though this target interface could
 * be extended in the future.  Each basic type (target, func, section) contains
 * enough information to uniquely identify the location within the file.  The
 * interfaces use libelf(3LIB) to do the actual processing of the file.
 */

/*
 * Symbol table entry type.  We maintain our own symbol table sorted by address,
 * with the symbol name already resolved against the ELF symbol table.
 */
typedef struct sym_entry {
	GElf_Sym	se_sym;		/* value of symbol */
	char		*se_name;	/* name of symbol */
	int		se_shndx;	/* section where symbol is located */
} sym_entry_t;

/*
 * Create a map of the virtual address ranges of every section.  This will
 * allow us to create dummpy mappings for unassigned addresses.  Otherwise
 * multiple sections with unassigned addresses will appear to overlap and
 * mess up symbol resolution (which uses the virtual address).
 */
typedef struct dis_shnmap {
	const char	*dm_name;	/* name of section */
	uint64_t	dm_start;	/* virtual address of section */
	size_t		dm_length;	/* address length */
	boolean_t	dm_mapped;	/* did we assign the mapping */
} dis_shnmap_t;

/*
 * Target data structure.  This structure keeps track of the ELF file
 * information, a few bits of pre-processed section index information, and
 * sorted versions of the symbol table.  We also keep track of the last symbol
 * looked up, as the majority of lookups remain within the same symbol.
 */
struct dis_tgt {
	Elf		*dt_elf;	/* libelf handle */
	Elf		*dt_elf_root;	/* main libelf handle (for archives) */
	const char	*dt_filename;	/* name of file */
	int		dt_fd;		/* underlying file descriptor */
	size_t		dt_shstrndx;	/* section index of .shstrtab */
	size_t		dt_symidx;	/* section index of symbol table */
	sym_entry_t	*dt_symcache;	/* last symbol looked up */
	sym_entry_t	*dt_symtab;	/* sorted symbol table */
	int		dt_symcount;	/* # of symbol table entries */
	struct dis_tgt	*dt_next;	/* next target (for archives) */
	Elf_Arhdr	*dt_arhdr;	/* archive header (for archives) */
	dis_shnmap_t	*dt_shnmap;	/* section address map */
	size_t		dt_shncount;	/* # of sections in target */
};

/*
 * Function data structure.  We resolve the symbol and lookup the associated ELF
 * data when building this structure.  The offset is calculated based on the
 * section's starting address.
 */
struct dis_func {
	sym_entry_t	*df_sym;	/* symbol table reference */
	Elf_Data	*df_data;	/* associated ELF data */
	size_t		df_offset;	/* offset within data */
};

/*
 * Section data structure.  We store the entire section header so that we can
 * determine some properties (such as whether or not it contains text) after
 * building the structure.
 */
struct dis_scn {
	GElf_Shdr	ds_shdr;
	const char	*ds_name;
	Elf_Data	*ds_data;
};

/* Lifted from Psymtab.c, omitting STT_TLS */
#define	DATA_TYPES      \
	((1 << STT_OBJECT) | (1 << STT_FUNC) | (1 << STT_COMMON))
#define	IS_DATA_TYPE(tp)	(((1 << (tp)) & DATA_TYPES) != 0)

/*
 * Save the virtual address range for this section and select the
 * best section to use as the symbol table.  We prefer SHT_SYMTAB
 * over SHT_DYNSYM.
 */
/* ARGSUSED */
static void
tgt_scn_init(dis_tgt_t *tgt, dis_scn_t *scn, void *data)
{
	int *index = data;

	*index += 1;

	tgt->dt_shnmap[*index].dm_name = scn->ds_name;
	tgt->dt_shnmap[*index].dm_start = scn->ds_shdr.sh_addr;
	tgt->dt_shnmap[*index].dm_length = scn->ds_shdr.sh_size;
	tgt->dt_shnmap[*index].dm_mapped = B_FALSE;

	/*
	 * Prefer SHT_SYMTAB over SHT_DYNSYM
	 */
	if (scn->ds_shdr.sh_type == SHT_DYNSYM && tgt->dt_symidx == 0)
		tgt->dt_symidx = *index;
	else if (scn->ds_shdr.sh_type == SHT_SYMTAB)
		tgt->dt_symidx = *index;
}

static int
sym_compare(const void *a, const void *b)
{
	const sym_entry_t *syma = a;
	const sym_entry_t *symb = b;
	const char *aname = syma->se_name;
	const char *bname = symb->se_name;
	size_t alen;
	size_t blen;

	if (syma->se_sym.st_value < symb->se_sym.st_value)
		return (-1);

	if (syma->se_sym.st_value > symb->se_sym.st_value)
		return (1);

	/*
	 * Prefer functions over non-functions
	 */
	if (GELF_ST_TYPE(syma->se_sym.st_info) !=
	    GELF_ST_TYPE(symb->se_sym.st_info)) {
		if (GELF_ST_TYPE(syma->se_sym.st_info) == STT_FUNC)
			return (-1);
		if (GELF_ST_TYPE(symb->se_sym.st_info) == STT_FUNC)
			return (1);
	}

	/*
	 * For symbols with the same address and type, we sort them according to
	 * a hierarchy:
	 *
	 *	1. weak symbols (common name)
	 *	2. global symbols (external name)
	 *	3. local symbols
	 */
	if (GELF_ST_BIND(syma->se_sym.st_info) !=
	    GELF_ST_BIND(symb->se_sym.st_info)) {
		if (GELF_ST_BIND(syma->se_sym.st_info) == STB_WEAK)
			return (-1);
		if (GELF_ST_BIND(symb->se_sym.st_info) == STB_WEAK)
			return (1);

		if (GELF_ST_BIND(syma->se_sym.st_info) == STB_GLOBAL)
			return (-1);
		if (GELF_ST_BIND(symb->se_sym.st_info) == STB_GLOBAL)
			return (1);
	}

	/*
	 * As a last resort, if we have multiple symbols of the same type at the
	 * same address, prefer the version with the fewest leading underscores.
	 */
	if (aname == NULL)
		return (-1);
	if (bname == NULL)
		return (1);

	while (*aname == '_' && *bname == '_') {
		aname++;
		bname++;
	}

	if (*bname == '_')
		return (-1);
	if (*aname == '_')
		return (1);

	/*
	 * Prefer the symbol with the smaller size.
	 */
	if (syma->se_sym.st_size < symb->se_sym.st_size)
		return (-1);
	if (syma->se_sym.st_size > symb->se_sym.st_size)
		return (1);

	/*
	 * We really do have two identical symbols, choose the one with the
	 * shortest name if we can, heuristically taking it to be the most
	 * representative.
	 */
	alen = strlen(syma->se_name);
	blen = strlen(symb->se_name);

	if (alen < blen)
		return (-1);
	else if (alen > blen)
		return (1);

	/*
	 * If all else fails, compare the names, so that we give a stable
	 * sort
	 */
	return (strcmp(syma->se_name, symb->se_name));
}

/*
 * Construct an optimized symbol table sorted by starting address.
 */
static void
construct_symtab(dis_tgt_t *tgt)
{
	Elf_Scn *scn;
	GElf_Shdr shdr;
	Elf_Data *symdata;
	int i;
	GElf_Word *symshndx = NULL;
	int symshndx_size;
	sym_entry_t *sym;
	sym_entry_t *p_symtab = NULL;
	int nsym = 0; /* count of symbols we're not interested in */

	/*
	 * Find the symshndx section, if any
	 */
	for (scn = elf_nextscn(tgt->dt_elf, NULL); scn != NULL;
	    scn = elf_nextscn(tgt->dt_elf, scn)) {
		if (gelf_getshdr(scn, &shdr) == NULL)
			break;
		if (shdr.sh_type == SHT_SYMTAB_SHNDX &&
		    shdr.sh_link == tgt->dt_symidx) {
			Elf_Data	*data;

			if ((data = elf_getdata(scn, NULL)) != NULL) {
				symshndx = (GElf_Word *)data->d_buf;
				symshndx_size = data->d_size /
				    sizeof (GElf_Word);
				break;
			}
		}
	}

	if ((scn = elf_getscn(tgt->dt_elf, tgt->dt_symidx)) == NULL)
		die("%s: failed to get section information", tgt->dt_filename);
	if (gelf_getshdr(scn, &shdr) == NULL)
		die("%s: failed to get section header", tgt->dt_filename);
	if (shdr.sh_entsize == 0)
		die("%s: symbol table has zero size", tgt->dt_filename);

	if ((symdata = elf_getdata(scn, NULL)) == NULL)
		die("%s: failed to get symbol table", tgt->dt_filename);

	tgt->dt_symcount = symdata->d_size / gelf_fsize(tgt->dt_elf, ELF_T_SYM,
	    1, EV_CURRENT);

	p_symtab = safe_malloc(tgt->dt_symcount * sizeof (sym_entry_t));

	for (i = 0, sym = p_symtab; i < tgt->dt_symcount; i++) {
		if (gelf_getsym(symdata, i, &(sym->se_sym)) == NULL) {
			warn("%s: gelf_getsym returned NULL for %d",
			    tgt->dt_filename, i);
			nsym++;
			continue;
		}

		/*
		 * We're only interested in data symbols.
		 */
		if (!IS_DATA_TYPE(GELF_ST_TYPE(sym->se_sym.st_info))) {
			nsym++;
			continue;
		}

		if (sym->se_sym.st_shndx == SHN_XINDEX && symshndx != NULL) {
			if (i > symshndx_size) {
				warn("%s: bad SHNX_XINDEX %d",
				    tgt->dt_filename, i);
				sym->se_shndx = -1;
			} else {
				sym->se_shndx = symshndx[i];
			}
		} else {
			sym->se_shndx = sym->se_sym.st_shndx;
		}

		/* Deal with symbols with special section indicies */
		if (sym->se_shndx == SHN_ABS) {
			/*
			 * If st_value == 0, references to these
			 * symbols in code are modified in situ
			 * thus we will never attempt to look
			 * them up.
			 */
			if (sym->se_sym.st_value == 0) {
				/*
				 * References to these symbols in code
				 * are modified in situ by the runtime
				 * linker and no code on disk will ever
				 * attempt to look them up.
				 */
				nsym++;
				continue;
			} else {
				/*
				 * If st_value != 0, (such as examining
				 * something in /system/object/.../object)
				 * the values should resolve to a value
				 * within an existing section (such as
				 * .data).  This also means it never needs
				 * to have st_value mapped.
				 */
				sym++;
				continue;
			}
		}

		/*
		 * Ignore the symbol if it has some other special
		 * section index
		 */
		if (sym->se_shndx == SHN_UNDEF ||
		    sym->se_shndx >= SHN_LORESERVE) {
			nsym++;
			continue;
		}

		if ((sym->se_name = elf_strptr(tgt->dt_elf, shdr.sh_link,
		    (size_t)sym->se_sym.st_name)) == NULL) {
			warn("%s: failed to lookup symbol %d name",
			    tgt->dt_filename, i);
			nsym++;
			continue;
		}

		/*
		 * If we had to map this section, its symbol value
		 * also needs to be mapped.
		 */
		if (tgt->dt_shnmap[sym->se_shndx].dm_mapped)
			sym->se_sym.st_value +=
			    tgt->dt_shnmap[sym->se_shndx].dm_start;

		sym++;
	}

	tgt->dt_symcount -= nsym;
	tgt->dt_symtab = realloc(p_symtab, tgt->dt_symcount *
	    sizeof (sym_entry_t));

	qsort(tgt->dt_symtab, tgt->dt_symcount, sizeof (sym_entry_t),
	    sym_compare);
}

/*
 * Assign virtual address ranges for sections that need it
 */
static void
create_addrmap(dis_tgt_t *tgt)
{
	uint64_t addr;
	int i;

	if (tgt->dt_shnmap == NULL)
		return;

	/* find the greatest used address */
	for (addr = 0, i = 1; i < tgt->dt_shncount; i++)
		if (tgt->dt_shnmap[i].dm_start > addr)
			addr = tgt->dt_shnmap[i].dm_start +
			    tgt->dt_shnmap[i].dm_length;

	addr = P2ROUNDUP(addr, 0x1000);

	/*
	 * Assign section a starting address beyond the largest mapped section
	 * if no address was given.
	 */
	for (i = 1; i < tgt->dt_shncount; i++) {
		if (tgt->dt_shnmap[i].dm_start != 0)
			continue;

		tgt->dt_shnmap[i].dm_start = addr;
		tgt->dt_shnmap[i].dm_mapped = B_TRUE;
		addr = P2ROUNDUP(addr + tgt->dt_shnmap[i].dm_length, 0x1000);
	}
}

/*
 * Create a target backed by an ELF file.
 */
dis_tgt_t *
dis_tgt_create(const char *file)
{
	dis_tgt_t *tgt, *current;
	int idx;
	Elf *elf;
	GElf_Ehdr ehdr;
	Elf_Arhdr *arhdr = NULL;
	int cmd;

	if (elf_version(EV_CURRENT) == EV_NONE)
		die("libelf out of date");

	tgt = safe_malloc(sizeof (dis_tgt_t));

	if ((tgt->dt_fd = open(file, O_RDONLY)) < 0) {
		warn("%s: failed opening file, reason: %s", file,
		    strerror(errno));
		free(tgt);
		return (NULL);
	}

	if ((tgt->dt_elf_root =
	    elf_begin(tgt->dt_fd, ELF_C_READ, NULL)) == NULL) {
		warn("%s: invalid or corrupt ELF file", file);
		dis_tgt_destroy(tgt);
		return (NULL);
	}

	current = tgt;
	cmd = ELF_C_READ;
	while ((elf = elf_begin(tgt->dt_fd, cmd, tgt->dt_elf_root)) != NULL) {
		size_t shnum = 0;

		if (elf_kind(tgt->dt_elf_root) == ELF_K_AR &&
		    (arhdr = elf_getarhdr(elf)) == NULL) {
			warn("%s: malformed archive", file);
			dis_tgt_destroy(tgt);
			return (NULL);
		}

		/*
		 * Make sure that this Elf file is sane
		 */
		if (gelf_getehdr(elf, &ehdr) == NULL) {
			if (arhdr != NULL) {
				/*
				 * For archives, we drive on in the face of bad
				 * members.  The "/" and "//" members are
				 * special, and should be silently ignored.
				 */
				if (strcmp(arhdr->ar_name, "/") != 0 &&
				    strcmp(arhdr->ar_name, "//") != 0)
					warn("%s[%s]: invalid file type",
					    file, arhdr->ar_name);
				cmd = elf_next(elf);
				(void) elf_end(elf);
				continue;
			}

			warn("%s: invalid file type", file);
			dis_tgt_destroy(tgt);
			return (NULL);
		}

		/*
		 * If we're seeing a new Elf object, then we have an
		 * archive. In this case, we create a new target, and chain it
		 * off the master target.  We can later iterate over these
		 * targets using dis_tgt_next().
		 */
		if (current->dt_elf != NULL) {
			dis_tgt_t *next = safe_malloc(sizeof (dis_tgt_t));
			next->dt_elf_root = tgt->dt_elf_root;
			next->dt_fd = -1;
			current->dt_next = next;
			current = next;
		}
		current->dt_elf = elf;
		current->dt_arhdr = arhdr;

		if (elf_getshdrstrndx(elf, &current->dt_shstrndx) == -1) {
			warn("%s: failed to get section string table for "
			    "file", file);
			dis_tgt_destroy(tgt);
			return (NULL);
		}

		if (elf_getshdrnum(elf, &shnum) == -1) {
			warn("%s: failed to get number of sections in file",
			    file);
			dis_tgt_destroy(tgt);
			return (NULL);
		}

		current->dt_shnmap = safe_malloc(sizeof (dis_shnmap_t) *
		    shnum);
		current->dt_shncount = shnum;

		idx = 0;
		dis_tgt_section_iter(current, tgt_scn_init, &idx);
		current->dt_filename = file;

		create_addrmap(current);
		if (current->dt_symidx != 0)
			construct_symtab(current);

		cmd = elf_next(elf);
	}

	/*
	 * Final sanity check.  If we had an archive with no members, then bail
	 * out with a nice message.
	 */
	if (tgt->dt_elf == NULL) {
		warn("%s: empty archive\n", file);
		dis_tgt_destroy(tgt);
		return (NULL);
	}

	return (tgt);
}

/*
 * Return the filename associated with the target.
 */
const char *
dis_tgt_name(dis_tgt_t *tgt)
{
	return (tgt->dt_filename);
}

/*
 * Return the archive member name, if any.
 */
const char *
dis_tgt_member(dis_tgt_t *tgt)
{
	if (tgt->dt_arhdr)
		return (tgt->dt_arhdr->ar_name);
	else
		return (NULL);
}

/*
 * Return the Elf_Ehdr associated with this target.  Needed to determine which
 * disassembler to use.
 */
void
dis_tgt_ehdr(dis_tgt_t *tgt, GElf_Ehdr *ehdr)
{
	(void) gelf_getehdr(tgt->dt_elf, ehdr);
}

/*
 * Return the next target in the list, if this is an archive.
 */
dis_tgt_t *
dis_tgt_next(dis_tgt_t *tgt)
{
	return (tgt->dt_next);
}

/*
 * Destroy a target and free up any associated memory.
 */
void
dis_tgt_destroy(dis_tgt_t *tgt)
{
	dis_tgt_t *current, *next;

	current = tgt->dt_next;
	while (current != NULL) {
		next = current->dt_next;
		if (current->dt_elf)
			(void) elf_end(current->dt_elf);
		if (current->dt_symtab)
			free(current->dt_symtab);
		free(current);
		current = next;
	}

	if (tgt->dt_elf)
		(void) elf_end(tgt->dt_elf);
	if (tgt->dt_elf_root)
		(void) elf_end(tgt->dt_elf_root);

	if (tgt->dt_symtab)
		free(tgt->dt_symtab);

	free(tgt);
}

/*
 * Given an address, return the section it is in and set the offset within
 * the section.
 */
const char *
dis_find_section(dis_tgt_t *tgt, uint64_t addr, off_t *offset)
{
	int i;

	for (i = 1; i < tgt->dt_shncount; i++) {
		if ((addr >= tgt->dt_shnmap[i].dm_start) &&
		    (addr < tgt->dt_shnmap[i].dm_start +
		    tgt->dt_shnmap[i].dm_length)) {
			*offset = addr - tgt->dt_shnmap[i].dm_start;
			return (tgt->dt_shnmap[i].dm_name);
		}
	}

	*offset = 0;
	return (NULL);
}

/*
 * Given an address, returns the name of the corresponding symbol, as well as
 * the offset within that symbol.  If no matching symbol is found, then NULL is
 * returned.
 *
 * If 'cache_result' is specified, then we keep track of the resulting symbol.
 * This cached result is consulted first on subsequent lookups in order to avoid
 * unecessary lookups.  This flag should be used for resolving the current PC,
 * as the majority of addresses stay within the current function.
 */
const char *
dis_tgt_lookup(dis_tgt_t *tgt, uint64_t addr, off_t *offset, int cache_result,
    size_t *size, int *isfunc)
{
	int lo, hi, mid;
	sym_entry_t *sym, *osym, *match;
	int found;

	*offset = 0;
	*size = 0;
	if (isfunc != NULL)
		*isfunc = 0;

	if (tgt->dt_symcache != NULL &&
	    addr >= tgt->dt_symcache->se_sym.st_value &&
	    addr < tgt->dt_symcache->se_sym.st_value +
	    tgt->dt_symcache->se_sym.st_size) {
		sym = tgt->dt_symcache;
		*offset = addr - sym->se_sym.st_value;
		*size = sym->se_sym.st_size;
		if (isfunc != NULL)
			*isfunc = (GELF_ST_TYPE(sym->se_sym.st_info) ==
			    STT_FUNC);
		return (sym->se_name);
	}

	lo = 0;
	hi = (tgt->dt_symcount - 1);
	found = 0;
	match = osym = NULL;
	while (lo <= hi) {
		mid = (lo + hi) / 2;

		sym = &tgt->dt_symtab[mid];

		if (addr >= sym->se_sym.st_value &&
		    addr < sym->se_sym.st_value + sym->se_sym.st_size &&
		    (!found || sym->se_sym.st_value > osym->se_sym.st_value)) {
			osym = sym;
			found = 1;
		} else if (addr == sym->se_sym.st_value) {
			/*
			 * Particularly for .plt objects, it's possible to have
			 * a zero sized object.  We want to return this, but we
			 * want it to be a last resort.
			 */
			match = sym;
		}

		if (addr < sym->se_sym.st_value)
			hi = mid - 1;
		else
			lo = mid + 1;
	}

	if (!found) {
		if (match)
			osym = match;
		else
			return (NULL);
	}

	/*
	 * Walk backwards to find the best match.
	 */
	do {
		sym = osym;

		if (osym == tgt->dt_symtab)
			break;

		osym = osym - 1;
	} while ((sym->se_sym.st_value == osym->se_sym.st_value) &&
	    (addr >= osym->se_sym.st_value) &&
	    (addr < osym->se_sym.st_value + osym->se_sym.st_size));

	if (cache_result)
		tgt->dt_symcache = sym;

	*offset = addr - sym->se_sym.st_value;
	*size = sym->se_sym.st_size;
	if (isfunc)
		*isfunc = (GELF_ST_TYPE(sym->se_sym.st_info) == STT_FUNC);

	return (sym->se_name);
}

/*
 * Given an address, return the starting offset of the next symbol in the file.
 * Only needed on variable length instruction architectures.
 */
off_t
dis_tgt_next_symbol(dis_tgt_t *tgt, uint64_t addr)
{
	sym_entry_t *sym;

	sym = (tgt->dt_symcache != NULL) ? tgt->dt_symcache : tgt->dt_symtab;

	while (sym != (tgt->dt_symtab + tgt->dt_symcount)) {
		if (sym->se_sym.st_value >= addr)
			return (sym->se_sym.st_value - addr);
		sym++;
	}

	return (0);
}

/*
 * Iterate over all sections in the target, executing the given callback for
 * each.
 */
void
dis_tgt_section_iter(dis_tgt_t *tgt, section_iter_f func, void *data)
{
	dis_scn_t sdata;
	Elf_Scn *scn;
	int idx;

	for (scn = elf_nextscn(tgt->dt_elf, NULL), idx = 1; scn != NULL;
	    scn = elf_nextscn(tgt->dt_elf, scn), idx++) {

		if (gelf_getshdr(scn, &sdata.ds_shdr) == NULL) {
			warn("%s: failed to get section %d header",
			    tgt->dt_filename, idx);
			continue;
		}

		if ((sdata.ds_name = elf_strptr(tgt->dt_elf, tgt->dt_shstrndx,
		    sdata.ds_shdr.sh_name)) == NULL) {
			warn("%s: failed to get section %d name",
			    tgt->dt_filename, idx);
			continue;
		}

		if ((sdata.ds_data = elf_getdata(scn, NULL)) == NULL) {
			warn("%s: failed to get data for section '%s'",
			    tgt->dt_filename, sdata.ds_name);
			continue;
		}

		/*
		 * dis_tgt_section_iter is also used before the section map
		 * is initialized, so only check when we need to.  If the
		 * section map is uninitialized, it will return 0 and have
		 * no net effect.
		 */
		if (sdata.ds_shdr.sh_addr == 0)
			sdata.ds_shdr.sh_addr = tgt->dt_shnmap[idx].dm_start;

		func(tgt, &sdata, data);
	}
}

/*
 * Return 1 if the given section contains text, 0 otherwise.
 */
int
dis_section_istext(dis_scn_t *scn)
{
	return ((scn->ds_shdr.sh_type == SHT_PROGBITS) &&
	    (scn->ds_shdr.sh_flags == (SHF_ALLOC | SHF_EXECINSTR)));
}

/*
 * Return a pointer to the section data.
 */
void *
dis_section_data(dis_scn_t *scn)
{
	return (scn->ds_data->d_buf);
}

/*
 * Return the size of the section data.
 */
size_t
dis_section_size(dis_scn_t *scn)
{
	return (scn->ds_data->d_size);
}

/*
 * Return the address for the given section.
 */
uint64_t
dis_section_addr(dis_scn_t *scn)
{
	return (scn->ds_shdr.sh_addr);
}

/*
 * Return the name of the current section.
 */
const char *
dis_section_name(dis_scn_t *scn)
{
	return (scn->ds_name);
}

/*
 * Create an allocated copy of the given section
 */
dis_scn_t *
dis_section_copy(dis_scn_t *scn)
{
	dis_scn_t *new;

	new = safe_malloc(sizeof (dis_scn_t));
	(void) memcpy(new, scn, sizeof (dis_scn_t));

	return (new);
}

/*
 * Free section memory
 */
void
dis_section_free(dis_scn_t *scn)
{
	free(scn);
}

/*
 * Iterate over all functions in the target, executing the given callback for
 * each one.
 */
void
dis_tgt_function_iter(dis_tgt_t *tgt, function_iter_f func, void *data)
{
	int i;
	sym_entry_t *sym;
	dis_func_t df;
	Elf_Scn *scn;
	GElf_Shdr	shdr;

	for (i = 0, sym = tgt->dt_symtab; i < tgt->dt_symcount; i++, sym++) {

		/* ignore non-functions */
		if ((GELF_ST_TYPE(sym->se_sym.st_info) != STT_FUNC) ||
		    (sym->se_name == NULL) ||
		    (sym->se_sym.st_size == 0) ||
		    (sym->se_shndx >= SHN_LORESERVE))
			continue;

		/* get the ELF data associated with this function */
		if ((scn = elf_getscn(tgt->dt_elf, sym->se_shndx)) == NULL ||
		    gelf_getshdr(scn, &shdr) == NULL ||
		    (df.df_data = elf_getdata(scn, NULL)) == NULL ||
		    df.df_data->d_size == 0) {
			warn("%s: failed to read section %d",
			    tgt->dt_filename, sym->se_shndx);
			continue;
		}

		if (tgt->dt_shnmap[sym->se_shndx].dm_mapped)
			shdr.sh_addr = tgt->dt_shnmap[sym->se_shndx].dm_start;

		/*
		 * Verify that the address lies within the section that we think
		 * it does.
		 */
		if (sym->se_sym.st_value < shdr.sh_addr ||
		    (sym->se_sym.st_value + sym->se_sym.st_size) >
		    (shdr.sh_addr + shdr.sh_size)) {
			warn("%s: bad section %d for address %p",
			    tgt->dt_filename, sym->se_sym.st_shndx,
			    sym->se_sym.st_value);
			continue;
		}

		df.df_sym = sym;
		df.df_offset = sym->se_sym.st_value - shdr.sh_addr;

		func(tgt, &df, data);
	}
}

/*
 * Return the data associated with a given function.
 */
void *
dis_function_data(dis_func_t *func)
{
	return ((char *)func->df_data->d_buf + func->df_offset);
}

/*
 * Return the size of a function.
 */
size_t
dis_function_size(dis_func_t *func)
{
	return (func->df_sym->se_sym.st_size);
}

/*
 * Return the address of a function.
 */
uint64_t
dis_function_addr(dis_func_t *func)
{
	return (func->df_sym->se_sym.st_value);
}

/*
 * Return the name of the function
 */
const char *
dis_function_name(dis_func_t *func)
{
	return (func->df_sym->se_name);
}

/*
 * Return a copy of a function.
 */
dis_func_t *
dis_function_copy(dis_func_t *func)
{
	dis_func_t *new;

	new = safe_malloc(sizeof (dis_func_t));
	(void) memcpy(new, func, sizeof (dis_func_t));

	return (new);
}

/*
 * Free function memory
 */
void
dis_function_free(dis_func_t *func)
{
	free(func);
}
/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */

/*
 * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 *
 * Copyright 2011 Jason King.  All rights reserved.
 */

#ifndef	_DIS_TARGET_H
#define	_DIS_TARGET_H

#ifdef __cplusplus
extern "C" {
#endif

#include <gelf.h>
#include <sys/types.h>

/*
 * Basic types
 */
typedef struct dis_tgt dis_tgt_t;
typedef struct dis_func dis_func_t;
typedef struct dis_scn dis_scn_t;

/*
 * Target management
 */
dis_tgt_t *dis_tgt_create(const char *);
void dis_tgt_destroy(dis_tgt_t *);
const char *dis_tgt_lookup(dis_tgt_t *, uint64_t, off_t *, int, size_t *,
    int *);
const char *dis_find_section(dis_tgt_t *, uint64_t, off_t *);
const char *dis_tgt_name(dis_tgt_t *);
const char *dis_tgt_member(dis_tgt_t *);
void dis_tgt_ehdr(dis_tgt_t *, GElf_Ehdr *);
off_t dis_tgt_next_symbol(dis_tgt_t *, uint64_t);
dis_tgt_t *dis_tgt_next(dis_tgt_t *);

/*
 * Section management
 */
typedef void (*section_iter_f)(dis_tgt_t *, dis_scn_t *, void *);
void dis_tgt_section_iter(dis_tgt_t *, section_iter_f, void *);

int dis_section_istext(dis_scn_t *);
void *dis_section_data(dis_scn_t *);
size_t dis_section_size(dis_scn_t *);
uint64_t dis_section_addr(dis_scn_t *);
const char *dis_section_name(dis_scn_t *);
dis_scn_t *dis_section_copy(dis_scn_t *);
void dis_section_free(dis_scn_t *);

/*
 * Function management
 */
typedef void (*function_iter_f)(dis_tgt_t *, dis_func_t *, void *);
void dis_tgt_function_iter(dis_tgt_t *, function_iter_f, void *);
dis_func_t *dis_tgt_function_lookup(dis_tgt_t *, const char *);

void *dis_function_data(dis_func_t *);
size_t dis_function_size(dis_func_t *);
uint64_t dis_function_addr(dis_func_t *);
const char *dis_function_name(dis_func_t *);
dis_func_t *dis_function_copy(dis_func_t *);
void dis_function_free(dis_func_t *);

#ifdef __cplusplus
}
#endif

#endif /* _DIS_TARGET_H */
/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */

/*
 * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 *
 * Copyright 2018 Jason King.
 * Copyright 2018, Joyent, Inc.
 */

#include <dlfcn.h>
#include <stdarg.h>
#include <stdio.h>
#include <stdlib.h>
#include <demangle-sys.h>

#include "dis_util.h"

int g_error;	/* global process exit status, set when warn() is called */

/*
 * Fatal error.  Print out the error with a leading "dis: ", and then exit the
 * program.
 */
void
die(const char *fmt, ...)
{
	va_list ap;

	(void) fprintf(stderr, "dis: fatal: ");

	va_start(ap, fmt);
	(void) vfprintf(stderr, fmt, ap);
	va_end(ap);

	(void) fprintf(stderr, "\n");

	exit(1);
}

/*
 * Non-fatal error.  Print out the error with a leading "dis: ", set the global
 * error flag, and return.
 */
void
warn(const char *fmt, ...)
{
	va_list ap;

	(void) fprintf(stderr, "dis: warning: ");

	va_start(ap, fmt);
	(void) vfprintf(stderr, fmt, ap);
	va_end(ap);

	(void) fprintf(stderr, "\n");

	g_error = 1;
}

/*
 * Convenience wrapper around malloc() to cleanly exit if any allocation fails.
 */
void *
safe_malloc(size_t size)
{
	void *ret;

	if ((ret = calloc(1, size)) == NULL)
		die("Out of memory");

	return (ret);
}


/*
 * Since the -C flag explicitly says C++, for now at least, force language to
 * C++
 */
const char *
dis_demangle(const char *name)
{
	static char *demangled_name = NULL;

	/*
	 * Since demangled_name is static, it may be preserved across
	 * invocations.  As such, make sure any memory that might be present
	 * from previous invocations is freed.
	 */
	free(demangled_name);
	demangled_name = sysdemangle(name, SYSDEM_LANG_AUTO, NULL);
	return ((demangled_name != NULL) ? demangled_name : name);
}
/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */

/*
 * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */

#ifndef _DIS_UTIL_H
#define	_DIS_UTIL_H

#include <sys/types.h>

#ifdef	__cplusplus
extern "C" {
#endif

void die(const char *, ...);
void warn(const char *, ...);

void *safe_malloc(size_t);
const char *dis_demangle(const char *);

extern int g_error;

#define	BUFSIZE	1024

#ifdef	__cplusplus
}
#endif

#endif /* _DIS_UTIL_H */